Laser Weeding Technology in Cropping Systems: A Comprehensive Review

Abstract

Weed infestations pose significant challenges to global crop production, demanding effective and sustainable weed control methods. Traditional approaches, such as chemical herbicides, mechanical tillage, and plastic mulches, are not only associated with environmental concerns but also face challenges like herbicide resistance, soil health, erosion, moisture content, and organic matter depletion. Thermal methods like flaming, streaming, and hot foam distribution are emerging weed control technologies along with directed energy systems of electrical and laser weeding. This paper conducts a comprehensive review of laser weeding technology, comparing it with conventional methods and highlighting its potential environmental benefits. Laser weeding, known for its precision and targeted energy delivery, emerges as a promising alternative to conventional control methods. This review explores various laser weeding platforms, discussing their features, applications, and limitations, with a focus on critical areas for improvement, including dwell time reduction, automated navigation, energy efficiency, affordability, and safety standards. Comparative analyses underscore the advantages of laser weeding, such as reduced environmental impact, minimized soil disturbance, and the potential for sustainable agriculture. This paper concludes by outlining key areas for future research and development to enhance the effectiveness, accessibility, and affordability of laser weeding technology. In summary, laser weeding presents a transformative solution for weed control, aligning with the principles of sustainable and environmentally conscious agriculture, and addressing the limitations of traditional methods.

1. Introduction

The prevalence of weeds globally represents a major obstacle to crop production, significantly hindering efforts to increase crop yields and ensure food security for a massively growing population [1]. Agricultural weed control strategies encompass a range of methods, including chemical applications, mechanical tillage, plastic or dry crop mulching, and manual labor [2]. While effective and economical weed control is essential, traditional methods often have detrimental effects on the environment because they collectively pollute the environment, kill healthy plants, harm beneficial organisms, and disturb soil health [3]. The usage of the herbicides, in particular, can lead to environmental contamination, as weeds usually occupy a small portion of the treated area, resulting in herbicide drift or impact on non-target organisms [4]. Despite this, chemical control remains the dominant yet expensive and potentially dangerous method for weed management [5,6]. The global rise in herbicide-resistant weed species has further exacerbated this challenge [7], with little progress in developing new herbicide modes of action since the 1980s [8]. Consequently, concerns over the negative impacts of herbicide use have led to more stringent regulations and policy efforts to restrict their application [9,10]. While various physical weed control methods, such as electrocution, exist, laser weeding is specifically highlighted due to its precision targeting capabilities, minimal soil disturbance, and potential for use in automated systems, making it particularly suitable for high-precision and sustainable farming.

In mechanical weed control, tillage adversely affects beneficial organisms such as predatory insects and wanderers on the soil surface, as well as angleworms within the earth [1]. Additionally, widespread tillage enhances the potential for soil erosion and the leaching of essential plant nutrients. This method also encourages the unnecessary mineralization of soil organic matter and contributes to the desiccation of soils with limited moisture content [2]. The use of fire for weeding requires a substantial amount of gas to burn, which raises more concerns about its long-term environmental sustainability due to the carbon dioxide (CO₂) emissions associated with its operation [3]. While shallow-depth mechanical weeding reduces some environmental impacts, laser weeding offers a precision alternative with potentially lower greenhouse gas emissions when powered by renewable energy sources, which should be compared alongside other methods.

Mulch weed control methods utilize an organic or inorganic cover material, which could be plastic sheets or dry crop biomass, and which is placed directly on the crop rows to control weeds [4]. The application of mulch on the soil surface functions as a physical barrier, blocking the passage of light and resulting in the decreased germination of small-seeded weed species [5], but plastic mulch polyethylene sheets are costly, difficult to handle, and also degrade the quality of the soil [6]. Also, mulches have adverse effects on soil pH, allelopathy, the struggle for nutrients and CO₂, disease induction, flammability, nitrogen insufficiency, pest, and weed infestation [4,7]. Woody materials used in nurseries can generate acids, like phenolic acids, during decomposition. However, under field conditions, the acidifying effect is minimal due to localized acid production and thus reduces the overall impact on soil acidity [8]. Allelochemicals leached from mulches can harm seedlings and shallow-rooted plants [9]. Plants and plant materials inside the mulch can also transfer disease to healthy crops [10]. Certain mulches can also serve as carriers for numerous weed seeds [8]. This paper focuses on laser weeding for its precision and technological potential, but acknowledges that dead mulching and living mulch systems, widely used in organic and conservation agriculture, are highly effective and warrant further comparison in future studies.

Weed control methods vary in effectiveness and environmental impact. Mechanical methods, such as tilling or hoeing, are widely accessible and effective for weed removal, but they can cause soil compaction and erosion over time [2,3]. Chemical herbicides are highly efficient and easy to apply, making them popular among farmers, but their long-term use raises concerns about environmental contamination, weed resistance, and potential harm to non-target organisms [4]. Biological control, including the use of natural predators or competitive plants, offers an eco-friendly solution but is often slower and less predictable in effectiveness [5,6,7,8]. Laser weeding provides a precision approach, eliminating weeds without disturbing the soil or introducing chemicals [11,12,13]. However, it has challenges, such as high initial costs and technical limitations in crops with dense or overlapping foliage. Each method has its own advantages and limitations, requiring careful consideration depending on the specific needs of the farming system.

Hence, there is a necessity to develop new techniques that complement or replace existing weed control methods [11,12]. The development of new techniques is essential to address challenges like herbicide resistance and environmental impact, complementing existing methods to enhance sustainability and precision in weed control.

Laser sources have the capability to deliver high-density energy to targeted objects or locations. By directing a laser beam towards a weed plant, the plant tissue can be heated, leading to harm or the death of the plant [13]. It is crucial to identify the target accurately to minimize energy consumption. Recognition tools, powered by artificial intelligence, enable real-time differentiation between weeds and crop plants in agricultural settings [14]. High-resolution cameras are capable of identifying the meristem, while precise laser scanners ensure accurate positioning [15]. As a result, laser weeding technology development has expanded, and field robots equipped with these technologies are hitting the market. Laser systems offer a precise, chemical-free alternative for weed control, potentially reducing long-term costs and environmental impact, though initial investment and performance efficiency are high and must be carefully considered against other methods.

