基于最大类间方差自适应阈值的五花肉肥瘦自动识别方法

Automatic Recognition Method of Five-flowered Meat with Adaptive Threshold Based on Maximum Inter-class Variance

摘要:

对五花肉的检测常使用化学检测或人工检测方法，但这两种常用方法都需要花费较大的成本和检测时间，且误差较大。为了提高识别准确度，降低检测成本，实现自动化检测。本文以猪肉腹部的五花肉为研究对象，使用最大类间方差自适应阈值法对图像R基色图层进行背景分离，并对图像进行中值滤波处理，从而获取图像的五花肉区域。针对五花肉图像中肥肉与瘦肉对比度不强的特点，采用有限对比度自适应直方图函数来增强肥瘦肉之间的对比度，再使用最大类间方差自适应阈值法分割五花肉图像的肥肉与瘦肉区域。通过实际图像样本的实验结果表明，本文方法比传统新近方法的识别准确率高。这说明本文方法通过自动阈值进行图像滤波处理方法可以有效区分肥肉和瘦肉区域，对其进行有效检测和识别。

Abstract:

Chemical or manual detection methods are often used in the detection of five-flowered meat. The both methods are costly and time-consuming, and the errors are large. In order to improve the accuracy of this research, reduce the cost and achieve automatic detection, we used maximum interval variance adaptive threshold approach to separate the background of R-based color layer and filtered the image by median filtering in this paper. Aiming at the non-strong contrast between fat and lean meat in the five-flower meat image, we used the adaptive histogram function with limited contrast to enhance the contrast between fatness and lean meat, and then use the maximum interval adaptive threshold method to separate the fat and lean meat regions. The experimental results on the actual image samples showed that the proposed method is more accurate than the traditional new method. It shows that our method based on automatic threshold filtering can distinguish the fat and lean meat areas effectively.