Key points are not available for this paper at this time.
Due to rising population, industrialization, environmental losses, and declining interest in low-profit farming, food security faces critical challenges, mostly in developing and underdeveloped countries like India. To address these limitations, it is imperative to promote the adoption of organic farming practices alongside the implementation of timely and judicious disease management strategies, including the precise application of pesticides in appropriate quantities. In a multilingual developing country like India, limited awareness and insufficient knowledge of plant diseases and their prevention remain significant challenges. The increasing demand for sustainable, efficient, and intelligent agricultural practices has driven the rapid advancement and adoption of deep learning (DL), machine learning (ML), internet-of-things (IoT) technologies across the agricultural sector. This review provides a comprehensive examination of recent advances in artificial intelligence (AI), ML, DL, IoT, & computer vision, highlighting their role in transforming agriculture into an automated, efficient, & data-driven system for diverse applications such as plant disease detection, fruit ripeness classification, weed discrimination, and freshness or spoilage assessment. Significant emphasis is given on the development and deployment of convolutional neural network (CNN), transfer learning approaches, vision transformers, and hybrid DL- ML frameworks that leverage high-resolution image/signals acquired through mobile devices, drones, satelites, and field sensors. This review provides a systematic analysis of diverse methodologies applied across multiple crop categories (including cereals, vegetables, and fruits), with a particular focus on evaluating the performance of both custom-designed and pre-trained DL models on datasets obtained from publicly available repositories as well as those collected through field-based sensing technologies, offering a comprehensive comparison of model effectiveness under varying data acquisition conditions. In addition, this literature survey explores the integration of attention mechanisms, feature fusion strategies, image augmentation techniques, and explainable AI (XAI) tools for enhancing model accuracy and interpretability. The analysis extends to emerging applications in post-harvest quality control, with a focus on fruit grading systems and quality assessment of fish and meat using vision and sensor data. While the reviewed literature demonstrates significant advancements in classification accuracy and real-time deployment capabilities, several persistent challenges remain. Key issues include dataset imbalance, limited model generalization across heterogeneous environmental conditions, and insufficient explainability of DL models, all of which constrain their large-scale adoption and practical reliability in diverse agricultural settings. The manuscript discusses various publicly available agricultural datasets for future research along with the performance metric for evaluating the DL/ML models’ performance ability for classification and segmentation. This review concludes by identifying key future research directions, including the advancement of model interpretability to enhance trust and transparency, the integration of multi-modal data fusion for improved decision-making, and the development of robust deployment strategies tailored to low-resource environments. It further emphasizes the need for scalable, crop-agnostic intelligent systems capable of generalizing across diverse agricultural contexts, thereby enabling more effective and sustainable precision agriculture practices.
Nath et al. (Mon,) studied this question.