This paper provides a detailed review of laser weeding technology and discusses its features, usage, research gaps, and comparisons with available conventional methods for weed control. Laser weeding is a new emerging technology; therefore, the cost of currently produced prototype laser weeders is highly variable and will not be presented for comparison. The mass production of laser weeding robots will decrease the price and help in making laser weeding economically competitive with current weed-control methods, and it may lead to replacing or supplementing commonly employed weed-control approaches.

2. Weeding Platforms

Several types of weeding platforms are available for field use that vary depending upon their structure, drive system, power source, working width, instrumentation, and weed control system (mechanical, flame, electric, etc.).

2.1. Autonomous/Self-Propelled

One of the common research approaches across many different studies was the development of a mobile laser weeding platform operating under some form of autonomy. Xiong et al. developed a prototype laser weeding robot comprising three main subsystems: a robotic platform, a machine vision system, and a laser targeting mechanism powered by a combination of a gasoline engine and batteries. The remote-controlled laser weeder featured a frame measuring 49.5 × 39.5 × 39.5 cm (length × width × height) and was mounted at the front of the robotic platform. The robot’s overall dimensions were 150 × 65 × 65 cm [16]. In North America, Carbon Robotics introduced two types of field-scale laser weeders. One is a tractor rear-driven laser weeder equipped with 42 high-resolution cameras [17]. The second is a self-propelled laser weeder powered by Nvidia GPUs with deep learning models mounted on an autonomous platform with a 74 hp diesel engine driving four hydraulic drive motors [18]. Zhu et al. created a self-guided laser weeder tailored for corn fields, taking into account factors such as planting area, inter-row spacing, and the robot’s mechanical design. The weeder was engineered with external dimensions of 550 × 450 × 400 mm (length × width × height) [19]. Wang et al. introduced a Hyper Weeding system which was integrated at the rear of a tractor, housed within the hood [20]. Within this setup, a multispectral vision system, based on a mono-camera [20], and a laser-loaded gimbal was positioned on a three-axis gantry, enabling adaptable relocation of the system in response to field conditions. Elstone et al. utilized a laser treatment systems specifically designed to target the intra-row area, whereas a non-laser weeding method was employed for the inter-row area [21]. Rakhmatulin and Andreasen developed an affordable, compact prototype platform for a laser-based weeding system, which was experimentally tested on a combination of couch grass (Elytrigia repens (L.) Desv. ex Nevski) and tomatoes [22].

The agricultural robot, often referred to as Agribot, is a robotic system employed in farming [23]. AVO is an autonomously operated weedkiller robot powered by solar photovoltaic (PV) panels and batteries with an auto navigation system and RGB cameras. This robot communicates through Wi-Fi for shorter distances and for longer distances using a 3G/4G system. It covers four rows in a single pass with a field capacity of 10 hectares a day [24]. There is a prevailing belief that advancements in robotics science and engineering will significantly transform the landscape of agriculture. Notably, global investment and research in this field are undergoing nearly exponential growth [25].

Table 1. Weeding robots that have been available in the market over the past nine years.

Robot Name	Origin	Scope of Work	Year	Systems	Sensors	Accuracy and Speed
Tertill [26]	-	Parks and gardens	2017	Mechanical	Capacitive sensors	80–95% and 0.5–1.5 ac/h
Laser Weeder [17]	America	-	2021	Laser	RGB camera	99% and 2 ac/h @ 1 MPH
EcoRobotix [27]	Switzerland	-	2015	Electrical	Camera	95% and 7 km/h
AVO [24]	Switzerland	Rapeseeds, bean, and cotton	2020	Chemical	RGB camera	95% and 10 ha/day
TerraSentia [28]	America	Soybean	2018	Laser	RGB camera	-
BoniRob [29]	Germany	Sweet beet	2015	Mechanical	Ultrasonic sensor and NIR cameras	91% and 5 km/h
WeedRobot 1 [30]	Japan	Rice field	2018	Mechanical	-	20 m/min
WeedRobot 2 [31]	Japan	Paddy	2018	Mechanical	-	71% and 500 m/s
AgBotII [32]	Australia	Thistle, feather-top Rhodes grass, and wild oats	2015	Mechanical	RGB camera	92.3% and 5–10 km/h

provides a summary of both commercial and research-based robots within the past eight years developed for weed control using a variety of end effectors.

2.2. Key Components

Laser source, optical system, control system, microprocessors, power supply, gantry and vision system, cameras, sensors, GPS, and mechanical drive motors are the key components in a laser weeder [16]. All these components are required to run and control laser weeders. Cameras are used to detect and identify objects and multiple types of crops and weed plants. Individual processing units are helpful to control the system by linking different devices on a single bus. To kill different weeds, an Arduino-based microcontroller boards can be used to control the entire system including the intensity of the laser through programming [33]. Edge AI cameras have been used for weed plant recognition [34]. Lithium batteries are commonly used as the main power source for small robots equipped with lasers [35]. Tertill is a mechanical weeding robot introduced by Franklin Robotics to control weeds in parks and home gardens which is equipped with capacitive sensors and trimmer string. Weeds were targeted by the combined action of robot wheels and string which disturbs the soil but this research did not report the detection of weed plants [26]. EcoRobotix developed an electrically controlled smart sprayer attached with cameras to detect and locate the crop and weed plants in row crops. It moved with a forward speed of 7 km/h and reported a field capacity of 4 ha/h [27]. TerraSentia is a small autonomous robot equipped with four HD RGB cameras and a laser to detect and collect data on under-canopy plants, i.e., maize and soybean. It has auto navigation (GPS) and communication systems [28]. Weed robot 1 was designed and tested to eliminate weeds from rice fields in Japan. This weeder was installed with four capacitive touch sensors to detect seedling and an azimuth sensor to provide directional guidance [30]. Weed robot 2 was developed for the eradication of weeds from rice and was equipped with a GNSS sensor, Raspberry pi 3, and an Arduino microcontroller. This robot communicated through Wi-Fi and was installed with magnetic, gyro, and accelerometric sensors to control the orientation and motion [31]. A prototype of a weeding platform called AgBotII was designed in Australia and tested on thistle, feather-top Rhodes grass, and wild oats. AgBotII showed 92.3% accuracy in weed detection and classification with various species [32].

