A Real-Time Semantic Segmentation Method of Sheep Carcass Images Based on ICNet

Abstract

How to realize the accurate recognition of 3 parts of sheep carcass is the key to the research of mutton cutting robots. The characteristics of each part of the sheep carcass are connected to each other and have similar features, which make it difficult to identify and detect, but with the development of image semantic segmentation technology based on deep learning, it is possible to explore this technology for real-time recognition of the 3 parts of the sheep carcass. Based on the ICNet, we propose a real-time semantic segmentation method for sheep carcass images. We first acquire images of the sheep carcass and use augmentation technology to expand the image data, after normalization, using LabelMe to annotate the image and build the sheep carcass image dataset. After that, we establish the ICNet model and train it with transfer learning. The segmentation accuracy, MIoU, and the average processing time of single image are then obtained and used as the evaluation standard of the segmentation effect. In addition, we verify the generalization ability of the ICNet for the sheep carcass image dataset by setting different brightness image segmentation experiments. Finally, the U-Net, DeepLabv3, PSPNet, and Fast-SCNN are introduced for comparative experiments to further verify the segmentation performance of the ICNet. The experimental results show that for the sheep carcass image datasets, the segmentation accuracy and MIoU of our method are 97.68% and 88.47%, respectively. The single image processing time is 83 ms. Besides, the MIoU of U-Net and DeepLabv3 is 0.22% and 0.03% higher than the ICNet, but the processing time of a single image is longer by 186 ms and 430 ms. Besides, compared with the PSPNet and Fast-SCNN, the MIoU of the ICNet model is increased by 1.25% and 4.49%, respectively. However, the processing time of a single image is shorter by 469 ms and expands by 7 ms, respectively.

Image Semantic Segmentation Method Based on Deep Fusion Network and Conditional Random Field

“…In principle, pixels with similar distances will be assigned the same semantic labels as much as possible, and pixels with obvious differences need to be assigned different labels. The evaluation index of “distance” defined by color difference and spatial relative distance ensures that the image can be accurately cut at the edge with a large gradient to a certain extent [ 25 , 26 ]. Different from the ordinary conditional random field, the bivariate term in the fully connected conditional random field expresses the correlation between each pixel and all other pixels in the image.…”

Research on Automatic Classification and Detection of Mutton Multi-Parts Based on Swin-Transformer

“…The computer vision object detection method based on deep learning, which has strong independent extraction, learning, and reasoning capabilities for deep and shallow features of sample images, can better solve the aforementioned problems. Zhao et al [ 18 ] generated sheep skeleton images using generative adversarial networks and conducted image semantic segmentation research based on ICNet for the key parts of the sheep skeleton in various scenes; they achieved an average segmentation accuracy of >90%. Meng et al [ 19 ] used image processing technology combined with the back propagation (BP) neural network method to distinguish the sheep back, foreleg, and hindleg meat under different storage time gradients.…”

SheepFaceNet: A Speed–Accuracy Balanced Model for Sheep Face Recognition

“…Although these works effectively reduce the number of model parameters and complexity, related studies [ 14 ] show that reducing the number of parameters and computational complexity cannot effectively improve the speed of the model during inference, and the model may not meet the real-time requirements of actual production. Therefore, [ 15 , 16 , 17 , 18 , 19 , 20 ] designed real-time recognition models, which effectively improve the recognition speed of the model; however, the recognition accuracy needs to be improved. Therefore, we urgently need a recognition model that strikes a good balance between recognition accuracy and recognition speed.…”