Research on Pig Breed Identification Algorithm Based on Image Segmentation and Pattern Recognition Technology

This paper discusses the application of image segmentation and pattern recognition technology in pig breed identification, and proposes a new framework combining NAM-YOLO algorithm and case segmentation network YOLACT. We first adopted the target detection network YOLOv5 for target positioning. The experimental results show that the preliminary application of YOLOv5 in the identification of pig breeds has achieved remarkable results. However, in practice, there are still some challenges, such as missing detection (not detecting all pigs) and false detection (incorrectly identifying non-pig objects as pigs). These problems are usually caused by a change in the size of the target and a shift in the position of the detection box. To solve these problems, this paper proposes a NAM-YOLO algorithm based on the improved normalized attention mechanism. By introducing the normalized attention mechanism to enhance the adaptability of the model to different scale targets, the false detection rate and missing detection rate are effectively reduced. In addition, the position error of the detection frame is corrected by boundary extension technique to ensure that each individual can be correctly identified. After the initial inspection, we divided the entire image into several small images and processed these small images using the YOLACT instance segmentation network to achieve accurate segmentation of individual pigs. The ability of YOLACT to generate instance masks directly from images without relying on additional segmentation modules or complex training processes makes it ideal for the individual pig segmentation task in this study. By applying YOLACT to the small images obtained above, it is possible to achieve fine segmentation of each individual pig, so as to obtain more accurate morphological characteristics information. To sum up, the framework based on NAM-YOLO and YOLACT proposed in this paper shows good potential in solving complex problems in pig breed identification, the recognition accuracy is significantly higher than that of traditional methods based on artificial features, and it can maintain high stability under different lighting conditions and shooting angles.