New Developments in Smart Farming Applied in Sustainable Agriculture

1. Introduction

Sustainable agriculture aims to increase agricultural productivity while minimising negative environmental impacts. One element that supports this approach to food production is precision agriculture. Contemporary research is being conducted in many directions to support sustainable agriculture [1,2,3]. Research concerns the development of innovative technologies for the precise management of crops and livestock, reducing the use of energy as well as natural resources such as water, fertilisers and plant protection products. Agricultural engineering research work is supported by disciplines such as computer science, automation and robotics, or artificial intelligence. The achievements of engineering science are increasingly being implemented in modern agriculture [4,5,6,7].

Sustainable agriculture is a response not only to the deleterious impact of agricultural production on water, soil, and air quality [8,9,10,11,12], but also to the reduction in biodiversity. The introduction of large-scale farming with monocultures has been demonstrated to reduce the diversity of plants found in these areas, and is also a threat to animals. One example of this is pollinating insects, which have difficulty accessing food for most of the year due to monocultures. Within crop species, there is also a decline in diversity, because farmers prefer a limited number of varieties that are easy to grow, disease-resistant, and economically viable. It is therefore very important to take care of agrobiodiversity, meaning diversity in the living organisms in agricultural systems (Contribution 1).

A pivotal component of sustainable agriculture pertains to the precise optimisation of agro-technical treatments. Thanks to precision farming technology, it is possible to adjust the amount of fertiliser to the actual needs of the plants, thus reducing production costs and the negative impact on the ecosystem. This makes use of soil abundance and yield maps, as well as machines that implement variable fertiliser application. Performing treatments such as variable-depth tillage allows significant fuel savings while maintaining adequate soil quality for plant growth [13]. A significant element of crop production, namely plant protection, can be effectively supported by machines equipped with vision systems and artificial intelligence algorithms. These systems facilitate the identification of plant diseases, pest infestations, and the presence of weeds, thereby enabling the mechanical removal of weeds or the selective spraying of weeds. This approach facilitates an early response and precise crop management. The integration of robotics, vision systems, and machine learning (ML) methods also enables the development of autonomous harvesters. These machines have the capacity to harvest not only cereals, but also pressure-sensitive fruits such as strawberries [14]. The employment of drones facilitates the precise monitoring of agricultural fields and the application of crop protection products and fertilisers, in accordance with the specific requirements of the vegetation.

The utilisation of artificial intelligence algorithms, encompassing ML, constitutes a pivotal element in the analysis of substantial agricultural data sets. The employment of predictive models facilitates the forecasting of meteorological conditions, the optimisation of sowing and harvesting dates, and the mitigation of losses incurred due to drought or crop diseases. These algorithms are capable of processing data generated by diverse sensor types, including those affixed to agricultural machinery, in addition to camera data and spectral data across a range of spectral domains [15].

Sustainable agriculture is an important line of research, as it allows increasing agricultural productivity while reducing the consumption of natural resources. This is of great importance given the negative environmental impact of crop and livestock production, and the need to feed the world’s growing human population with resources reduced due to climate change. Therefore, the development of precision agriculture and AI technologies can contribute to the further development of sustainable agricultural practices.

The objective of this Special Issue was to publish high-quality research articles that address issues pertaining to state-of-the-art solutions for agriculture that support precision farming techniques for sustainable agriculture.

2. An Overview of Published Articles

A significant trend in scientific research concerning precision farming techniques for sustainable agriculture is the development of methodologies for determining the optimal parameters for agricultural operations. The objective of this is to minimise energy consumption and negative environmental impact. For instance, in Contribution 2, the researchers examined the impact of moisture content (42%, 56%, and 69%), forage length (183 mm (chopped) and 312 mm (long)), and baling pressure (0.9 MPa, 1.4 MPa, and 1.8 MPa) on the density of haylage bales. This study investigated the implications of these factors on fermentation, storage, and transport performance. Furthermore, a series of experiments were conducted to evaluate the impact of a silage additive on the compaction of bales. Drawing upon the findings of these experiments, the authors reported the optimal conditions for the baling of haylage, with the objective of maximising density and storage efficiency. These conditions entailed a moisture content of 56%, forage lengths of 183 mm (chopped), and a pressing pressure of 1.8 MPa, with the intention of facilitating fermentation, reducing spoilage, and enhancing transport efficiency. In the contemporary context of agricultural practice, achieving optimal liquid distribution and uniformity during the process of spraying is a primary objective. Contribution 3 presents a model based on optimisation results, providing a practical framework for the characterisation of the distribution of liquid sprayed with single-stream nozzles. The model facilitates the control of nozzle operation and the assessment of wear, thereby ensuring uniform application. Experiments were conducted using anti-drift, air-induction, and standard nozzles. The model demonstrated strong agreement with experimental data (R² > 0.95), thus confirming its applicability in improving spraying uniformity.

