Modified U-Net for plant diseased leaf image segmentation

Early detection and recognition of plant disease is a prerequisite for controlling plant disease, and one of the key steps is to segment plant diseased leaf images. However, this task is challenging because diseased leaf images are often very complex, with irregular shapes, variable sizes, various shapes, rich colors, fuzzy boundaries and messy backgrounds. An improved U-Net (MU-Net) is constructed for plant diseased leaf image segmentation by introducing a residual block (Resblock) and a residual path (Respath). Resblock is introduced into U-Net to overcome gradient disappearance and explosion problems, and 2 Respaths are used instead of 2 skip connections to improve the transformation of corresponding feature information between the contraction path and the expansion path. Furthermore, Resblock and Respath are combined, which can increase the network depth and improve the network's expression ability. Experimental results on a plant diseased leaf image dataset show that the proposed method can improve the accuracy and efficiency of plant diseased leaf image segmentation.

Automated summary

Early detection and recognition of plant disease is crucial for controlling plant disease. However, segmenting diseased leaf images is challenging due to their complexity. This study proposes an improved U-Net (MU-Net) for plant diseased leaf image segmentation by introducing a residual block (Resblock) and a residual path (Respath). Experimental results demonstrate that this method enhances the accuracy and efficiency of plant diseased leaf image segmentation.