基于计算机视觉的稻谷霉变程度检测

发布时间：2018-04-25 07:26

本文选题：计算机视觉 + 图像处理　；参考：《农业工程学报》2017年03期

【摘要】：为了实现无损检测稻谷储藏中的霉变,该研究以引起稻谷霉变的5种常见真菌(米曲霉、黑曲霉、构巢曲霉、桔青霉和杂色曲霉)为对象,首先进行真菌培养,制成悬浮液,然后将悬浮液接种到稻谷样品中,对稻谷样品模拟储藏,确定不同霉变程度的稻谷类型,划分为对照组(无霉变)、轻微霉变组和严重霉变组。利用计算机视觉系统对三组稻谷样品进行图像采集和图像处理,提取灰度、颜色和纹理特征,共获取68个图像特征。采用支持向量机(support vector machines,SVM)和偏最小二乘法判别分析(partial least squares discriminant analysis,PLS-DA)构建模型,分别用于无霉变稻谷与霉变稻谷的区分和稻谷霉变类型区分。为了降低模型复杂度和数据冗余,利用连续投影算法(successive projections algorithm,SPA)来消除原始数据变量间的共线性,优选特征值。结果表明:利用所有参数构建的SVM模型能够很好的区分对照组与霉变组,其中建模集和验证集总体区分准确率分别为99.7%和98.4%;SVM模型对于稻谷严重霉变类型的区分效果要优于轻微霉变稻谷,其中对稻谷轻微霉变类型建模集和验证集总体区分的准确率分别为99.3%和92.0%,对稻谷严重霉变类型区分的总体准确率分别为100%和94%,且整体上SVM模型的效果要优于PLS-DA模型。而基于SPA优选特征构建的模型区分结果表明,SVM模型区分效果优于PLS-DA模型,其中,在建模集和验证集中,对无霉变和霉变稻谷总体区分准确率分别为99.8%和99.5%,对稻谷轻微霉变种类区分总体准确率分别为99.8%和90.5%,对稻谷严重霉变种类区分总体准确率分别为100%和95.0%。因此,基于计算机视觉对稻谷霉变检测是可行的,而且SPA优选特征能够较好反映稻谷霉变特征,基于优选特征和SVM模型能够较好地稻谷霉变进行识别和区分,结果较好,可以为实际应用提供技术支持和参考。
[Abstract]:In order to realize nondestructive detection of mildew in rice storage, five common fungi (Aspergillus oryzae, Aspergillus Niger, Aspergillus oryzae, Penicillium citrinus and Aspergillus chromoides) were studied. Then the suspension was inoculated into the rice samples, and the rice samples were stored in a simulated way. The rice types with different mildew degrees were determined and divided into control group (no mildew, mild mildew and severe mildew). Three groups of rice samples were collected and processed by computer vision system. The grayscale, color and texture features were extracted, and 68 image features were obtained. The model was constructed by using support vector machine (SVM) and partial least squares discriminant analysis (PLS-DAA). The models were used to distinguish mildew and mildew of rice and the type of mildew, respectively. In order to reduce model complexity and data redundancy, successive projections algorithm (spa) is used to eliminate the collinearity between the original data variables and select the eigenvalues. The results showed that the SVM model constructed with all the parameters could distinguish the control group from the mildew group. The accuracy of modeling set and verification set is 99.7% and 98.4% respectively for the classification of severe mildew type of rice, which is better than that for mildly mildew type of rice, and the accuracy of SVM model is 99.7% and 98.4% respectively for the classification of severe mildew type of rice. Among them, the total accuracy of modeling set and verification set for mildly mildew of rice was 99.3% and 92.0%, respectively, and the total accuracy for classification of severe mildew was 100% and 94%, respectively, and the effect of SVM model was better than that of PLS-DA model on the whole. The result of model differentiation based on SPA optimal selection features shows that the classification effect of SPA model is better than that of PLS-DA model, among which, in the modeling set and validation set, The overall accuracy of classification of mildew and mildew was 99.8% and 99.5% respectively, that of slight mildew was 99.8% and 90.5%, and that of severe mildew was 100% and 95.0% respectively. Therefore, it is feasible to detect rice mildew based on computer vision, and SPA optimal selection features can better reflect the characteristics of rice mildew. Based on the optimal selection feature and SVM model, rice mildew can be better identified and distinguished, and the result is better. It can provide technical support and reference for practical application.
【作者单位】：南京农业大学食品科技学院;南京财经大学食品科学与工程学院江苏省现代粮食流通与安全协同创新中心;
【基金】：中央高校基本科研业务费专项(KYLH201504) 国家自然科学基金项目(31671926,31671925) 国家粮食公益性行业科研专项(201313002-01)
【分类号】：S435.11;TP391.41

【相似文献】