Progress and Challenges in Research on Key Technologies for Laser Weed Control Robot-to-Target System

Abstract

The development of precise and sustainable agriculture has made non-chemical, highly selective laser weed control technology a hot research topic. The core of this technology lies in the overall performance of the targeting system, which consists of three key technologies, namely, target identification, dynamic positioning, and precise removal, which are interrelated and jointly determine the overall performance of the weed control system. In this paper, the key technologies of the targeting system are systematically analyzed to clarify the coupling relationship among the technologies and their role in performance optimization. This review systematically compares the mainstream recognition algorithms for the needs of laser weeding for specific parts, reveals the performance bottleneck of the existing algorithms in the laser weeding environment, and points out new research directions, such as developing weed apical growth zone recognition algorithms. The influence of laser beam control technology on weeding accuracy is analyzed, the advantages of vibroseis technology are explored, and the applicability problems of existing vibroseis technology in farmland environments are revealed, such as the shift of irradiation point caused by ground undulation. The key laws of laser parameter optimization are summarized, guiding the optimal design of the system. Through the systematic summary and in-depth analysis of the related research, this review reveals the key challenges facing the development of laser technology. It provides a prospective outlook on the future research direction, aiming to promote the development of laser weed control technology in terms of high efficiency, precision, and intelligence.

1. Introduction

Weeds are a persistent and challenging biological threat to agricultural production with high adaptability and reproduction capacity, posing a continuous risk to global food security [1,2,3]. Although chemical herbicides can rapidly control weed infestations, their overuse has led to a series of ecological crises, including soil degradation [4,5,6], sharp reductions in biodiversity [7], and the proliferation of herbicide-resistant weeds [8]. Data show that more than 3 million tonnes of herbicides are used globally each year [8], leading to the development of resistance in more than 500 weed species, involving 272 weeds and 168 herbicides in 100 crops in 72 countries [9]. These data indicate that traditional chemical weed control methods present a serious sustainability challenge.

To cope with this predicament, the agricultural and scientific research fields have actively explored alternatives to chemical herbicides [10], among which mechanical weed control [11,12,13] and biological control technologies [14,15] have gradually gained attention. Although these technologies have made breakthroughs in environmental friendliness, they are limited by significant technical shortcomings. Mechanical weeding methods may lead to high seedling injury rates [16], biological control technologies are costly and environmentally incompatible [17], and manual weeding is difficult to apply on a large scale due to labor costs of more than 30% [18]. Therefore, there is an urgent need to develop new, efficient, and environmentally friendly weed control technologies.

In this context, laser weed control technology is regarded as a key technological path to breaking through the existing dilemma due to the unique advantages of non-contact extermination, zero chemical pollution, and intelligent operation [19,20,21]. This technology precisely strikes the apical meristematic tissue of weeds through the directional emission of high-energy laser beams, and the laser energy is absorbed by the meristematic tissue cells and converted into high thermal energy, which destroys the cellular structure through the thermal effect and realizes the rapid extermination of the weeds [19,20,21]. The key to achieving this technology lies in the high-precision targeting system, which completes the weed removal operation through the three-level structure of ‘target identification—dynamic positioning—accurate removal’, and its technical performance directly determines the practical level of laser weed control technology. Therefore, analyzing the technical requirements of the targeting system is of great significance in promoting the development and application of laser weeding technology.

This paper focuses on the three key technologies of laser weeding targeting system, focuses on the research progress of the core links, such as target identification algorithm, laser beam control structure, and laser parameter regulation, reveals the key issues restricting the development of the technology, and provides valuable references for the future development of laser weeding technology.

2. Methodology

The literature in this review was obtained from the Web of Science database, Knowledge Network, and related professional websites. The objective of the literature search was to obtain the research results in the field of laser weeding from 1990 to 2025, and the types of articles covered included reviews, research papers, books, and conference papers. The search interface was set up with ‘laser weeding’, ‘weed identification’, ‘precision agriculture’, ‘weeding robots’, ‘laser weeding’, and ‘laser weeding’ as the search terms. Identification included ‘precision agriculture’, ‘weed robots’, ‘non-chemical weed’ and so on. To understand the development process of laser weed control technology, this study used bibliometrics and scientific mapping methods [22] to analyze the literature on smart weed control in the core database SSCI of the Web of Science (WOS) platform. Using CiteSpace 6.3. R1 (64-bit) Basic software, the number of publications, year, country, and keyword co-occurrence maps were obtained.

Figure 1 shows the countries with no less than 10 publications in the field of laser weeding. From the figure, it can be observed that China has the most active research investment in this field, with the highest number of publications, followed by the United States, showing the continuous attention of these two countries in the research and application of laser weeding technology. In addition, countries such as Spain, India, Germany, and Australia also maintain high research activity in this field, reflecting the fact that laser weeding, as an important part of precision agriculture technology, has received extensive attention worldwide.

Figure 2, Figure 3 and Figure 4 show the keyword co-occurrence maps of the three key technology modules of ‘target identification’, ‘dynamic localization’, and ‘precision removal’ respectively, which reveal the research hotspots and development trends of each sub-field. The keywords in the map are indicated in different colors. The keywords are shown in different colors to indicate their relevance, ranging from purple (weakest relevance) to red (strongest relevance). From the visualization results, high-frequency keywords, such as ‘precision agriculture’, ‘deep learning’, ‘classification’, among others, are widely used. Other high-frequency keywords are widely distributed in the three maps, reflecting the high degree of relevance and intersectionality of these topics in the research process of laser weed control technology. These keyword co-occurrence maps not only reveal the research focus and technological evolution direction of laser weed control technology in different links but indicate that there is a close synergistic relationship between the various technical modules, reflecting that laser weed control research is gradually developing toward systematization and intelligence at the theoretical and application levels. This trend is highly compatible with the actual demand for efficient, precise, and green weeding technology in current agricultural production, which further verifies the broad application prospect of laser weeding technology in modern agriculture.