Another important trend in precision agriculture is the search for alternative, rapid methods for assessing soil and plant physico-chemical and biological parameters. These methods often combine measurement techniques with artificial intelligence methods, and allow accurate determination of selected parameters faster and cheaper than traditional laboratory methods. Amaral et al. (Contribution 4) investigated the potential of hyperspectral remote sensing to estimate nutrient levels in cowpea (Vigna unguiculata) leaves, focusing on the content of phosphorus, potassium, calcium, and zinc across different phenological stages (V4, R6, and R9). Leaf reflectance spectra in the range 350–2500 nm were measured using a spectroradiometer. The study utilised three modelling methods—single-band models, band ratio models, and Partial Least Squares Regression (PLSR)—to identify relationships between spectral data and nutrient levels in leaves. The findings revealed that phosphorus, potassium, and zinc levels diminished during the crop development phase, while calcium levels remained stable or increased, presumably due to its constrained mobility within plant tissues. The study concluded that the accuracy of the models was contingent on the phenological stage. Generally, PLSR produced the most accurate models, with high quality for estimation of phosphorus, potassium, and calcium. However, zinc prediction remains challenging. The vision-based method of red and yellow sweet pepper maturity classification was proposed by van Essen et al. (Contribution 5). The rule-based method for dynamically selecting viewpoints aims to improve classification accuracy while minimising economic costs. The research focuses on integrating next-best-view planning, where an additional viewpoint is captured only if it is profitable, based on misclassification risks. Sweet peppers were classified into three maturity stages based on RGB-D camera imaging. The random forest classifier was trained on colour features extracted from principal components analysis, and the economic model was utilised to determine whether acquiring an additional viewpoint was cost-effective. The dynamic viewpoint selection approach enhanced the classification accuracy by 6% for red peppers and 5% for yellow peppers, in comparison with the use of a single viewpoint. The economic costs were reduced by 52% (red) and 12% (yellow), thereby demonstrating the efficiency of selective viewpoint acquisition. The insights provided by this research are of significant value for the field of agricultural robotics, particularly in the context of optimising machine vision techniques for the automated harvesting of sweet peppers.

A pivotal soil parameter that influences plant growth and yield is soil compaction, the measurement of which by conventional methods is extremely labour-intensive. An alternative approach involves the utilisation of geophysical data to predict soil mechanical parameters, with geophysical soil features being measured using scanners such as the Geonics EM38 conductivity meter. Machine learning methods can be employed to develop models for predicting soil mechanical parameters. In the research conducted by Pentoś et al. (Contribution 6), the relationship between soil electrical properties (electrical conductivity and magnetic susceptibility) and soil compaction and shear stress was investigated. Multiple linear regression and ML approaches were applied to predict these soil properties. The findings indicate that neural network models, particularly the multilayer perceptron and radial basis function networks, can produce models of high accuracy, achieving a correlation coefficient (R) of 0.846 for soil compaction and 0.680 for shear stress. These findings have practical applications in precision agriculture, allowing for improved soil management and reduced energy losses in agricultural machinery operations. Another significant parameter influencing plant growth and yield is soil moisture. The study by Alibabaei et al. (Contribution 7) evaluates the effectiveness of a deep learning approach, specifically a Bidirectional Long Short-Term Memory (BiLSTM) model, in predicting the Fraction of Transpirable Soil Water (FTSW) in vineyards. The FTSW was estimated based on different input parameters, including relative humidity, reference evapotranspiration, rainfall, and vapour pressure deficit. The authors developed models of high accuracy for an independent test data set (R² of 87% and an RMSE of 10.36%). The results suggest that BiLSTM models are a reliable tool for predicting FTSW in vineyards, which can support irrigation management decisions and optimise water use efficiency. The study highlights the potential of deep learning in precision agriculture and the need for further research to adapt these models to different environmental conditions.

