Low-Level Feature Representation in the RCSU-Net Model Using Channel-Spatial Attention Mechanism for Semantic Segmentation of Plant Leaves

Smart agriculture leverages advanced technologies like the Internet of Things (IoT), deep learning, and computer vision to address challenges in plant disease detection and management. Semantic image segmentation has become an essential tool for monitoring plant health by detecting subtle anomalies and patterns through visual data analysis. While encoder-decoder architectures such as U-Net show great potential, they face challenges like the loss of low-level features, incomplete texture analysis, and weak edge detection, which limit segmentation accuracy. To overcome these limitations, this study proposes an enhanced U-Net model that integrates ResNet-50 as the encoder and incorporates a channel-spatial attention block in the decoder. This architecture preserves low-level features like edges and textures, enabling precise segmentation of diseased regions in plant leaves. Experimental results demonstrate the model's effectiveness, achieving an Intersection over Union (IoU) of 93.35% and a Dice coefficient of 96.45%, indicating its superior performance. These advancements highlight the model's significant potential for real-world applications in disease detection and efficient crop management. The code for this research is publicly available at https://github.com/Faphnut/Plant-Disease-semantic-segmentation.git