Leaf Segmentation Algorithm Based on Improved U-shaped Network under Complex Background

Abstract

In order to segment leaf image under complex background and improve the accuracy of leaf image segmentation, an image segmentation method based on improved U-shaped network is proposed. Based on the Pytorch deep learning framework, the U-shaped network model FPN is improved, the model adopts the encoder-decoder structure, ResNet50 is used as the trunk network, the encoder receives the image input, the feature extraction is accomplished by convolution, and the decoder uses the bilinear interpolation to complete the image reconstruction and outputs the segmentation results. In order to integrate the underlying position features and high-level semantic features better, the feature fusion module is introduced in the decoder. The experimental results show that the model has a significant effect in plant leaf segmentation, and the technical index is better than most traditional image segmentation algorithms.