基于多特征融合和水平集的碧根果品质检测

碧根果在生产加工过程中易酸败，误食会对人体造成多方面危害。针对此问题，提出一种基于多特征融合和水平集的碧根果品质检测方法。以薄壳碧根果为研究对象，首先，对采集的原始图像进行预处理，解决目标对象与背景区域比例不匹配问题；然后，通过改进边缘指示函数的自适应距离正则化水平集算法（Distance regularized level set evolution, DRLSE）对图像进行感兴趣区域（Region of interest, ROI）分割，最后提取图像灰度直方图统计特征、灰度共生矩阵、Tamura和局部二值模式等多特征，并进行融合分析，建立支持向量机（Support vector machine, SVM）判别模型，实现碧根果无损品质检测。试验采集了200个正常、酸败碧根果样本图像，对其进行图像酸败及多特征分析。结果表明，采用本文方法判别碧根果酸败的分类准确率高达96.15%，在此基础上识别碧根果酸败程度，平均识别率为90.81%。

Detection of Pecan Quality Based on Multi-feature Fusion and Level Set

Pecan is one of the top ten nuts in the world. Because of its good taste and rich nutrition, it is loved by people. But pecan is easy to deteriorate in the process of production and processing. Mistaken food can cause many hazards to human body. To solve this problem, a method for detecting the quality of pecans was proposed based on multi feature fusion and level set. Taking thin shelled pecans as research object, and the original image was preprocessed to solve the problem that the target object did not match the background area. The adaptive DRLSE method with improved edge indication function was used to segment the pecans in the image, and the statistical features of the gray histogram of the image were extracted. Multi features such as co occurrence matrix, Tamura and local binary mode were combined and analyzed. The SVM discriminant model was established to realize the non destructive quality detection of pecans. The experiment collected 200 normal, rancid pecans sample images, and subjected to image rancidity and multi feature analysis. The experimental results showed that the adaptive DRLSE segmentation method with improved edge indication function can complete the segmentation better than the traditional method even inside or outside the target. The accuracy of the method was as high as 96.15% in judging whether pecan was rancid or not, and on this basis, the average recognition rate was 90.81% in judging the degree of pecan rancidity.