Specifically, in the literature related to target recognition, keywords such as ‘machine learning’, ‘artificial intelligence’, and ‘identification’ frequently appear, indicating that this field is increasingly relying on computer vision and AI technologies to enhance the accuracy and efficiency of weed identification. In the area of dynamic positioning, the keywords ‘accuracy’, ‘robot’, and ‘system’ have emerged as research focal points, reflecting a current trend toward robotic technology and system integration, to achieve precise positioning and real-time response in dynamic environments. Meanwhile, in studies on precise removal, high-frequency keywords remain centered on ‘weed control’, ‘imagery’, and ‘robot’, with a noticeable lack of in-depth exploration of the underlying mechanisms and response characteristics of laser weeding. This situation suggests that future research should focus on strengthening the study of the interaction mechanisms between lasers and plant tissues, to reveal the intrinsic patterns of biological responses induced by laser irradiation, thereby providing a more scientific theoretical foundation for optimizing precision laser weeding technology.

In summary, these keyword co-occurrence diagrams not only reveal the research focus and technological evolution direction of laser weed control technology in different aspects but indicate the close synergistic relationship between various technological modules, reflecting that the laser weed control research is gradually developing towards systematization and intelligentization at the theoretical and application levels. This trend is highly compatible with the actual demand for efficient, precise, and green weeding technology in current agricultural production, which further verifies the broad application prospect of laser weeding technology in modern agriculture.

3. Progress of Key Technology Research on Targeting System

The development of laser weed control technology as an emerging tool for precision agriculture is shown in Figure 5. The technology germinated in the 1990s [21], and at the beginning of the 21st century, researchers carried out a series of basic experiments on laser–plant interactions in a laboratory environment [20,21,23]. Regulating different parameters of the laser irradiation of weed samples allowed for an in-depth study of its ability to control weeds, thereby verifying the feasibility of laser weed control, and laying a theoretical foundation for the technology towards practical application [24,25]. In recent years, rapid advances in cutting-edge technologies, such as artificial intelligence and automated robotics, have provided a strong impetus for laser weed control technology [26,27], resulting in significant breakthroughs in target identification, dynamic localization, and efficient removal. With the deepening of the research, the successful development of laser weeding prototypes has further promoted the development of this technology towards application [28]. Modern laser weeding equipment is capable of autonomously recognizing the ability of weeds and crops to accurately locate and efficiently remove weeds [29]. The advancement of this technology has not only brought revolutionary changes in agricultural weeding methods but has provided new directions and opportunities for the development of agricultural automation and intelligence.

Although laser weed control technology has made some progress in recent years, it is still in the early stage of development, and it still faces many challenges in system performance optimization and large-scale application [30]. The core of laser weeding technology is to build a targeting system that integrates target identification, dynamic positioning, and precise removal, and its overall performance directly depends on the seamless integration and synergistic operation between these three links [20]. As shown in Figure 6, the system first relies on deep learning and other intelligent perception technologies to capture the distribution of weeds in real-time in the farmland environment and accurately distinguish between crops and weeds, as well as to accurately identify the top growth area of weeds. Immediately after that, the recognition results are transmitted to the dynamic positioning module, which uses real-time position data to calculate the optimal irradiation path and angle of the laser beam and quickly drives the laser control unit to make adjustments to achieve accurate tracking of the target area. Finally, in the precision removal phase, the system regulates the timing, duration, and energy parameters of the laser beam based on the outputs of the previous two phases to ensure that the laser beam acts accurately on the target area. Overall, the realization of this highly efficient weed control process relies on the organic integration of three core indicators: the efficient identification of the top area of the weed, the precise positioning of the laser beam in space, and the fine-tuning of the laser parameters and the response of the weed. These three elements are interdependent and closely coupled, and together form the basis for the efficient and precise operation of the laser weed control system, which is also a key challenge for the current system integration and optimization.

To further clarify the vein of technological development, this paper systematically combs and compares the relevant review literature published in recent years (see

Table 1. Summary of review articles related to laser weed control technology.

Source	Country	Article Contributions
Yaseen and Long (2024) [32]	USA	Provided a comprehensive review of laser weeding technology, focusing on various laser platforms, their functionalities, applications, limitations, and key areas for improvement (e.g., reduced dwell time, automation, energy efficiency, affordability, and safety).
Zhang et al. (2024) [41]	USA	Reviewed the application of lasers and optical radiation in weed control, discussing their mechanisms, models, technical challenges, and future directions.
Krupanek et al. (2024) [42]	Poland	Provided a lifecycle environmental impact assessment of laser weeding robots, revealing potential environmental challenges in real-world applications.
Andreasen et al. (2022) [45]	Denmark	Explored the feasibility and pros/cons of using small autonomous vehicles with laser technology for weed control, emphasizing advantages in reducing non-target organism impacts and preserving soil health.
Upadhyay et al. (2024) [31]	USA	Summarized the research progress in laser weeding technology, including various laser types (e.g., CO2 lasers), robotic applications, and field trial results.
Jiao et al. (2024) [33]	China	Summarized research progress on laser weeding technology, covering different laser types (e.g., CO2, diode, and fiber lasers), platforms (robots and drones), and application examples.
Lytridis and Pachidis (2024) [34]	Greece	Mentioned the use of machine vision in laser weeding robots for weed detection.
Gao and Su (2024) [35]	China	Discussed the principles of laser weeding technology, application cases, effectiveness evaluation, and future development directions.
Muchhadiya et al. (2024) [36]	Gujarat	Introduced the integration of AI and robotics in laser weeding technology, highlighting its principles, advantages, and potential applications.
Mohan et al. (2023) [37]	India	Described the principles, advantages, and application scenarios of laser weeding technology.
Christensen et al. (2009) [38]	Denmark	Emphasized the combination of laser weeding technology with AI, computer vision, and deep learning.
Slaughter et al. (2008) [30]	United States	Mentioned laser weeding as a potential alternative to chemical herbicides, among other physical methods of weed control.
Xu et al. (2025) [39]	China	Presented the development history and research progress of laser weeding technology, including equipment design, algorithm optimization, and expanded application scenarios.
Tran et al. (2023) [40]	Belgium	Provided a comprehensive assessment of the implementation potential of laser weeding technology in Europe.
Michaliszyn-Gabryś et al. (2024) [43]	Poland	Evaluated the market acceptance and implementation barriers of laser weeding technology from the perspective of farmers, highlighting regional differences in acceptance and demand.
Qu and Su (2024) [44]	China	Reviewed the application of deep learning in agricultural equipment, showcasing the potential of laser weeding in smart agriculture.
Abouziena and Haggag (2016) [46]	Egypt	Highlighted the potential of laser weeding as an emerging technology, but noted that it is still in the developmental stage and has not yet been widely applied in large-scale agricultural production.

