Integrating Squeeze-and-Excitation Network with Pretrained CNN Models for Accurate Plant Disease Detection

Abstract

The increasing global population and the challenges posed by climate change have intensified the demand for sustainable food production. Traditional agricultural practices are often insufficient, leading to significant crop losses due to diseases and pests, despite the widespread use of pesticides and other chemical interventions. This paper introduces a new approach that integrates deep learning techniques, specifically Convolutional Neural Networks (CNNs) with Squeeze and Excitation (SE) networks, to enhance the accuracy of disease detection in fig leaves. By leveraging three pre-trained CNN models—MobileNetV2, InceptionV3, and Xception—this framework addresses data scarcity issues and improves feature representation while minimizing the risk of overfitting. Data augmentation techniques were employed to counteract data imbalance, and visualization tools like Grad-CAM and t-SNE were utilized for model interpretability. The proposed CNN-SE model was trained and evaluated on a fig leaf dataset comprising 1,196 images of healthy and diseased fig leaves, achieving an accuracy of 92.90% with MobileNet-SE, 91.48% with Inception-SE, and 89.62% with Xception-SE. Our model demonstrates superior performance in detecting fig leaf diseases, presenting a robust solution for sustainable agriculture by providing accurate, efficient, and scalable disease management in crops. The code of the proposed framework is available at https://github.com/lafta/SE-block-with-CNN-Models-for-Plant-Disease-Detection.