2.3. Working Widths

Most of the laser weeders have small working widths, but a few weeders with large working widths over 6 m have also been reported. When considering the cultivation of a corn crop and the row spacing for corn planting, Zhu et al. engineered a weeding robot with specific external dimensions of 550 × 450 × 400 mm (length × width × height) based on its mechanical design [19]. The robot arm’s operational area measured 400 × 400 mm, and the weeding robot had a weight of 20 kg. The prototype for this laser weeding robot comprised three main subsystems: a robotic platform, a machine vision system, and a laser targeting mechanism, all powered by a combination of a gasoline engine and batteries [16]. A robotic laser platform with a working width of 65 cm was operated to kill the weeds in a greenhouse nursery [16]. A small prototype with a single laser was tested for the precise weeding of small weed plants targeting tissue fibrosis [33]. Working width and speed are the two main drivers of theoretical capacity, but field efficiency due to operating losses also drives actual capacity. Hence, working width is an important factor to increase the theoretical field capacity of equipment, which also helps to increase the field efficiency by reducing the number of headland turns in the field.

Table 2. Comparative summary of reviewed studies on laser weeding technologies.

Laser Type	Laser Parameters	Objective	Crops Studied, Weed Species	Weed Control Efficiency	Environmental Impact	Location	Results	Year
CO2 Laser	50 W	Cutting weeds with a CO2 laser	N/A, general weed species	~90%	Minimal disturbance	Denmark	Laser cutting was effective for multiple species, better with higher power levels	2008 [13]
Static laser	N/A	Development of a robot and fast path-planning for weeding	General crops, general weed species	High efficiency	Reduced herbicide use	China	A robot with path planning optimized for fast laser weed control	2017 [16]
Fiber Laser	70 W, 1070 nm	Precision laser weeding in vineyards	Grapevine, common thistle	88%	Reduced erosion	Spain	Useful for vineyards due to precision and no soil impact	2019 [36]
Diode Laser	40 W, 980 nm	Low-energy laser weeding for row crops	Tomato and pepper, foxtail and crabgrass	80%	No soil disturbance	South Korea	Effective for soft weeds but needed precision targeting	2020 [37]
Diode Laser	30 W, 808 nm	Laser weed control in rice fields	Rice, barnyard grass	85%	No harm to non-target plants	China	Best results at seedling stage; mild impact on water	2021 [34]
Nd Laser	60 W, 1064 nm	Evaluation of organic farming systems	Soybean, lambs quarters	92%	Reduced soil disturbance	Canada	Beneficial for organic farming systems	2022 [38]
Blue laser	Wavelength 445 nm	Blue laser weeding robot in corn	Corn	~85%	No damage to corn crops	China	Laser weeding system specifically designed for corn fields	2022 [19]
Diode Laser	Low-energy, continuous	Diode laser treatment to manage weeds in row crops	Soybean, maize, general weed species	~80%	Reduced soil disturbance	USA	Diode laser effective for row crops but power management required	2022 [39]
CO2 Laser	100 W, 10.6 µm	Thermal impact of lasers on weed control	Wheat, wild oat	93%	Minor carbon emissions	India	High power was needed for tougher species; good efficiency	2023 [40]

presents the comparative summary of reviewed studies on laser weeding technologies used over the years in the field of agriculture.

3. Weed Detection

Weed detection systems employing laser technology are versatile tools used across various weed control methods, including mechanical and chemical approaches. In response to the rising costs of labor and heightened concerns regarding health and environmental considerations, the adoption of site-specific weed management (SSWM) has gained considerable appeal. The initial and crucial phase in the establishment of SSWM involves the detection and recognition of weeds. Weed detection approaches are generally divided into two primary categories: machine learning (ML) and deep learning (DL). Figure 1 illustrates the differences between these two methods.

Previous weed recognition methods used traditional computer vision and machine learning techniques, relying on manually selected image features, such as shape, often requiring controlled conditions for capturing images [41]. However, with the success of deep learning, these manual features, which are laborious and time-consuming, are no longer needed. Deep learning can now automatically extract useful features directly from images, making the process more efficient and adaptable.

Deep learning and traditional machine learning have advantages and disadvantages. Deep learning offers higher accuracy due to its ability to process complex, high-dimensional data like images. Models such as convolutional neural networks (CNNs) are particularly effective in recognizing patterns in weed and crop images under various field conditions, leading to more accurate weed detection. Another advantage of deep learning is its ability to automatically extract features from raw data, eliminating the need for manual feature engineering that traditional machine learning requires. This makes it well suited for complex environments where weeds and crops might appear similar [14].

However, deep learning comes with notable challenges, such as the need for significant computational resources. These models require powerful hardware and large datasets to perform effectively, which can be a limitation in real-time, field-based agricultural applications [35]. Additionally, the data-hungry nature of deep learning makes it less feasible for cases where large labeled datasets are not available.

On the other hand, traditional machine learning models, such as Support Vector Machines (SVMs) and Decision Trees, have lower computational costs, making them more accessible for environments with limited processing power. These models can also work effectively with smaller datasets, which reduces the time and cost associated with data labeling [42]. However, their reliance on manual feature extraction limits their adaptability in diverse and dynamic agricultural environments. Moreover, traditional machine learning tends to have lower accuracy in weed detection tasks compared to deep learning, particularly in cases where weeds and crops have similar appearances [43]. Thus, while both methods have their place in agricultural applications, deep learning is becoming the more favored approach due to its superior accuracy and ability to handle complex data, despite its higher resource demands.

3.1. Methods of Weed Detection

This section offers a comprehensive review of machine learning techniques are increasingly used in the development of systems for classification and weed detection. Various approaches, including neural networks (NNs), recurrent neural networks (RNNs), convolutional neural networks (CNNs), and random forest regressions have been explored for this purpose [44,45]. For instance, weed detection systems utilizing the CNN architectures, such as ResNet152 and VGG16, have attained an impressive precision of 98.4% in the classification of leaf species. Although these results are promising, there remains room for further improvement. Advancing plant identification could be realized by incorporating more robust algorithms capable of accurately distinguishing a wider variety of leaf species, regardless of their color or shape [46].

