Region segmentation of sheep ribs based on fully convolutional neural network

Abstract

The accurate segmentation of the lamb rib area is one of the key technologies for the research of split intelligent sorting robots. In order to accurately segment the lamb ribs on the conveyor belt, this paper takes lamb ribs as the research object and proposes a lamb rib image segmentation model based on fully convolutional neural networks (FCN). First, the lamb rib image is collected, and after preprocessing, the lamb rib image data set is established. Then, based on VGG16 as the basic network to build FCN8s model, and used the Tensorflow deep learning framework to achieve model training. Finally, by introducing precision (PA), mean pixel precision (MPA), average cross-combination ratio (MIoU) three image semantic segmentation standards, the segmentation performance of the FCN model is evaluated. The results show that for the lamb rib data set, the FCN8s model has an average merge ratio (MIoU) of 84.26%. Compared with level set and FCM, Miou of fcn8s model is improved by 6.47% and 11.12% respectively.