Artificial Intelligence and Machine Learning Approaches for Plant Disease Detection: A Comprehensive Review of Image, Sensor, IoT and VOC-Based Techniques

Abstract

Plant diseases are a major constraint to agricultural productivity, crop quality and food security. Conventional diagnosis is commonly based on visual inspection by farmers or experts and, when required, laboratory techniques such as microscopy, culturing and molecular assays. These approaches can be laborious, time-consuming, subjective and difficult to scale for continuous field monitoring. Recent advances in Artificial Intelligence (AI), Machine Learning (ML), Deep Learning (DL), computer vision, Internet of Things (IoT), chemical sensing and electronic-nose technology have created new opportunities for automated, rapid and non-destructive plant disease detection. Classical ML algorithms such as Support Vector Machine (SVM), Random Forest (RF), k-Nearest Neighbour (kNN), Decision Tree and Artificial Neural Network (ANN) have been used with engineered image and sensor features. DL models, including Convolutional Neural Networks (CNNs), ResNet, MobileNet, EfficientNet, YOLO and Vision Transformers (ViTs), can learn discriminative representations directly from images. Recent systematic reviews show rapid growth in this field but also identify important limitations related to dataset diversity, environmental variability, class imbalance, model generalisation and the gap between laboratory and field performance. In parallel, electronic noses and volatile organic compound (VOC) sensors provide a complementary chemical-sensing modality that can detect changes associated with plant stress and disease. This review examines image-based ML/DL, VOC and E-nose approaches, IoT integration, evaluation metrics, current challenges and future research directions. Particular attention is given to multimodal AI that combines plant images, VOC signatures and environmental parameters for early tomato disease detection. The review identifies multimodal sensing, field validation, sensor-drift compensation, explainable AI and lightweight edge deployment as important research opportunities.