), and summarizes the core viewpoints of each publication on laser weed control technology [30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46]. These studies generally affirmed the potential of laser weed control technology in enhancing agricultural productivity, reducing chemical pesticide dependence, and promoting the development of green agriculture [31,32,33,34,35], and provided a good foundation for subsequent studies. However, a systematic analysis revealed that only four reviews have provided relatively complete descriptions of laser weed control technology (the first four rows in the table), which provide in-depth analyses in terms of platform types, technical challenges, environmental challenges, and feasibility, respectively, while the rest of the literature treats laser weed control technology as a comparison rather than the main body of research. Against this background, this paper provides a detailed analysis of each key technology of the technical architecture of the targeting system, aiming to provide a more targeted reference for the optimization and practical application of laser weeding technology.

3.1. Target Identification

As the key front link to achieve precise operation in the laser weeding system, the performance of the target recognition module directly determines the overall positioning accuracy and operational reliability of the system. Most of the current studies adopt the YOLO model and its variants as the core technology for weed identification in laser weeding (see Table 2). For example, YOLOv5 [47,48] has met the real-time processing requirements in terms of inference speed (0.4 s/frame) and outperformed the traditional color segmentation method (0.063 s/frame), but there is still room for improvement in its robustness and accuracy in unstructured scenes. Further, the DIN-LW-YOLO model [49] effectively improves the multi-scale target detection performance by introducing a dual-task detection mechanism, achieving a recognition level of 88.5% for region mAP and 85.0% for point target mAP [49]. However, in general, the current research focuses on the optimization of detection performance indexes for weed and crop differentiation, and less on the intrinsic relationship between recognition results and weed control efficacy from the perspective of the task requirements of laser weeding. The core mechanism of laser weeding relies on the accurate irradiation of the apical growth zone of weeds [20], and only the distinction between crops and weeds is not enough to meet the actual operational requirements of laser weeding, The recognition system needs to be further refined to the accurate recognition of the apical growth zone, such as the fusion of YOLO instance segmentation and image processing technology, by first extracting the contour of the weed, and then obtaining the weed contour center of mass method [50]. Such methods include the skeleton extraction method [51], the stem emergence point recognition algorithm [52], the improved center of mass detection algorithm [53], among others. The related recognition results are shown in Figure 7, and Table 3 provides an exhaustive classification and description of each type of method.

In addition, the effectiveness of laser weed control varies significantly among different weed species. Studies have shown that dicotyledonous weeds are more sensitive to laser energy than monocotyledonous weeds, and are more likely to be effectively removed at the same dose [24]. Therefore, to achieve more efficient weed control, the system needs to have the ability to dynamically adjust the laser output energy according to the weed species. As the first part of the laser weed control process, target recognition should not only achieve the distinction between crops and weeds but further refine the identification and classification of the weed types. The current recognition algorithm research in this direction is relatively weak, and there is still a lack of high-precision models that can meet the needs of practical applications. There is an urgent need to invest more research resources and technical strength in the development of algorithms and model training.

Table 2. Status of recognition algorithms used in laser weeding robots.

Source	Country	Target of Detection	Sample Size	Algorithm	Performance
Zhao et al. (2025) [49]	China	Strawberry seedlings (crop) and weeds	2153	DIN-LW-YOLO	Average mAP for region and point detection: 88.5% and 85.0%, respectively.
Zhu et al. (2022) [54]	China	Weeds and corn seedlings (crop)	15,000	YOLOX	Average detection rates for corn seedlings and weeds: 92.45% and 88.94%, respectively.
Xiong et al. (2017) [55]	United Kingdom	Dicotyledon and clover	_	Color Segmentation Algorithm/Shape Extraction Algorithm	Processing time for color segmentation: 0.0958 s (based on area) and 0.0251 s (based on erosion and dilation); Shape extraction: 0.063 s.
Hussain et al. and Fatima et al. (2023) [47,48]	Pakistan	Three crops (okra, bitter gourd, sponge gourd) and four weed species (horseweed, Herb Paris, grasses, small weeds)	9000	YOLOv5	Average mAP: 88%, Inference time: 0.4 s.
Qin et al. (2024) [56]	China	Veronica didyma (weed)	234	YOLOv7	Accuracy: 94.94%, Recall: 95.65%, mAP@0.5: 0.972.

Table 2. Status of recognition algorithms used in laser weeding robots.

Source	Country	Target of Detection	Sample Size	Algorithm	Performance
Zhao et al. (2025) [49]	China	Strawberry seedlings (crop) and weeds	2153	DIN-LW-YOLO	Average mAP for region and point detection: 88.5% and 85.0%, respectively.
Zhu et al. (2022) [54]	China	Weeds and corn seedlings (crop)	15,000	YOLOX	Average detection rates for corn seedlings and weeds: 92.45% and 88.94%, respectively.
Xiong et al. (2017) [55]	United Kingdom	Dicotyledon and clover	_	Color Segmentation Algorithm/Shape Extraction Algorithm	Processing time for color segmentation: 0.0958 s (based on area) and 0.0251 s (based on erosion and dilation); Shape extraction: 0.063 s.
Hussain et al. and Fatima et al. (2023) [47,48]	Pakistan	Three crops (okra, bitter gourd, sponge gourd) and four weed species (horseweed, Herb Paris, grasses, small weeds)	9000	YOLOv5	Average mAP: 88%, Inference time: 0.4 s.
Qin et al. (2024) [56]	China	Veronica didyma (weed)	234	YOLOv7	Accuracy: 94.94%, Recall: 95.65%, mAP@0.5: 0.972.

Figure 7. Potential algorithm examples for identifying the apical growth region of weeds: (a) An example of the F-YOLOv8n-seg-CDA algorithm detecting the apical growth region of weeds; the red dot indicates the recognition result [50]. (b) An example of the seedling stem emergence point recognition algorithm; the red dot indicates the recognition result [52]. (c) An example of the skeleton extraction algorithm identifying the central point; the green dot indicates the recognition result [51]. (d) An example of the LettWd-YOLOv8l algorithm detecting the central point of lettuce; the red dot indicates the recognition result [57]. (e) Example of combining 2G-R-B with a connectivity element to detect lettuce centroids; in the figure, the solid line indicates the upper boundary, the dashed line indicates the lower boundary, the ‘*’ indicates the center position of the acquired crop, and the red cross indicates the center of mass of the connectivity domain of the plant [58]. (f) An example of the SE-YOLOv5x algorithm detecting the seedling stem emergence point; the red dot indicates the recognition result [59]. (g) An example of the CNN-based PSEP localization algorithm for locating the seedling stem emergence point; the red dot indicates the recognition result [60]. (h) An example of the improved centroid detection algorithm identifying the corn canopy; the red box indicates the recognition result [53].

