SCAG-Enhanced U-Net for Wheat Yellow-Rust Semantic Segmentation in Multispectral Remote Sensing

The wheat yellow-rust disease poses a serious risk to global wheat production, making effective detection methods essential. This study aims to enhance wheat yellow-rust detection accuracy by investigating the use of spatial-channel attention gates (scAGs) in semantic segmentation with multispectral remote sensing images. While scAGs find applications in medical image segmentation and precision agriculture, this study extends usage for wheat yellow rust detection. Integrated into the skip connections of the U-Net model, scAGs aim to refine feature extraction and improve segmentation performance. Furthermore, to address a limitation in prior work that used only one upsampling method, this study explores multiple techniques—bilinear, bicubic, nearest neighbor, and transposed convolution—optimizing performance. According to experimental results, bicubic interpolation delivers the best performance, significantly enhancing wheat yellow-rust disease detection accuracy.