“Smart Agriculture” Information Technology and Agriculture Cross-Discipline Research and Development

1. Introduction

Digital technology applications in agriculture and biology are a dynamic area of research interest, with topics including, but not limited to, agriculture, data collection, data mining, bioinformatics, genomics and phenomics, as well as applications of machine learning and artificial intelligence.

The development of a community to support this goal requires the cross linking and integration of multiple sources of agricultural research, including 3S technologies (remote sensing—RS, geographic information systems—GIS, global positioning systems—GPS). This broad framework provides a basis for the detection of crop pathogens, weeds and pests (insects) using multi-spectrum techniques and the exploitation of remote sensing technology to create and analyze multiple heterogeneous structured data sets, which subsequently enables effective cross linking and phenomics classification. It is essential to study the growth models of plants/crops and exploit expert support to develop smart production and management decision systems to achieve real-time, quantified and precise decisions.

This SI’s topics of high interest included the capture and curation of biological “big data” research on multi-spectral data analysis, the assembly of complex genetic sequencing fragments, and structural gene predictions coupled with intermediate structures to predict phenotypes. In this context, novel data structures are required to capture predictive structures in the path from genotype to phenotype, concurrently with new techniques to capture and identify the regularity of biological data.

Finally, multiple-sources-based monitoring and decision making for plants, water and nutrients are required, with a research focus on the utilization of remote sensing and drone sensing to compute and predict plant water usage. This framework will lead to the development of precision models of crop water/nutrient management systems and form the foundation for the digitalization of agricultural water/nutrient applications.

The topic is expending rapidly and has a potential high impact. Detection, sensing and decision making in the context of modern agriculture practice is benefiting from smart/digital agriculture research and will contribute to the improvement of sustainable development. Many colleagues have contributed this Special Issue by submitting their research and studies.

2. Results

A total of 31 manuscripts were submitted to this Special Issue, with 20 manuscripts accepted and published. The content of these manuscripts includes artificial intelligence and decision making, sensor and sensing, imaging, and geographic information technology (https://www.mdpi.com/journal/agronomy/special_issues/Smart_Agriculture_Technology (accessed on 26 August 2023)). Using imaging and decision-making technologies, Bi et al. reported a novel method of corn seed identification [1]. Cai et al. determined the varying maturity levels of strawberries by utilizing image segmentation methods from the improved DeepLabV3+ [1]. Crop seed vigor monitoring is revolutionized by optical sensor technology, which enhances performance and ensures consistent production. This technology supports reliable and non-destructive calibration techniques that allow for accurate seed use through agronomic evaluation [2]. A combination of deep learning with machine vision and Swin transformer-based models is capable of achieving a high assortment accuracy through attention to specific characteristics and multi-scale feature fusion networks [3]. To facilitate automated weed management systems, a faster R-CNN network model for weed detection in cropping regions has been introduced, increasing recognition accuracy to over 95%. Additionally, a Swin-DeepLabv3+ model has been utilized for weed recognition in soybean fields, incorporating a Swin transformer and a convolution block attention module. It resolves border contour identification concerns, ameliorates accuracy by 2.94% and achieves an average intersection ratio of 91.53% [4,5]. A deep learning grading approach has been used to identify ginseng, a crucial component of Chinese medicine, with an accuracy of 97.39% and a loss value of 0.035, making it a valuable tool for ginseng appearance quality identification. Using an ATmega 328 microcontroller and an SIM900A GSM modem, an automated system is being developed to mass produce panchagavya, a traditional organic fertilizer used in India. This system will benefit farmers and society [6,7]. A smartphone-based machine vision system is able to accurately simulate the quantities of sand, clay and silt in soil, featured in this Special Issue. This system is using learning models to determine soil texture [8]. In order to remotely monitor atmospheric CO₂, CH₄ and N₂O above rice paddy fields, two high-resolution laser heterodyne radiometers (LHRs) were set up in Hefei. The emissions from rice fields increased the amount of CO₂. Due to the negative correlations between CH₄ and N₂O emissions, rice fields play a part in carbon sequestration during the rice growth season. The LHRs are promising for monitoring emissions from agricultural fields due to their accuracy in observing air concentrations [9]. The network-based automatic algorithm for classifying and detecting rice pests has a recognition accuracy of 98.28%. The model illustrates how deep learning-based classification methods perform better at detection, while insect images are segmented effectively. With a CA attention module and STR detection head, the YOLOv5n model, CTR YOLOv5n, will detect maize disease in mobile applications. It decreases memory to 5.1 MB with an average recognition accuracy of 95.2%. With detection errors of less than 2.0% for both large- and small-diameter soybeans, a photoelectric sensor-based real-time monitoring system assesses the efficiency of soybean seed metering devices. The technique benefits from the system’s assistance for device evaluation and seeding monitoring. An Efficient-Net-B5 network has been employed to describe citrus leaf diseases and foster the sustainability of the citrus industry. The network can handle small, unevenly distributed samples [10,11,12,13] and is able to obtain high accuracy rates, precision, recall and F1 scores. Hyperspectral data are used to determine plant diseases like Septoria tritici and Stagonospora nodorum blotch in cereal crops. A trained neural network can classify between healthy and damaged plants with high accuracy. This technique might help with crop estimation before harvest [14]. The development of ICT has had a huge impact on the world, especially on India’s sugarcane harvest. The goal is to anticipate soil moisture and categorize sugarcane production using two-level ensemble classifiers and an ensemble model that combines support vector machines, convolutional neural networks and the Gaussian probabilistic technique. The suggested approach outperforms existing classifiers by 89.53%, enabling farmers and agricultural authorities to enhance sugarcane farming productivity, thus increasing production. The phenotype monitoring technology uses an internal gradient algorithm to accurately measure the target region and diameter of maize stems. The technology uses color images captured during the small bell stage, extracting color information and applying a morphological gradient algorithm [15,16]. A simple detection system is created for automated yield estimation and picking in small-target apple orchards using the public MinneApple dataset. The program employs 829 photos with difficult weather circumstances and reduces the size by 15.81%, while attaining an mAP of 80.4%. A single-leaf labelling technique and a seedling detection network based on YOLOv5 and a transformer mechanism are presented to identify agricultural seedlings in challenging field situations [17]. The transformer mechanism module increased the detection capability by 1.5%, while the single-leaf labelling approach increased the model’s mAP0.5 by 1.2%. By the 23 ms/frame, the optimized model increased computing speed. In recognizing early fruit body diseases in edible fungus, the study introduces a ShuffleNetV2 + SE model. This model improves disease classification performance, accuracy, precision, recall and Macro-F1 value, making it acceptable for devices with low resources [18,19]. A unique algorithm for void detection in yield maps is presented. The algorithm exhibits 100% sensitivity, 91% specificity and 82% accuracy. The method allows for smooth incorporation into real-time big data quality assessment systems based on different dimensions by mapping geographical mistakes to two common data quality dimensions [20].

3. Future Perspectives

This Special Issue, entitled ““Smart Agriculture” Information Technology and Agriculture Cross-Discipline Research and Development”, has come to a close. We are very grateful for the efforts of the journal editors, peer reviewers and hard-working authors. We also would like to thank all colleagues who contributed to this section; without their efforts, this research topic would not have extended to Agronomy readers. This Special Issue has initiated much attention from research communities towards this topic and, we believe, has been completed during a remarkable time of development. We believe there will be more presentations that could be shared in the future.