Figure 7. Potential algorithm examples for identifying the apical growth region of weeds: (a) An example of the F-YOLOv8n-seg-CDA algorithm detecting the apical growth region of weeds; the red dot indicates the recognition result [50]. (b) An example of the seedling stem emergence point recognition algorithm; the red dot indicates the recognition result [52]. (c) An example of the skeleton extraction algorithm identifying the central point; the green dot indicates the recognition result [51]. (d) An example of the LettWd-YOLOv8l algorithm detecting the central point of lettuce; the red dot indicates the recognition result [57]. (e) Example of combining 2G-R-B with a connectivity element to detect lettuce centroids; in the figure, the solid line indicates the upper boundary, the dashed line indicates the lower boundary, the ‘*’ indicates the center position of the acquired crop, and the red cross indicates the center of mass of the connectivity domain of the plant [58]. (f) An example of the SE-YOLOv5x algorithm detecting the seedling stem emergence point; the red dot indicates the recognition result [59]. (g) An example of the CNN-based PSEP localization algorithm for locating the seedling stem emergence point; the red dot indicates the recognition result [60]. (h) An example of the improved centroid detection algorithm identifying the corn canopy; the red box indicates the recognition result [53].

Table 3. Summary of potential weed apical growth zone identification methods.

Source	Country	Target of Detection	Sample Size	Methodologies	Performance
Zhang et al. (2024) [50]	China	Weed Top Growth Zone	6978	F-YOLOv8n-seg-CDA	Overall detection accuracy: 81%, Average inference speed: 82.9 FPS.
Midtiby et al. [52]	Denmark	Lettuce Stem Emergence Point	805 leaves/223 data points	Beet PSEP estimation system based on leaf detection	95% of cases had error less than 3.58 mm.
Zhao et al. [57]	China	Lettuce Center	584	LettWd-YOLOv8l	Average localization accuracy: 86.05%.
Zhang et al. [59]	China	Lettuce Stem Emergence Point	1918	SE-YOLOv5x	Accuracy: 97.14%.
Karimi et al. (2018) [60]	Iran	Cereal Plant Emergence Point	5719	CNN-based PSEP localization system	F1 score: 82% (Field 1), 86% (Field 2), 74% (Field 3) under varying PSEP densities.
Wu et al. (2022) [51]	China	Cereal Stem Center (Seedling Stage)	100	HSV and Zhang-Sue skeleton extraction method	Achieved precise recognition and localization of seedling stage crop stems, with positioning error less than 12 mm.
Jin and Tang (2009) [61]	USA	Corn Plant Center	122	Skeleton extraction algorithm	Recognition accuracy: 96.7%, Maximum positioning error: 5 cm (74.6%) and 1 cm (62.3%).
Hu et al. (2013) [53]	China	Lettuce Plant Center	80	2G-R-B combined with connected domain	Correct recognition rate for cotton seedlings: 95.8%, for lettuce seedlings: 100%.
Zong et al. (2019) [58]	China	Corn Canopy at Seedling Stage	8000	Improved centroid detection algorithm	Average recognition rate: 92%, Average processing time per frame: 17 s, Centroid positioning error: ≤1 pixel.
Wei et al. (2017) [62]	China	Corn Plant Center	108	Super Green Enhancement—Otsu Threshold Segmentation—Horizontal Set Localisation—Partitioned Search for Accurate Recognition of Maize Plant Hearts	Recognition accuracy: 96%.

Table 3. Summary of potential weed apical growth zone identification methods.

Source	Country	Target of Detection	Sample Size	Methodologies	Performance
Zhang et al. (2024) [50]	China	Weed Top Growth Zone	6978	F-YOLOv8n-seg-CDA	Overall detection accuracy: 81%, Average inference speed: 82.9 FPS.
Midtiby et al. [52]	Denmark	Lettuce Stem Emergence Point	805 leaves/223 data points	Beet PSEP estimation system based on leaf detection	95% of cases had error less than 3.58 mm.
Zhao et al. [57]	China	Lettuce Center	584	LettWd-YOLOv8l	Average localization accuracy: 86.05%.
Zhang et al. [59]	China	Lettuce Stem Emergence Point	1918	SE-YOLOv5x	Accuracy: 97.14%.
Karimi et al. (2018) [60]	Iran	Cereal Plant Emergence Point	5719	CNN-based PSEP localization system	F1 score: 82% (Field 1), 86% (Field 2), 74% (Field 3) under varying PSEP densities.
Wu et al. (2022) [51]	China	Cereal Stem Center (Seedling Stage)	100	HSV and Zhang-Sue skeleton extraction method	Achieved precise recognition and localization of seedling stage crop stems, with positioning error less than 12 mm.
Jin and Tang (2009) [61]	USA	Corn Plant Center	122	Skeleton extraction algorithm	Recognition accuracy: 96.7%, Maximum positioning error: 5 cm (74.6%) and 1 cm (62.3%).
Hu et al. (2013) [53]	China	Lettuce Plant Center	80	2G-R-B combined with connected domain	Correct recognition rate for cotton seedlings: 95.8%, for lettuce seedlings: 100%.
Zong et al. (2019) [58]	China	Corn Canopy at Seedling Stage	8000	Improved centroid detection algorithm	Average recognition rate: 92%, Average processing time per frame: 17 s, Centroid positioning error: ≤1 pixel.
Wei et al. (2017) [62]	China	Corn Plant Center	108	Super Green Enhancement—Otsu Threshold Segmentation—Horizontal Set Localisation—Partitioned Search for Accurate Recognition of Maize Plant Hearts	Recognition accuracy: 96%.

