基于深度学习的新型可视耳标识别方法

A new visual ear tag recognition method based on deep learning

针对目前主流的耳标识别方式抗环境干扰能力差的问题,结合通信系统中的信源编码,提出了一种耳标轮廓自设计的新型可视耳标,并利用深度学习方法采集识别确定猪只个体号码。系统设计包括耳标样式设计、数据采集、耳标轮廓分割提取、码位复原、译码输出5个部分:耳标样式采用了结合卷积码编码的轮廓编码方式;数据采集部分采用了帧差检测与耳标分类方式;耳标轮廓分割提取部分采用OCRNet−HRNet18语义分割算法,码位复原采用图像处理技术,译码输出部分采用维特比译码,获得耳标号码结果。系统部署至Jetson Xavier NX上,实现了系统落地。研究结果表明,本文方法实现了新型自设计可视耳标的识别,使得耳标识别抗环境干扰能力更强,为智能养殖中的个体监测提供了新思路。

In order to solve the problem that the current mainstream ear tag identification method has poor antienvironmental interference ability,in this paper,an innovative method is proposed by combining the source code in the communication system.It is a new type of visual ear tag with self-designed ear tag outline,which uses the deep learning method to collect,identify and determine the individual number of pigs.The system includes five parts:ear tag style design,data collection,ear tag contour segmentation and extraction,code bit recovery,and decoding output.The ear tag style adopts contour coding combined with convolution code coding,and the data selection part adopts frame difference detection and ear tag classification methods.The ear tag contour segmentation and extraction part adopts OCRNet-HRNet18 semantic segmentation algorithm.The code bit restoration adopts image processing technology,and the decoding output part adopts Viterbi decoding to obtain the result of ear tag number.The system is deployed on the Jetson Xavier NX to realize the system landing.The results show that the new self-designed visual ear tag identification is realized,which makes the ear tag identification more resistant to environmental interference,providing a new idea for individual monitoring in intelligent breeding.