基于纹理和颜色特征的甜瓜缺陷识别

为了提高硬皮甜瓜缺陷分类的正确率,提取基于纹理和颜色的综合特征,采用支持向量机分类器构造了甜瓜缺陷的自动检测系统。对甜瓜图像可疑区进行了纹理分析,提取灰度共生矩阵的4个特征参数,经过比较实验得出,对比度和角二阶矩2个参数对甜瓜瓜蒂、花萼、擦伤和霉变有明显的可区分性。在可疑区域上提取了由R、G、B分量及其算术运算组成的12种颜色特征,通过实验筛选出4种具有较好区分性的颜色特征。实验结果表明,由这些优选出的纹理与颜色特征组成的综合特征及支持向量机分类器对甜瓜缺陷的识别正确率达到92.2%。

In order to improve the accuracy of muskmelon's defect detection, an automatic defect detection system based on support vector machine (SVM) was set up by adopting complex features of texture and color. Four textural parameters and twelve color features of combinations from RGB were tested for the discriminability in stem, calyx, bruise and mildew. Through the experiments, two textural and four color features with good discriminability were selected and treated as the complex features. The results indicated that with the complex features and SVM, the accuracy of classification on muskmelons was up to 92.2%.