3.2. Laser Beam Control Technology

Accurate control of the laser beam is the key to achieving precise laser irradiation on the top growing area of the weed. This technique can be achieved by integrating the laser into the robotic arm structure and calculating the motor displacement angle through inverse kinematics for targeting in response to changes in weed position [54]. This method can achieve an operating speed of 0.2 m/s under flat ground conditions with a weed removal rate of 85% and an average seedling injury rate of 4.68%. However, the robotic arm system is limited by the structural response time and motor control precision, and its dynamic adaptability in complex terrain is insufficient to guarantee high-frequency, stable, and accurate irradiation. The laser irradiation of a single weed can be completed within 0.64 s with a hit rate of 97% and an average positioning error of 1.97 mm by the dual-head control structure [55]. Although the system is superior in terms of control accuracy, its operating speed is only 0.03 m/s, which makes it difficult to meet the demand for efficiency in large-area field operations. A laser weeding platform based on the 2-DoF 5R RPM framework achieved an average positioning accuracy of 0.62 mm and a dynamic weeding efficiency of 0.72 s/plant, which balanced the performance of laboratory and field scenarios with high structural complexity [63]. However, the complexity and cost investment of this system poses an obstacle to its practical dissemination. By contrast, an autonomous laser weeding robot with a vibroseis system has the best overall performance [49], with an operating speed of 0.5 m/s and a weeding rate of 92.6%, while maintaining a low crop damage rate (1.2%). The relevant technical scheme is shown in Figure 8.

In complex farmland environments, vibroseis technology is undoubtedly a promising solution for precise control of laser beams in laser weeding systems, taking into account factors such as weeding rate, real-time performance, and cost. However, despite the high accuracy and fast response characteristics of galvanometer technology [65,66], there are still adaptability problems when applying it to the variable environment of farmland. Ground undulations may cause deviations in laser aiming and precise irradiation of the target, thus affecting the weeding effect. As shown in Figure 9, the laser beam is reflected by mirrors in the X- and Y-axis and then focused by the F-Theta lens to the working surface. When the height changes, if the weeds are located directly below the laser, the height change has less impact on the irradiation coordinates; however, if the weeds are located diagonally below the laser, the irradiation coordinates will change as the height changes. For example, when the height changes from h0 to h1, the coordinates of the weed change from P(x, y) to EP(x′, y′). Future research could focus on developing vibroseis technology that is more suitable for complex environments, constructing real-time feedback and adaptive control models, and effectively compensating for the interference of terrain undulation on laser positioning, so as to enhance the ability of high-precision weed control in dynamic environments.

3.3. Laser Modulation Related Research

Further development of laser weed control technology relies on in-depth studies of laser modulation mechanisms (Table 4). Studies have shown that the absorption properties of plant tissues for laser radiation are influenced by the laser wavelength, and Figure 10a illustrates the absorption spectra of fresh and dried leaves, indicating that only lasers capable of generating high absorption in plant tissues can be used effectively for weed control [67]. The response of different weed species to laser wavelengths varied significantly. For example, Stellaria media, Tripleurospermum inodorum, and Brassica napus generally absorbed the 532 nm laser better than the 810 nm laser, with Stellaria media having the strongest absorption effect and Brassica napus the weakest [20]. This result shows that different weeds have different sensitivity to laser wavelengths, so the most suitable laser wavelength can be selected according to the weed species to achieve more accurate and efficient control.

The efficiency of laser weed control also depends on the growth stage of the weed. As shown in Figure 10b [68], the energy density required increases significantly as the weed growth stage advances. Annual ryegrass required an energy density of only 19.1 J/mm² to reduce biomass by 93.3% at the three-leaf stage, while at the same energy density at the seven-leaf stage, biomass was reduced by only 23.3%. At the middle and late growth stages, even when the highest energy density (76.4 J/mm²) was applied, the biomass reductions were only 60.2% and 10.4%. Further studies found a log-logical relationship between laser energy density and plant biomass reduction, suggesting the existence of a critical energy density threshold, which, once exceeded, would result in a sharp decrease in plant biomass [68]. Current laser weeding robots generally adopt a uniform high-energy output strategy when performing weeding tasks, treating weeds with the highest laser power regardless of their species or growth stage, resulting in a large waste of energy. It is suggested that further optimization of laser weeding systems should focus on the hierarchical regulation of laser energy. Specifically, for weeds in the early growth stage, only lower energy is needed to achieve the desired killing effect, thus effectively reducing energy consumption and improving operational efficiency; while for weeds in the middle and late growth stages, high-precision identification technology should be relied on to determine whether they are within the threshold of the laser lethal energy, to ensure the precise application of energy, avoiding unnecessary waste of resources and ineffective irradiation. This on-demand energy management strategy will help promote the development of laser weed control technology in a more efficient and intelligent direction.

In addition, spot size and spot position equally play a key role in laser weed control effectiveness (Figure 10c [24]). Spot size directly affects the efficiency of laser energy distribution in the target area. It has been shown that a medium-diameter spot (∅4.2 mm) achieves a more uniform heat transfer on the target surface, thus heating and destroying plant tissues more efficiently under the same energy conditions. At the same time, the location of the spot determines the lethality of the weed control. When the laser can precisely cover the target area (e.g., 100% coverage at p1), it causes the most significant lethal damage; on the contrary, if the positioning deviation is large and the coverage is only 49% (p2) or 8% (p3), the laser energy needs to be significantly increased to achieve a similar killing effect [24]. Thus, in the design and application of a laser weeding system, it is necessary to pay attention to the reasonable setting of spot size and the improvement of positioning accuracy at the same time and combine the real-time sensing and feedback mechanism to build an adaptive control strategy, to optimize the energy consumption while guaranteeing the weeding efficiency.

Figure 10. Relevant factors affecting laser weed control: (a) Absorption spectra of fresh and dried leaves (A. retroflexus) and the absorption coefficients of water, the red dashed line indicates the case of water absorption spectra [67]. (b) Influence of increasing laser energy density doses on the biomass of annual ryegrass plants when applied at four plant growth stages expressed as the percentage of the untreated control. Growth stages treated included three-leaf, seven-leaf, mid-tillering, and late-tillering [68]. (c) Predicted probabilities of lethal damages dependent on weed species, laser spot position (p1: 100%, p2: 49%, p3: 8%), growth size (w1: BBCH 10, w2: BBCH 12, w3: BBCH 14), laser spot area (s1: 7 mm², s2: 14 mm², s3: 28 mm²) [24].

