A Chicken Part Segmentation Method Based on CP-U Net
Chicken part segmentation is the main task of chicken segmentation.This article proposes a chicken part image seg-mentation algorithm based on U-Net with the aim of chicken part segmentation.According to the needs of chicken part detection,appropriate types of industrial cameras,cameras,light sources,and PC mechanisms are selected to form an image acquisition sys-tem for chicken parts.3000 images of chicken parts are collected,and then data augmentation and expansion are carried out to con-struct a chicken part dataset.A CP(Chicken parts)-U Net chicken part image segmentation model is constructed to extract chick-en part features.In order to achieve end-to-end semantic segmentation of chicken part images,the following steps are taken.First,through pooling calculation,deep and shallow features of chicken parts can be obtained.Then,through multiple deconvolution pro-cesses,feature fusion can be obtained.Finally,a binary image of the chicken part area is generated.In order to evaluate the segmen-tation performance,three evaluation criteria are used,which are mean intersection union(MIoU),mean pixel precision(MPA),and precision(PA).The CP-U Net computational network is compared with three classic algorithms.The experiment shows that the PA,MPA,and MIou of CP-U Net for semantic segmentation of chicken parts are 93.65%,89.12%,and 85.37%,respectively.The three other image semantic segmentation methods compared are 11.23%,8.74%,and 6.68%higher,respectively,and the process-ing time for a single chicken part image is 44 ms shorter than SegNet.
chicken partssemantic segmentationU-Netdeep and shallow feature fusion
国家科技期刊平台
NSTL
万方数据