Sustainable agriculture activities closely related to precision farming include robotic solutions and autonomous harvesters, supported by machine learning-based vision systems. The Nano Aerial Vehicle employed for pollination purposes (Nano Aerial Bee) (Contribution 8) is a particularly intriguing case study. The researchers propose a bio-inspired approach, involving the design of miniature drones capable of replicating natural pollination processes. The article presents the hardware implementation and a vision system based on YOLO neural networks trained for flower detection. The study concludes that while NAVs present a viable solution to pollination challenges, further research is necessary to enhance their efficiency, scalability, and environmental integration. The findings emphasise the potential for robotics to contribute to the sustainability of agricultural productivity in the context of pollinator decline. Another example of the use of Unmanned Aerial Vehicles in agriculture is drones for precision spraying of crop protection products. Such drones can detect weeds in crops and carry out precision spraying, minimising the negative impact on the environment and reducing the problem of herbicide-resistant weeds. An important component of such drones are vision systems that allow accurate weed classification. Deep neural networks are employed for this purpose, but their optimal training remains challenging. Fathimathul et al. (Contribution 9) put forward the proposition of utilising the Harris Hawks Optimisation algorithm as a sophisticated approach for weed detection among crops by employing drone-captured images and Convolutional Neural Networks (CNNs). The experimental outcomes demonstrate that the proposed method enhances classification accuracy, computational efficiency, and generalisation capability. The findings suggest that integrating metaheuristic algorithms with CNNs can significantly enhance precision agriculture by enabling automated weed detection, leading to reduced herbicide use and improved crop yields.

Sustainable agriculture employs a range of technologies to simultaneously optimise resource use, environmental impact and production volumes. In a paper by Santos et al. (Contribution 10) the authors demonstrate that optimised Heating, Ventilation, and Air Conditioning (HVAC) system designs can significantly enhance sustainability, reduce water waste, and improve overall energy efficiency in controlled-environment agriculture. The findings highlight the potential for sustainable mushroom farming to contribute to climate change mitigation and more eco-friendly food production systems. Tombe et al. (Contribution 11) propose a digitalization framework for agricultural value chains with a view to enhancing inclusivity and economic growth in agricultural communities. The research highlights the role of agricultural social networks (ASNs) in knowledge sharing, market access, and financial inclusion for small-scale farmers. Machine learning algorithms are very helpful in the development of precision agriculture. The quality of the solutions obtained using ML depends very much on the quantity and quality of the data utilised during the training process. The advent of the Internet of Things (IoT), satellite imagery, and farm management systems has greatly facilitated data access. However, the heterogeneity of these data poses a significant challenge. In addressing this challenge, Žuraulis et al. (Contribution 12) developed a data aggregation and conversion model specifically tailored for smart farming, with the aim of enhancing the processing of heterogeneous agricultural data from multiple sources. Another issue that has been identified in the context of acquiring data for ML models is the necessity of manual labelling. For large-scale time-series data, this process is both labour-intensive and prone to errors. Jung et al. (Contribution 13) proposed a novel approach to diagnosing device status in smart agriculture. This approach utilises pseudo-labelling and time-series data analysis, achieving an accuracy of 89% in agricultural environments and a 30% reduction in data processing time.

The use of technologies to support sustainable agriculture often requires additional costs. In order for farmers to be willing to incur these costs, they need to be confident that environmentally beneficial measures will also be economically beneficial. Karydas et al. (Contribution 14) proposed an application, namely ProFit, which provides a simple way to assess the profitability of precision agriculture applications. This type of application, equipped with a user-facilitated interface, has the potential to stimulate interest among farmers in the adoption of precision agriculture methodologies.

3. Conclusions

Agriculture is an industry with a significant detrimental effect on the environment. Consequently, the development of precision farming technology, which supports sustainable agriculture, is of paramount importance. Precision farming utilises the latest automation, robotics, and IT solutions to facilitate food production while minimising resource utilisation and environmental contamination. Machine learning algorithms play a pivotal role in this regard, as they facilitate the creation of precise models to optimise the parameters of agro-technical operations. Furthermore, computer vision systems supported by ML have proven to be highly beneficial in practical applications.