Figure 10. Relevant factors affecting laser weed control: (a) Absorption spectra of fresh and dried leaves (A. retroflexus) and the absorption coefficients of water, the red dashed line indicates the case of water absorption spectra [67]. (b) Influence of increasing laser energy density doses on the biomass of annual ryegrass plants when applied at four plant growth stages expressed as the percentage of the untreated control. Growth stages treated included three-leaf, seven-leaf, mid-tillering, and late-tillering [68]. (c) Predicted probabilities of lethal damages dependent on weed species, laser spot position (p1: 100%, p2: 49%, p3: 8%), growth size (w1: BBCH 10, w2: BBCH 12, w3: BBCH 14), laser spot area (s1: 7 mm², s2: 14 mm², s3: 28 mm²) [24].

Table 4. Summary of mechanistic studies related to laser weed control.

Laser Type	Laser Wavelength	Experimental Parameters	Experimental Objects	Growth Stage	Conclusion	Reference
CO2 Laser	10.6 μm	Power: 4/10/20 W Speed: 1/5/10 mm/s Spot Diameter: 0.6 mm (24 cm focal length)	Chenopodium album, Sinapis arvensis, Lolium perenne	Cotyledon, 2-leaf, 4-leaf stages	Chenopodium album: 0.9 J/mm dose reduced 90% biomass. Sinapis arvensis: 2.3 J/mm dose reduced 90% biomass.	Heisel et al. (2002) [21]
Thulium-doped Fiber Laser	2000 nm	Power: 50 W Spot Diameter: 2 mm Angle: 45° Dose: 0–15.9 J/mm2	Elymus repens	1-leaf, 2-leaf, 3-leaf stages	≥1.6 J/mm2 effectively kills 3-leaf plants.	Andreasen et al. (2024) [69]
Fiber-coupled Diode Laser	975 nm	Power: 25 W Spot Diameter: 5 mm Irradiation Time: 1–60 s Energy Density: 1.3–76.4 J/mm2	Lolium rigidum	3-leaf, 7-leaf, late tillering stage	76.4 J/mm2 achieves full control.	Coleman et al. (2021) [68]
CO2 Laser, Diode Laser, Solid-state Laser	10,600 nm, 940 nm, 532 nm	Energy: 125–1400 J/plant Spot Diameter: 2 mm.	Amaranthus retroflexus	Seedling, 2-leaf, 4-leaf stages	CO2 laser has the highest absorption efficiency, with low-power long-duration exposure being more effective.	Marx et al. (2012) [67]
CO2 Laser, Diode Laser, Fiber Laser	10,600 nm, 940 nm, 980 nm, 1908 nm, 532 nm	Energy: 125–1400 J/plant Spot Diameter: 2 mm	Amaranthus retroflexus, Echinochloa crus-galli, Chenopodium album	Cotyledon to mature stages	CO2 laser has the highest efficiency (125 J/plant); 532 nm has the lowest efficiency (1400 J/plant).	Andreasen et al. (2022) [45]
CO2 Laser, Diode Laser	10,600 nm, 940 nm	CO2: 20 W × 1 s (6 mm) Diode: 250 W × 2 s (6 mm)	Echinochloa crus-galli, Nicotiana tabacum	Cotyledon, 2-leaf, 3-leaf stages	CO2 laser absorption rate nearly 100%, significantly more efficient than diode laser.	Wöltjen et al. (2008) [23]
Diode Laser	532 nm, 810 nm	532 nm: 5 W (0.9/1.8 mm) 810 nm: 90 W (1.2/2.4 mm)	Stellaria media, Tripleurospermum inodorum, Brassica napus	cotyledon stage	Stellaria media: 532 nm (0.9 mm) ED90 = 1.4 J. Tripleurospermum inodorum: 532 nm (1.8 mm) ED90 = 2.7 J.	Mathiassen et al. (2006) [20]
He-Ne Laser, CO2 Laser	623 nm, 10,600 nm	CO2: 1–40 W Speed: 1–10 mm/s Spot: 0.6 mm2	Solanum nigrum	Cotyledon, two true-leaf stages	At 250 plants/m2 density, 22.5 kJ/m2 required to achieve 90% kill.	Heisel et al. (2022) [21]
Continuous Fiber Laser	1064 nm	Power: 100 W Speed: 80 mm/s Line Distance: 0.5 mm Spot: 8 mm	Veronica didyma	Seedling stage	Complete carbonization of leaves and growth points.	Qin et al. (2024) [56]
Blue Laser	400–500 nm	Dose: 13–15 J/mm	Echinochloa colonum, Amaranthus retroflexus, Plantago asiatica	2-leaf, 4-leaf stages	Dry weight control > 85%, more effective on grasses.	Zhu et al. (2022) [54]
CO2 Laser	10,600 nm	Energy Density: 0.08–5.00 J/mm2 Spot Diameter: Multiple sizes Irradiation Time: 500 ms	Echinochloa crus-galli, Amaranthus retroflexus	2–4 leaf stages	Highest efficiency when spot center covers 100%, dicotyledons more sensitive.	Marx et al. (2012) [24]
Semiconductor Laser	405 nm, 635 nm, 450 nm	Power: 0.3–5 W Irradiation Time: 0.5–7 s	Elytrigia repens	35-day seedlings	5 W suitable for >2 mm stems but with risk of collateral damage; 1 W suitable for <2 mm stems.	Rakhmatulin and Andreasen (2020) [25]
Diode Laser	450 nm	Power: 1.2–6.1 W Distance: 5–15 cm	Palmer amaranth, smallflower morningglory	3-week-old plants	4.2 W × 3 s achieves 100% kill, 6.1 W × 1.5 s achieves full control.	Mwitta et al. (2022) [70]

Table 4. Summary of mechanistic studies related to laser weed control.