To achieve precise weed elimination while avoiding crop damage, Quan et al. adapted Faster R-CNN, an object detection model, to pinpoint the locations of weeds in agricultural fields [44]. A precision rate of 97.71% was demonstrated by this model across diverse conditions, including varying weather and angles, and throughout the crop cycle. Despite these strong results, the study only focused on the detection speed of Faster R-CNN and a single cropping system (corn), that was limited to 7 frames per second, which presents real-time weed detection challenges [44].

Weed detection and identification is the most crucial part of laser weeding technology and requires embedded and single-board computers in mobile applications. An example of a single-board computer used for weed detection is the Raspberry Pi 3 Model B+, developed by the Raspberry Pi Foundation in Cambridge, UK. This device features a Quad-core Broadcom BCM2837B0 Cortex-A53 64-bit SoC running at 1.4 GHz and is equipped with 1 GB of LPDDR2 SDRAM [12]. This choice was driven by the device’s cost-effectiveness and compact form factor. The programming language used was Python 3.6, and the OpenCV 3.4.1 vision library from the University of Birmingham, UK, was utilized.

However, the limited resources of the Raspberry Pi 3 Model B+, with 1 GB of RAM and a processor speed of 1.5 GHz, make it challenging to implement deep neural networks for rapid plant recognition. To address weed detection, a combination of the Viola–Jones algorithm [45] and SqueezeNet [42] was utilized. The Viola–Jones algorithm, initially developed for rapid multi-view face detection, employs Haar feature-based cascades [45] and offers real-time image feature detection.

SqueezeNet, tailored for computer vision, is a type of compact deep neural network that was selected due to its small parameter count, making it suitable for the limited memory of the Raspberry Pi. The SqueezeNet model utilized 1 × 1 and 3 × 3 convolutional kernels and had a storage requirement of 5 MB [12]. Despite this setup, the time required to identify the target object in an image was around one second. This speed was deemed somewhat slow for efficient weed detection in typical field management conditions.

3.2. Methods of Machine Learning

In the early stages of research, outmoded machine learning (ML) algorithms were widely utilized by scholars for distinguishing between crops and weeds. A standard approach for weed detection using machine learning involves four essential steps: data collection, pre-processing, feature extraction, and classification, as identified in previous research [46]. Hamuda et al. conducted a detailed review of image segmentation techniques for plants from 2007 to 2016, highlighting the advantages and limitations of both threshold-based methods and color index-based approaches [47].

Means-based machine learning has become a favored approach in weed detection due to its low demands for computing power and data. This method’s ability to produce interpretable results has contributed to its widespread adoption. The visual features that help distinguish between weeds and crops can be grouped into four main categories: texture [48], color [49], shape, and spectral characteristics [50].

Unsupervised learning techniques have also been utilized in weed detection. Hall et al., for instance, proposed a clustering method that does not require prior knowledge, thereby removing the need to retrain systems for different weed species [51]. Due to the integral similarities between crops and weeds, particularly in the initial stages of growth, using a single feature for identification is often insufficient. Consequently, researchers have shifted toward multi-feature fusion methods, which have proven highly effective in the identification of weeds. The following section delivers an overview of multi-feature techniques employed in the past five years, emphasizing key developments and emerging patterns in the field.

3.3. Deep Learning Methods

The physical creation of features in machine learning requires domain expertise and knowledge, which can limit potential improvements in accuracy. However, with the rise in powerful computing resources and large datasets, neural networks are now able to automatically learn features and optimize weights throughout their layers, leading to significant enhancements in deep learning (DL) performance. Considering the current breakthroughs in DL, it is rational to apply this technique to weed detection using machine vision. Several established neural network architectures, such as DenseNet, GoogleNet, and ResNet, have achieved cutting-edge results in this field. Deep learning approaches are generally divided into two categories based on the type of input data: graph convolutional networks (GNNs) and convolutional neural networks (CNNs) [52].

Deep learning can be categorized based on the labeling of data into three main types: unsupervised learning, semi-supervised learning, and supervised learning [14]. Significant progress has been achieved, particularly in supervised and semi-supervised approaches. The following sections provide an in-depth examination of these two methods, highlighting their applications and contributions in recent years. This detailed review aims to capture recent emerging trends and advancements in the application of DL for the detection of weeds using machine vision.

Bargoti and Underwood evaluated three advanced detection algorithms, namely, Mask R-CNN, RetinaNet, and EfficientDet. Each model underwent training and testing for a binary problem on individual datasets. These networks, despite their distinctive features, share a foundational structure. Initially, a pre-trained classification network, known as the ‘backbone’, is utilized in a fully convolutional fashion to produce a dense representation of the input image [53]. Following this, Dollár et al. employed a modified version of the Feature Pyramid Network (FPN) to generate tensors with varying resolutions, enabling detection at multiple scales [54].

The models evaluated the presence of objects within predefined candidate rectangles, referred to as “anchors” which are utilized through parallel “head” modules. Specifically, a classification head is responsible for determining whether a subject belongs to the target group, while the “bounding-box refining” head adjusts the planned rectangle to improve align with the subject’s boundaries [55].

Mask R-CNN, an advanced model form of Faster R-CNN that incorporates selective object segmentation, operates using a two-stage model. A Region Proposal Network (RPN) filters potential anchors for further evaluation. During the training of the model, negative and positive subjects are selected to maintain a balance, and subject regions are processed in the second stage for bounding-box refinement and classification [56].

RetinaNet, similar to ‘ResNet-50’ CNN in its use of Mask R-CNN and Feature Pyramid Network (FPN) for multi-scale illustration, functions as a single-stage network with end-to-end training. It tackles the problem of class imbalance by implementing ‘Focal Loss’, a modified cross-entropy loss function that reduces the emphasis on ‘easy examples’ during training, allowing the model to concentrate on more challenging examples [57].

EfficientDet, another one-stage network like RetinaNet, employs ‘Focal Loss’ and achieves state-of-the-art results. It utilizes B4-EfficientNet as its backbone, known for higher ImageNet accuracy with fewer parameters and faster execution. EfficientDet also incorporates a modified Bi-FPN for improved feature fusion [58].

