Application of Pyramid Scene Parsing Network in leaf segmentation for Wheat Stripe Rust

This paper addresses wheat stripe rust leaf segmentation using the Pyramid Scene Parsing Network algorithm (PSPNet). Unlike previous YOLO-guided GrabCut and DeepLabV3 approaches, PSPNet excels in mining global contextual information, embedding challenging scenarios in a Fully Convolutional Network (FCN)-based pixel prediction framework for precise segmentation. Notably, PSPNet achieves a recognition accuracy of up to $\mathbf{9 5. 0 4 \%}$, surpassing the limitations of image-dependent methods. The study’s background discusses current segmentation shortcomings, motivating the exploration of PSPNet. The research aims to enhance accuracy, comparing PSPNet against YOLO-guided GrabCut and DeepLabV3, showcasing its superior performance. In conclusion, this research establishes PSPNet as a robust solution, effectively addressing challenges in wheat stripe rust leaf segmentation, offering improved accuracy, and demonstrating enhanced generalization capabilities.