基于YOLO目标检测的生猪多阈值Otsu分割方法  

Multi-threshold Otsu Segmentation Method of Pigs Based on YOLO Target Detection

作　　者：李德平 朱伟兴[1] LI De-ping;ZHU Wei-xing(School of Electrical and Information Engineering,Jiangsu University,Zhenjiang 212013,China)

机构地区：[1]江苏大学电气信息工程学院,江苏镇江212013

出　　处：《软件导刊》2021年第8期179-184,共6页Software Guide

基　　金：国家自然科学基金项目(31872399)。

摘　　要：采用一种基于YOLO目标检测的多阈值Otsu分割方法对采食区域的生猪进行分割。首先,利用自适应直方图均衡化对彩色图像进行增强并划分采食区域,训练一个YOLO目标检测网络,提取出采食区域内的生猪目标;然后在每个目标区域内利用Otsu方法计算多个阈值并对图片像素进行分类,将属于前景的像素叠加起来并通过形态学处理得到精确的分割结果。实验结果表明,当设置两个阈值时,分割精度和处理速度较好。挑选出不同情况下生猪在采食区域的500帧图片进行测试,平均分割率(所有在采食区域的生猪被完整分割)为88.65%,算法平均运行速度约为0.4s/帧,有效解决了粘连生猪分割问题,为类似图像的目标提取提供了一种新方法。Propose a multi-threshold Otsu segmentation method based on YOLO target detection,which is aimed at segmenting pigs in the feeding area.First,use adaptive histogram equalization to enhance the color image and divide the feeding area,and then train a YOLO network to detect the pig targets in the feeding area.In each target area,the Otsu method is used to calculate multiple thresholds to classify the picture pixels,and the pixels belonging to the foreground are superimposed and processed by morphology to obtain accurate segmentation results.Experimental results show that when two thresholds are set,the segmentation accuracy and processing speed are better.500 frames of pictures of pigs in the feeding area were selected for testing under different conditions.The average segmentation rate(all pigs in the feeding area were completely divided)was 88.65%,and the average running speed of the algorithm was about 0.4 seconds/frame.At the same time,it effectively solves the segmentation of sticky pigs,and also provides a new method for the extraction of targets in similar images.

关 键 词：猪 YOLO目标检测 多阈值 OTSU 彩色图像增强 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]