These three algorithms were selected for their reported accuracy and distinct architectural details, providing a comprehensive evaluation across different methodologies. Notably, other algorithms such as ‘You Only Look Once’ (YOLO) or ‘Single Shot Detectors’ (SSDs), focused on inference speed, were not chosen as they are less suitable for accuracy optimization [57].

In recent years, several studies have applied YOLO models to enhance weed detection in agricultural fields. A YOLOX-based blue laser weeding robot was developed specifically for corn fields, showing high precision in detecting weeds and distinguishing them from crops in real time, allowing for efficient weed removal with minimal damage to the crops [19]. A study focused on lettuce crops, using multispectral imaging in conjunction with YOLO to improve weed detection accuracy, outperforming traditional machine learning methods in complex field conditions [41]. Similarly, a lightweight, deep learning-based weed detection system was designed for a commercial laser weeding robot, utilizing YOLO for real-time detection across various field conditions, demonstrating YOLO’s adaptability and effectiveness in modern weed control systems [35]. These studies highlight YOLO’s growing role in achieving precise, real-time weed detection, and promoting sustainable agricultural practices. Some key performance metrics are presented in

Table 3. Comparative performance metrics of deep learning and traditional machine learning in weed control studies.

Method	Accuracy (%)	Precision (%)	Recall (%)	F1-Score (%)	Processing Time (s/Image)	Scalability
Deep Learning (CNN) [15]	95	93	92	92	0.02	High
Traditional ML (SVM) [43]	85	80	82	81	0.05	Moderate
Deep Learning (YOLO) [19]	98	96	95	95	0.01	High
Traditional ML (KNN) [42]	80	78	77	77	0.04	Low

for comparison.

3.4. Camera

Weed control systems have utilized a variety of cameras for detection, each with specific technical capabilities. One prominent camera used is the FLIR Blackfly 23S6C Gigabit Ethernet high-resolution camera, which features a 1920 × 1200 pixel resolution and a high dynamic range of 73.90 dB, ensuring detailed and clear imaging even under challenging light conditions. Paired with a Fujinon CF25HA-1 machine vision lens with a 25 mm focal length and a 450 × 280 mm field of view, this setup enables precise weed texture recognition at close ranges [34].

In another study, the Sony α-6000 camera, with a 24.7 MP resolution, was used for UAV-based weed detection. This camera provided high-quality images with an exposure time of 1/1250 s, ensuring rapid and accurate data capture. The camera was mounted on an octocopter platform, flying 1.5 to 3 m above the ground, taking images every second to detect weed species in real time [59]. These cameras, often optimized for agricultural environments, feature high frame rates of 10 to 60 frames per second, ensuring real-time processing capabilities crucial for deep learning-based weed detection models [60]. These studies highlight the role of advanced imaging technologies in improving precision agriculture.

4. Laser Technologies

A device known as a laser, which stands for “light amplification by stimulated emission of radiation”, operates as a complex apparatus generating light by means of optical amplification induced by the stimulated emission of electromagnetic radiation. This cutting-edge technology showcases unique features such as monochromaticity, coherence, directionality, and an elevated intensity through its interaction with diverse materials [61]. The laser beam emitted demonstrates singular traits, delivering substantial energy in a focused, narrow beam that diverges minimally, distinguishing lasers from typical light sources [62].

Lasers stand out due to their coherence; two sources of wave are coherent when sharing identical frequencies and waveforms. This coherence enables lasers to concentrate on extremely small areas, resulting in high irradiance, or to exhibit minimal divergence, concentrating power across significant distances [63]. The wavelength in a vacuum defines lasers, offering advantages such as customizable beam divergence and shape using optics for precise energy delivery in applications like weed control.

Moving beyond their underlying principles, lasers have permeated diverse industrial domains, serving purposes such as measurement, cutting, hole boring, etching, welding, and paint removal. Laser irradiation facilitates the meticulous surface adjustment of polymers with minimal surface impairment, contingent on the specific wavelength in use [33]. Certain laser types, including carbon dioxide (CO₂) lasers operating at 10,600 nm, have been instrumental in tissue burning due to their absorption by water in biological cells [64]. The micrometer-range laser wavelengths find application in corneal tissue removal and addressing abnormal blood vessels in diabetic patients, contributing to the prevention of associated blindness [65].

Various laser types employ distinct active media and emit at different wavelengths. Diode lasers, powered by a semiconductor as the active medium stimulated by electrical current, encompass wavelengths ranging from ultraviolet to infrared radiation. Fiber lasers, categorized as solid-state lasers featuring a doped glass fiber as the active medium, function within the visible and near-infrared light spectrum. CO₂ lasers, utilizing a gas mixture, typically nitrogen, helium, and carbon dioxide, are recognized for emitting in the far-infrared light spectrum [65,66]. The diverse applications and distinctive attributes of lasers emphasize their importance across numerous scientific, medical, and industrial domains.

Laser weeding technology offers significant potential across various cropping systems, but its application must be tailored to different environments. For row crops and permanent crops, the technology can be adapted for the precision targeting of weeds without disturbing the crop. In horticulture and controlled environments such as greenhouses, laser weeding systems need further refinement to ensure optimal performance in tighter spaces and under artificial conditions. By coordinating the design and operation of these systems to suit the specific needs of each cropping system, laser weeding can become a more versatile and effective solution in modern agriculture. Different types of laser weeding platforms and autonomous robots are shown in Figure 2.

4.1. Types of Lasers

Three primary laser classifications exist in the field of weed control, each characterized by unique attributes and applications including carbon dioxide lasers (CO₂ lasers), diode lasers, and fiber lasers [71]. Diode lasers function by exciting semiconductors through electrical currents, spanning a broad spectrum from ultraviolet to infrared wavelengths. Renowned for their adaptability, diode lasers find utility in telecommunications, medical interventions, and everyday gadgets like laser pointers. Fiber lasers, another category, leverage a doped glass fiber as the active medium, operating within the visible and near-infrared light spectrum. Fiber lasers are embraced in industry as the choice for manufacturing processes such as cutting and welding because fiber lasers offer substantial power and precise beam quality.

In the instance of both CO₂ and fiber lasers, the absorption of light energy by plants results in fatal damage. A CO₂ laser achieves the highest energy density by directly delivering energy to the plant’s surface. Conversely, a 2 µm fiber laser predominantly absorbs energy through the water inside the plant, causing heating across a broader plant area. A thulium-doped fiber laser operating at a 2 µm wavelength is advantageous for effective weed control because its radiation penetrates the epidermis rather than being absorbed only on the plant’s surface [72].

