Cascading Autoencoder With Attention Residual U-Net for Multi-Class Plant Leaf Disease Segmentation and Classification

Abstract

Plant leaf diseases pose a significant threat to global food security and cause substantial economic losses. The objective of this study is to develop an effective approach for early detection and accurate identification of plant leaf diseases using computer vision techniques. The proposed method, Cascading Autoencoder with Attention Residual U-Net (CAAR-UNet), leverages deep learning to achieve precise segmentation and classification of plant leaf diseases. By cascading Symmetric Autoencoders with Attention Residual U-Net model and training on a custom dataset, it surpassed existing methods in identifying four disease classes. The model achieves remarkable accuracy, with a mean pixel accuracy of 95.26% and a weighted mean intersection over union of 0.7451, accurately capturing individual pixels and delineating disease class boundaries. This approach holds great potential in facilitating early plant disease detection and improving crop management practices. Its adoption can significantly impact food security worldwide, addressing a critical gap in the agricultural sector. The results highlight the effectiveness of the proposed strategy in plant disease management and open the door for further research in this field.