Laser Type	Laser Wavelength	Experimental Parameters	Experimental Objects	Growth Stage	Conclusion	Reference
CO2 Laser	10.6 μm	Power: 4/10/20 W Speed: 1/5/10 mm/s Spot Diameter: 0.6 mm (24 cm focal length)	Chenopodium album, Sinapis arvensis, Lolium perenne	Cotyledon, 2-leaf, 4-leaf stages	Chenopodium album: 0.9 J/mm dose reduced 90% biomass. Sinapis arvensis: 2.3 J/mm dose reduced 90% biomass.	Heisel et al. (2002) [21]
Thulium-doped Fiber Laser	2000 nm	Power: 50 W Spot Diameter: 2 mm Angle: 45° Dose: 0–15.9 J/mm2	Elymus repens	1-leaf, 2-leaf, 3-leaf stages	≥1.6 J/mm2 effectively kills 3-leaf plants.	Andreasen et al. (2024) [69]
Fiber-coupled Diode Laser	975 nm	Power: 25 W Spot Diameter: 5 mm Irradiation Time: 1–60 s Energy Density: 1.3–76.4 J/mm2	Lolium rigidum	3-leaf, 7-leaf, late tillering stage	76.4 J/mm2 achieves full control.	Coleman et al. (2021) [68]
CO2 Laser, Diode Laser, Solid-state Laser	10,600 nm, 940 nm, 532 nm	Energy: 125–1400 J/plant Spot Diameter: 2 mm.	Amaranthus retroflexus	Seedling, 2-leaf, 4-leaf stages	CO2 laser has the highest absorption efficiency, with low-power long-duration exposure being more effective.	Marx et al. (2012) [67]
CO2 Laser, Diode Laser, Fiber Laser	10,600 nm, 940 nm, 980 nm, 1908 nm, 532 nm	Energy: 125–1400 J/plant Spot Diameter: 2 mm	Amaranthus retroflexus, Echinochloa crus-galli, Chenopodium album	Cotyledon to mature stages	CO2 laser has the highest efficiency (125 J/plant); 532 nm has the lowest efficiency (1400 J/plant).	Andreasen et al. (2022) [45]
CO2 Laser, Diode Laser	10,600 nm, 940 nm	CO2: 20 W × 1 s (6 mm) Diode: 250 W × 2 s (6 mm)	Echinochloa crus-galli, Nicotiana tabacum	Cotyledon, 2-leaf, 3-leaf stages	CO2 laser absorption rate nearly 100%, significantly more efficient than diode laser.	Wöltjen et al. (2008) [23]
Diode Laser	532 nm, 810 nm	532 nm: 5 W (0.9/1.8 mm) 810 nm: 90 W (1.2/2.4 mm)	Stellaria media, Tripleurospermum inodorum, Brassica napus	cotyledon stage	Stellaria media: 532 nm (0.9 mm) ED90 = 1.4 J. Tripleurospermum inodorum: 532 nm (1.8 mm) ED90 = 2.7 J.	Mathiassen et al. (2006) [20]
He-Ne Laser, CO2 Laser	623 nm, 10,600 nm	CO2: 1–40 W Speed: 1–10 mm/s Spot: 0.6 mm2	Solanum nigrum	Cotyledon, two true-leaf stages	At 250 plants/m2 density, 22.5 kJ/m2 required to achieve 90% kill.	Heisel et al. (2022) [21]
Continuous Fiber Laser	1064 nm	Power: 100 W Speed: 80 mm/s Line Distance: 0.5 mm Spot: 8 mm	Veronica didyma	Seedling stage	Complete carbonization of leaves and growth points.	Qin et al. (2024) [56]
Blue Laser	400–500 nm	Dose: 13–15 J/mm	Echinochloa colonum, Amaranthus retroflexus, Plantago asiatica	2-leaf, 4-leaf stages	Dry weight control > 85%, more effective on grasses.	Zhu et al. (2022) [54]
CO2 Laser	10,600 nm	Energy Density: 0.08–5.00 J/mm2 Spot Diameter: Multiple sizes Irradiation Time: 500 ms	Echinochloa crus-galli, Amaranthus retroflexus	2–4 leaf stages	Highest efficiency when spot center covers 100%, dicotyledons more sensitive.	Marx et al. (2012) [24]
Semiconductor Laser	405 nm, 635 nm, 450 nm	Power: 0.3–5 W Irradiation Time: 0.5–7 s	Elytrigia repens	35-day seedlings	5 W suitable for >2 mm stems but with risk of collateral damage; 1 W suitable for <2 mm stems.	Rakhmatulin and Andreasen (2020) [25]
Diode Laser	450 nm	Power: 1.2–6.1 W Distance: 5–15 cm	Palmer amaranth, smallflower morningglory	3-week-old plants	4.2 W × 3 s achieves 100% kill, 6.1 W × 1.5 s achieves full control.	Mwitta et al. (2022) [70]

4. Technological Bottlenecks and Challenges

With the continuous advancement of agricultural modernization, there is a rising demand for efficient, precise, and environmentally friendly weed control technologies. Laser weed control technology has become a research hotspot in the field of agriculture due to its advantages of precise targeting, low damage to crops, and reduced use of chemical herbicides, which can not only effectively remove weeds but reduce environmental pollution. However, despite the preliminary results of theoretical and experimental research, laser weed control technology still faces many technical bottlenecks and challenges in practical application, such as target identification, dynamic positioning, and energy regulation, which limits its large-scale promotion and practical application.

The precision of laser weed control relies on efficient target recognition techniques. Differences between weeds and crops in terms of morphology, color, and texture form the basis of recognition [71], but variable factors in complex farmland environments pose serious challenges to target recognition. For example, factors such as lighting conditions, weather variations, and growth stages can significantly affect image contrast and sharpness, which can easily lead to misidentification and localization bias under unfavorable conditions, such as strong light or shadows [44,72]. In addition, different varieties of crops and weeds may present similar visual features, which are particularly difficult to distinguish under complex backgrounds, further increasing the difficulty of accurate positioning of the top growing zone of weeds [73,74]. Although recognition algorithms based on machine vision and deep learning have been applied in laser weeding robots [75], their robustness and accuracy in real-life complex environments still need to be improved. Therefore, improving the adaptability of the algorithms in variable farmland environments has become a key technical challenge to be solved.

Precise dynamic positioning is one of the key challenges to be solved in laser weed control technology. To achieve the precise irradiation of a specific part of the weed (apical growth zone), the laser beam must be positioned with high accuracy during the dynamic operation. However, uneven terrain, crop occlusion, and errors introduced by robot motion in farmland environments may cause the laser beam to deviate from the target position [76,77]. In addition, non-uniformity in crop growth density and height can interfere with laser localization, causing it to misfire on non-target areas (e.g., the crop itself), which in turn affects weed control [78]. Especially in crop growing patterns with narrow row spacing, laser weeding equipment requires higher positioning accuracy and flexibility to prevent accidental damage to neighboring crops. Therefore, improving the accuracy of dynamic positioning, effectively reducing environmental interference, and enhancing the overall flexibility of the system have become important challenges that need to be overcome in the current application of laser weeding technology.