Furthermore, emerging approaches integrate plant species recognition tools with laser equipment on compact autonomous vehicles, introducing an innovative method for weed control. This fusion is vital for sustainable laser-based weed control, as the impracticality of treating entire areas with laser beams arises due to significant energy consumption. Each laser type plays a pivotal role in advancing precision applications across diverse scientific, medical, and industrial spheres [33].

4.2. Energy Requirements/Intensity

The power needs for laser-driven weed control are intricately linked to the electrical supply propelling the laser. This power is typically harnessed from fixed electrical grids for most applications, but mobile applications are much more limited. A pivotal element contributing to the efficiency of field robots in this context is the ability to efficiently utilize the electric energy in batteries or generated onboard the machine through the consumption of other fuels. The selection of laser wavelength emerges as a critical factor influencing thermal coupling and the least lethal doses for vegetation during weed control operations.

In an extensive study by Kaierle et al., focusing on the weed species Amaranthus retroflexes, it was unveiled that the overall energy demand was minimized when employing a wavelength of 10,600 nm, specifically with a CO₂ laser [73]. Conversely, wavelengths of 1908 nm, 940 nm, and 532 nm, corresponding to solid-state, fiber, and diode lasers, respectively, demanded doses of 230 J, 237 J, and 1400 J per weed for effective control, underscoring the importance of selecting appropriate wavelengths.

Research findings by Marx et al. emphasized the variability in energy requirements for lethal damage, and argued that factors such as weed species, laser spot position, plant growth size, spot area, and applied laser energy must be considered [74]. Effective weeding is optimally achieved when the meristem of the target plant is exposed during the cotyledon stage or the stage with two permanent leaves. Small weeds, particularly in these early growth stages, exhibit heightened sensitivity to lasers, requiring less energy for efficient treatment. In contrast, larger plants with multiple developed meristems have a higher potential for reestablishment after laser treatment [75].

A significant advantage of laser-based weed control, distinct from mechanical weeding, lies in its precision in targeting weeds in close proximity to crop plants without causing harm to leaf and roots [2]. This precision is achieved through the use of a small laser diameter and advanced optics that allow precise beam manipulation, ensuring that the meristem of the weed plant does not overlap with the crop plant. The optimization of energy usage in laser-based weed control is paramount, focusing on selectively targeting only those plants that pose a threat to crop yield or quality, thereby minimizing unnecessary energy consumption. Diode and fiber lasers are low-powered lasers ranging from 2 to 20 W which are easily operatable with a 12 VDC battery along with an additional pack to operate the system. In the comparison of herbicide and tillage methods, the sprayer requires the same amount of energy (Joules) but the mechanical weeder requires a high amount of power (J/s) to tilth and hoe the soil.

4.3. Targeted Area

Recent strides in artificial intelligence (AI) and the field of robotics have ushered in a new era of accuracy in identifying and locating weed and crop plants [11,22]. This technological leap allows for the precise steering of laser beams targeting the meristems of weed seedlings, enabling real-time laser control scenarios. Operating with a focus on a single spot and with restricted heat dispersion beyond the irradiated area, merely a small fraction of the field is exposed to the laser treatment. To illustrate, envision a beam of laser with a 2 mm diameter amidst 110 plants of weeds per meter square—this translates to approximately 0.31% of the total field area directly exposed [76]. This targeted precision minimizes exposure, ensuring that only the weed plant, not the soil, bears the direct impact. Such a focused approach stands out for its significant reduction in environmental impact compared to conventional methods like mechanical weeding, herbicide, and flame application. In the market, most of the mechanical weeders are unguided, which also affects the crop plants, while the target of herbicide droplets is adversely effective with sprayer and air velocity, which is the same in the case of flame application. Weeders driven by internal combustion engines produce more emission gasses.

The effectiveness of inducing harm to weed plants through lasers hinges on the beam of laser diameter. A smaller diameter concentrates the entire energy on a minute spot, amplifying its effectiveness [74]. However, navigating the fine line between efficacy and precision becomes crucial, as an excessively small diameter might not disturb enough cells to ensure the demise of the plant. This challenge is accentuated in field scenarios marked by uneven terrain, where hitting the meristem with precision becomes more intricate. Striking the right equilibrium between beam diameter and precision is pivotal for optimizing the efficacy of laser weed control technologies. In this delicate balance, the efficiency of laser-induced harm to weed plants finds its sweet spot.

A key challenge in laser weeding is not only detecting and discriminating weeds from crops but also accurately targeting the critical part of the weed, such as the meristem, for effective elimination [77]. While the article discusses target area identification, further emphasis is needed on the vision system’s role in pinpointing the precise attack zone. Developing a targeting system that ensures precision, speed, and reliability under varying field conditions remains essential. Future advancements in AI-driven vision systems and real-time processing will be crucial in meeting these demands for efficient and effective weed control.

4.4. Weeding Capacity

In the current landscape, the development of small autonomous vehicles tailored for laser weeding, which can also be used for tillage and herbicide applications, colloquially known as laser weeders, entails a design with a modest working width, for example, 2.2 m. Despite their compact size, these vehicles grapple with limitations in treating expansive areas when compared to their larger counterparts, such as sprayers boasting wider boom widths ranging from 24 to 40 m [71]. The intricacies of the real-time identification and precise targeting of both crop and weed plants pose challenges, potentially capping the process of weeding at a modest drive speed of 3.8–6.1 km.h⁻¹ [78]. In dense weed scenarios (>100 m²), further reductions in driving speed may become necessary. The optimization of laser weeding efficiency involves a delicate interplay between adjusting driving speed and laser power, where higher laser power demands less time for weed control. However, the inherent constraints of the laser’s detection and targeting system may impede efforts to escalate the operating speed.

In contrast, conventional herbicide applicators such as self-propelled sprayers can operate at speeds ranging from 6 to 12 km/h and, notably, surge to 40 km/h in regions like Australia and the United States [37]. Mechanical weed control, employing diverse implements with varying working widths, often operates at speeds ranging from 4 to 12 km/h, and the advent of autonomous vehicles provides the flexibility for 24 h operations, mitigating certain capacity challenges.

