A Lightweight Deep Learning Model for Accurate Plant Disease Detection in Real Applications

Plant diseases pose a significant threat to global agricultural productivity, especially in resource-limited regions where access to expert diagnostics is scarce. This paper presents USENet, a lightweight deep learning model for accurate and efficient plant disease detection using leaf images. USENet leverages MobileNetV3 inverted residual blocks and Spatial Pyramid Pooling (SPP) to minimize model size and memory footprint while maintaining high accuracy. The model is trained and evaluated using the PlantVillage dataset. USENet achieves a precision of 0.9926, recall of 0.9935, and an F1-score of 0.9930, while having only 41,446 parameters. The model is just 0.722 MB in size, making it suitable for edge deployment. The lightweight nature of USENet makes it ideal for deployment on low-cost microprocessors and mobile devices, providing a practical solution for plant disease detection in remote agricultural areas.