The effect of laser weed control is highly dependent on the precise regulation of laser parameters. There are obvious differences in the laser energy requirements of different weeds at different growth stages, so how to flexibly adjust the laser power and irradiation time according to the growth of weeds is the key to achieving precise removal [24]. For example, higher laser power and longer irradiation time are usually required for weeds with higher growth or thicker stems to fully destroy their cellular structure and achieve the desired weed control effect [68]. However, the increase in power is often accompanied by problems such as increased energy consumption and difficulties in heat dissipation of the equipment, which not only affects the economy of the system but may trigger overheating, leading to a decrease in the stability of the equipment. In addition, the quality of the laser beam is also crucial to the accuracy of weeding; ideally, the laser beam should have uniform light intensity distribution and stable beam quality but, in practice, due to the laser manufacturing process, the quality of the optical components, and the external environmental fluctuations and other factors, the laser beam may be dispersed and the light intensity uneven, which can lead to the reduction of positioning accuracy and affect the final weeding effect. In summary, under the premise of ensuring efficient weed control, how to optimize the laser energy output while reducing energy consumption and improving the quality of the beam is still a technical challenge that needs to be solved urgently.

Although laser weeding technology has demonstrated significant advantages, such as precision, efficiency, environmental protection, and non-pollution, its high equipment cost is still the main obstacle restricting its wide application. The core components of laser weeding systems include, among other things, lasers, optical systems, control systems, and robotic platforms, and the requirements for high quality and the precision of these components determine the high development and manufacturing costs [42]. For example, high-power lasers and precision optics are expensive [79], while advanced sensors and complex control systems are required to enhance target recognition and positioning accuracy, which further pushes up the overall equipment cost [80]. Such high inputs make the promotion of laser weed control technology in agricultural production face significant economic pressure, especially in some developing countries and regions, where farmers often find it difficult to afford such huge investments. Therefore, how to reduce the production cost of equipment and improving cost-effectiveness has become the key to realizing the large-scale application of laser weed control technology.

Another key challenge facing laser weed control technology is energy efficiency. Currently, some laser weed control equipment uses high-power lasers, but due to the low energy conversion efficiency, a large amount of laser energy is wasted during transmission and irradiation, which directly affects weed control efficiency and increases energy consumption [81]. In addition, laser energy is often interfered with by factors such as atmospheric scattering and weed resistance in actual field environments, and the high energy efficiency parameters verified in the laboratory may be significantly attenuated in practical applications, resulting in lower-than-expected weed control effects. Therefore, it is particularly important to improve the overall energy efficiency of laser weed control systems and to develop intelligent regulation systems that are adaptive to the environment.

5. Conclusions and Outlook

With the acceleration of agricultural modernization, there is an increasing demand for efficient, precise, and environmentally friendly weed control technologies. As an emerging non-chemical precision weed control method, laser weeding has become an important research direction in the field of agricultural intelligent equipment due to its strong selectivity, environmental friendliness, and wide adaptability. This paper presents a systematic review of the key technologies in laser weeding, encompassing target recognition, dynamic positioning, and precise removal, among other aspects.

In terms of target recognition, although the existing algorithms have made some breakthroughs in recognition accuracy and computational efficiency, it is still difficult to fully meet the special requirements of laser weeding for high-precision recognition. For example, it is still a big challenge to accurately identify the top growth area of weeds and distinguish dicotyledonous weeds from monocotyledonous weeds. Currently, some methods, such as stem emergence point localization and plant skeleton extraction, provide new research directions for target identification, but their practicality and robustness have yet to be verified. Future research should focus on developing more efficient, accurate, and adaptable recognition algorithms, such as combining multimodal sensing information (e.g., RGB, depth image, thermal imaging, etc.) for fusion sensing, which is also an important direction to improve the accuracy of target recognition.

In terms of dynamic positioning, the current research compares a variety of laser beam control technologies, including robotic arms, gimbal structures, and vibroseis technology. Among them, the vibrating mirror technology is considered to be the more promising solution due to its combined advantages in terms of weeding efficiency, response speed, and cost control. However, this technology still faces the problem of decreasing positioning accuracy when dealing with complex terrains in farmland, such as undulating ground. To improve its adaptability, future research should be devoted to constructing a laser beam regulation model that can cope with irregular terrain and developing a more versatile and flexible galvanometer control system. At the same time, through the introduction of high-efficiency batteries, solar energy, and other green energy to optimize the system power supply, and supplemented by the optimization of equipment design and manufacturing cost control, the overall system performance and economy can be effectively improved, and to further promote the scale application of laser weeding technology in agricultural production.

In terms of precision removal, based on the in-depth research on the mechanism of laser weed control, three core rules for the optimization of laser parameters have been clarified: firstly, the selection of laser type should seek the best balance between energy efficiency and economy; secondly, the energy thresholds corresponding to different weed species and their growth stages vary significantly; and lastly, the diameter of the laser spot and the location of the laser spot irradiation play an important role in the spatial distribution of the energy in the plant tissues. The results of this study are summarized as follows. To further improve the performance of laser weed control systems, future research should focus on the deeper mechanism of the interaction between laser and plant tissue, establish mathematical models to dynamically adjust the laser parameters, and develop intelligent laser systems with the ability of environmental self-adaptation. At the same time, strengthening research on the energy response to multiple weed species and improving the operational stability of the equipment in complex terrain, climatic conditions, and diversified planting modes will provide solid technical support for the practical application and large-scale promotion of this technology.

Overall, laser weeding technology is in a rapid development stage, but still faces multiple challenges in terms of target recognition accuracy, dynamic positioning stability, energy efficiency improvement, and equipment cost control. Future research should focus on breaking through these technological bottlenecks to promote the practicality and widespread diffusion of laser weeding systems. Through multidisciplinary collaborative innovation, combined with deep learning, precision control technology, and sustainable energy systems, laser weeding is expected to become an efficient, environmentally friendly, and economical weeding solution in future agricultural production.