Efficient weed control is a common goal across various methods, be it laser weeding, herbicide application, mechanical weeding, or flame weeding. The pivotal factor for success lies in targeting weed plants when they are small, a principle applicable across diverse weed control approaches. The efficacy of weed control is also intricately tied to environmental variables, including weather conditions and the developmental stage of both weeds and crops. Although research on the impact of environmental conditions on laser performance is limited, indications suggest that lower temperatures and wet plant conditions may undermine efficacy due to the reliance on heat-based control mechanisms [79]. Laser weeding offers high precision with minimal soil disturbance, particularly in dense crops, though it has higher energy consumption and costs compared to other methods [18]. Flaming, precision hoeing, and electrocution are effective alternatives but come with drawbacks such as CO₂ emissions, soil disturbance, and lower precision [80].

To overcome challenges related to capacity, the proposition of deploying a fleet of laser weeders concurrently working within the same field emerges. Alternatively, a synergistic approach entails combining laser weeding with other methods, albeit at the cost of diminishing their individual advantages. For instance, in row crops, mechanical weed control could be judiciously implemented between rows, allowing the laser weeder to focus on targeting weeds near crop plants in rows where mechanical intervention might pose risks to the delicate roots or leaves [81]. This adaptive strategy holds particular relevance in minor crops devoid of herbicide approvals and in herbicide-sensitive crops, with chemical treatment between rows using hooded sprayers and non-selective herbicides to protect crops while laser weeding takes center stage within the crop rows along with flaming and finger weeders.

4.5. Dwell Time

Dwell time is the length of time the laser remains on the same portion of the weed. Selecting a proper dwell time should result in plant tissue damage or kill. Diode-type lasers typically take 0.5–1.5 s to harm a plant, and this time duration determines the level of harm [39]. A tracked mobile platform, combined with a weed detection module and a robotic arm equipped with a laser emitter, produced a successful kill in 0.2 ms⁻¹ using the blue laser beam [19]. Three types of lasers were applied (0.3 W, 1 W, and 5 W) to kill the weed plants, and the 0.3 W laser required more time compared to the 5 W laser [22]. A weeding robot tested to kill the weeds in tomato plants used a CO₂ laser as a light source and completely destroyed the plant in 1 s [74]. A laser weeder-equipped AI system operated to destroy certain types of weed plants required 0.64 s of dwell time per weed plant with a forward speed of 30 mm/s [16]. A laser system using a YOLOv5 model in a cornfield completed the task in 54.44 h for one hectare with an average speed of 0.09 m per second [35]. A fiber laser was tested to check its effectiveness on the weeds and it completed the job in 1.5 s with a 25 W fiber-coupled diode laser with an efficiency of 93.3% [37]. A laser weeding gimbal based on two degrees of freedom was evaluated at a distance of 535 mm from the target weed. The weeding efficiency was recorded at 0.72 s/weed with a dwell time of 0.64 s and a tracking speed of 0.1 m/s [20].

To increase the feasible working time of laser robots in the field, several strategies can be considered. Enhancing battery life or incorporating solar panels can extend operational hours. Additionally, optimizing energy efficiency through improved power management systems and using lightweight materials for the robot’s design can reduce energy consumption. Another potential solution is implementing autonomous recharging stations in the field, allowing robots to recharge without manual intervention, thus minimizing downtime [82]. Advances in AI and machine learning could also help optimize route planning, reducing time and energy spent on non-essential tasks.

4.6. Efficacy of the Laser

The efficacy of laser-based weed control is intricately tied to several factors, with the diameter of the laser beam playing a pivotal role in determining its impact. Laser beams with smaller diameters concentrate energy on a smaller spot, enhancing their effectiveness in targeting weed plants [83]. However, a delicate balance must be struck, as excessively small diameters may fail to affect enough cells to ensure the demise of the targeted plant. This challenge is particularly pronounced in field scenarios with uneven terrain, where the precise targeting of the meristem becomes increasingly difficult.

The precision of laser weeding is a standout feature, allowing for the close targeting of weed plants without causing harm to the leaves and roots of neighboring crop plants [14]. This precision is achieved through the deployment of a small laser diameter and advanced optics, enabling the manipulation of the beam. The ability to avoid damage to crop plants while effectively targeting weeds is a distinctive advantage of laser weeding over traditional unguided mechanical methods.

To optimize efficiency, it is crucial to strike a balance between the diameter of the laser beam and its precision in field scenarios. While a smaller diameter allows for more focused energy delivery, it also increases the challenge of hitting the weed plant’s meristem accurately, especially in areas with rough terrain [22]. Therefore, finding the optimal combination of beam diameter and precision is essential for maximizing the efficacy of laser weed control technologies.

In comparison to chemical weeding, where the large-scale exposure of the field to treatment is common, laser weeding minimizes environmental impact by selectively targeting only the weed plants that pose a threat to crop yield or quality [12]. The targeted approach significantly reduces the area exposed to the laser treatment, mitigating the risk of unnecessary energy consumption. Moreover, the ability to treat weed plants in close proximity to crop plants without causing harm to the latter contributes to the overall efficiency of laser weeding in agricultural applications.

The use of laser weed control can be most problematic in crops that have dense foliage or closely spaced plants, such as wheat, leafy greens, or crops like lettuce and spinach. In these crops, the laser system may have difficulty distinguishing between the crop and the weeds, especially when the weeds grow in close proximity to the plants. Additionally, crops with irregular growth patterns or those grown in highly variable conditions might pose challenges for accurate weed detection. Further advancements in AI and deep learning-based image recognition could help mitigate some of these challenges in the future.

4.7. Safety

Safety considerations are paramount in the deployment of laser weeders, especially in the context of autonomous movement and navigation through fields. While autonomous vehicles, including laser weeders, have gained acceptance in certain experimental locations for on-road use, they pose unique challenges when operating in agricultural environments. Small autonomous agricultural robots, such as laser weeders, are subject to safety guidelines to ensure their responsible use on both private and public properties [84,85].

