There is an increasing demand for efficient and precise plant disease detection methods that can quickly identify disease outbreaks. For this, researchers have developed various machine learning and image processing techniques. However, real-field images present challenges due to complex backgrounds, similarities between different disease symptoms, and the need to detect multiple diseases simultaneously. These obstacles hinder the development of a reliable classification model. The attention mechanisms emerge as a critical factor in enhancing the robustness of classification models by selectively focusing on relevant regions or features within infected regions in an image. This paper provides details about various types of attention mechanisms and explores the utilization of these techniques for the machine learning solutions created by researchers for image segmentation, feature extraction, object detection, and classification for efficient plant disease identification. Experiments are conducted on three models: MobileNetV2, EfficientNetV2, and ShuffleNetV2, to assess the effectiveness of attention modules. For this, Squeeze and Excitation layers, the Convolutional Block Attention Module, and transformer modules have been integrated into these models, and their performance has been evaluated using different metrics. The outcomes show that adding attention modules enhances the original models' functionality.
Agriculture is essential to the economy and social well-being of any community. Early diagnosis of plant diseases is critical to crop protection and sustainable agriculture. Deep learning (DL) has demonstrated outstanding performance in disease detection, but its complexity has made field implementation difficult. To overcome this issue, lightweight DL models have been developed. This paper presents a comparative analysis of lightweight DL models for plant disease classification on three publicly available data sets: PlantDoc, Mango, and Soybean Leaf Disease. The PlantDoc dataset has 28 classes with different crops and disease conditions captured under real field conditions. The Mango data set has images of 7 disease classes and 1 healthy class, and the Soybean data set has 2 disease classes and 1 healthy class, with multiple leaves per frame, depicting real-world scenarios. This study compares lightweight classification models like ShuffleNetV2, MobileNetV2, and MobileNetV3Small and object detection models YOLOv8n and YOLOv11n, modified for classification via bounding box class filtering. The models were evaluated using conventional evaluation metrics such as accuracy, precision, recall, F1 score, and Cohen’s kappa score; additionally, model size and number of parameters are considered in comparison. Experiments demonstrate that MobileNetV2 achieves the highest accuracy of 89.86% on the PlantDoc dataset (augmented), and MobileNetV3Small achieves the highest accuracy on the Mango dataset (99.59% without augmentation) and achieves an accuracy of 96.49% on the Soybean dataset. Ablation studies were conducted by applying pruning and dynamic quantization on models. Models were also deployed on Internet of Things (IoT) devices, such as the Raspberry Pi 5, for performance evaluation. Further, this study states the current limitations present and suggests future directions for efficient deployment of DL models on edge devices for improved crop disease identification.