Research on Image Segmentation Based on ResUNet

In response to the challenges faced by traditional methods when dealing with complex backgrounds and irregular shapes, such as the problems caused by reliance on manual feature extraction, including insufficient feature extraction, overfitting, and information loss. This study proposes a new leaf segmentation method that integrates ResNet and U-Net, named Res-U-Net. This method combines the advantages of two classic deep learning networks, including upsampling layers, downsampling layers, residual network layers, and skip connections. The downsampling layer extracts multi-scale features through convolution and pooling operations, while the residual network layer utilizes residual connections to alleviate gradient vanishing and explosion issues, speed up training, and enhance the model's generalization capability. The upsampling layer restores the underlying feature map back to the original image size. And at the same time, the skip connection fuses encoder features with decoder features, effectively utilizing information from different scales, reducing information loss, and improving segmentation accuracy and robustness. Experimental results demonstrate that Res-U-Net effectively overcomes the shortcomings of traditional methods, significantly enhances the accuracy of leaf segmentation.