One crucial safety concern associated with laser weeders is the potential for the emitted collimated laser beam to generate high levels of energy, leading to heat production upon contact with surfaces. This heat can pose a fire hazard, particularly in fields containing dry materials like straw, leaves, or other organic matter. Continuous surveillance during and after laser weeding operations is essential to detect any signs of uncontrolled heating or fire [86]. Various sensors, such as smoke sensors and cameras, can be mounted on the vehicle to monitor and respond to potential risks, providing a layer of safety during and after the weeding process.

Beyond fire risks, laser beams pose potential harm to both humans and animals in the vicinity. Integrating infrared cameras and sensors onto the vehicle is crucial to detect and halt its movement when approaching humans, animals, or obstacles, ensuring a safe distance is maintained. The danger lies in the laser wavelength, where exposure to laser light can cause irreversible damage, including blindness. Visible and near-infrared laser light within the 400–1400 nm range presents a serious threat to the retina, with potential for permanent damage. Laser light in the ultraviolet (UV) or far-infrared (FIR) spectrum range can harm the cornea or lens. Protective measures, such as laser-protecting glasses (O.D 6+) tailored to specific wavelengths, become essential for operators working in close proximity to the laser weeder. Moreover, employing clothing, gloves, screens, and curtains is critical for skin protection and preventing laser beams from escaping the target area due to reflections, ensuring comprehensive safety during laser weeding operations.

Moreover, precautions must be taken to prevent beams of laser from escaping the working area due to reflections from stones, sand, or other reflective items. Screens and curtains can be installed to contain the laser beams within the intended treatment area, reducing the risk of unintended exposure and enhancing overall safety during laser weeding operations [12]. It is crucial to emphasize that laser beams may be invisible, and safety measures should be designed to account for these potential hazards while ensuring the well-being of operators, bystanders, and the environment.

4.8. Effect on Soil

Laser weeding presents advantages in terms of its impact on soil compared to traditional methods such as mechanical weed control and herbicide application [15]. Soil compaction, a common issue caused by larger and heavier tractors, is minimized with small, relatively light autonomous vehicles designed for laser weeding, other physical weed control methods, such as flaming, electrocution, and mechanical hoeing, can also be implemented using similar light platforms. These vehicles, equipped with laser technology, do not compress the soil like heavy tractors and implements used in conventional methods like long-spanned boom sprayers and 12-row inter-cultivars. Many farmers already use these methods with lighter tractors specifically to avoid the compaction issues associated with larger, heavier machinery. Therefore, the advantage of reducing soil compaction is not exclusive to laser weeding systems. A broader consideration of various physical weed control methods reveals that multiple approaches can be employed with lightweight equipment, offering viable alternatives to traditional, heavy machinery-based methods.

While laser weeding primarily heats a specific region of the soil surface, mitigating the impact on beneficial organisms such as predatory ground beetles, spiders, and earthworms, mechanical weed control, despite its efficacy, has adverse effects on these soil-dwelling organisms [87]. Additionally, mechanical weed control contributes to the unnecessary mineralization of organic matter, the leaching of plant nutrients, and the stimulation of dormant weed seed germination. Laser weeding’s focused approach minimizes the broader impact on beneficial organisms, presenting a reduced risk of inducing widespread weed seed germination.

Laser weeding’s limited interaction with the soil surface, both in terms of treated area and duration (milliseconds), contrasts with herbicide application, which typically covers the entire field area [79]. Herbicides may pose risks to various organisms over an extended period, including potential leaching into surface- and groundwater. In comparison, laser weeding’s localized impact, focused on the immediate treatment spot, allows for the rapid recolonization of microorganisms in the treated area [88,89]. While laser weeding may require multiple treatments throughout the season, especially for weed species with specific temperature requirements for seed germination, its precision and reduced impact on the overall soil environment make it a promising and environmentally friendly alternative in weed control strategies.

5. Summary/Conclusions

In conclusion, laser weeding emerges as a promising and environmentally friendly alternative in weed control, offering distinct advantages over traditional methods such as mechanical weeding and herbicide application. The technology’s use with small, lightweight autonomous vehicles minimizes soil compaction, a concern commonly associated with heavier tractors. Laser weeding’s targeted approach, focusing on small areas for brief periods, reduces its impact on beneficial organisms on the soil surface, distinguishing it from the negative effects typically observed in mechanical weed control. Furthermore, its minimal interaction with the soil surface reduces the risk of triggering widespread weed seed germination, providing a more controlled and precise solution.

However, it is important to acknowledge that other physical weed control methods, such as spraying, flaming, electrocution, and mechanical hoeing, can also be mounted on small, lightweight platforms, effectively reducing soil compaction. Farmers have already been using these technologies with lighter tractors for this purpose. Therefore, while laser weeding offers certain benefits, it is not the only solution for minimizing soil disturbance.

While multiple treatments with laser weeding may be required throughout the growing season, its ability to selectively target small areas for short durations helps minimize disruptions to the soil ecosystem. This contrasts with herbicide applications, which impact organisms across the entire field and can lead to prolonged environmental consequences. Additionally, laser weeding supports long-term sustainability goals by reducing soil disturbance and minimizing chemical inputs. Its precision, coupled with the rapid recolonization of microorganisms in treated areas, highlights its reduced impact on the overall soil environment, making it a valuable tool for modern, eco-conscious agriculture.

Despite these advantages, laser weeding technology is still in its early stages and requires advancements in several areas. Increasing the field capacity and reducing dwell time from hours per hectare to minutes per hectare are crucial for improving efficiency. Automated navigation systems must also be refined to function across a range of cropping systems, including row crops, permanent crops, and horticulture. Real-time decision-making capabilities need further development to ensure precision even in dense crop conditions, such as those found in greenhouses or controlled environments.

Energy efficiency is another challenge, particularly for mobile platforms operating in the field. Innovations in power supply systems, such as improved battery storage or integration with solar technologies, are essential. Additionally, expanding the accessibility of laser weeding systems to small and medium-sized farms, while establishing regulatory and safety standards, is key for widespread adoption.

Ultimately, laser weeding presents a progressive solution in sustainable farming practices, offering precision and reduced environmental impact. However, it should be viewed as part of a broader integrated weed management strategy that includes other physical and agronomical methods. Addressing the current challenges through targeted research, innovation, and collaboration will be vital for realizing the full potential of laser weeding in modern agriculture.