论文部分内容阅读
以240个“秦冠”苹果水心病果和好果为试材,采集900~1 700nm的近红外波段高光谱图像,选取高光谱图像中的苹果区域作为感兴趣区域(ROI)并计算其平均光谱,分别采用4种特征选择方法和3种核函数支持向量机(SVM)分类器对水心病果进行判别,以探讨利用近红外高光谱成像技术判别苹果水心病的可行性。结果表明:基于卡方检验和支持向量机递归消除(SVM-RFE)2种特征选择法优于基于F检验和决策树的方法。4种特征选择的3种核函数支持向量机(SVM)分类器在1~200个波段下对水心病果的判别正确率分别为:48.6%~70.2%、48.6%~72.0%、33.3%~71.8%、47.2%~70.8%。基于SVM-RFE检验的特征选择下,SVM对水心病果的正确识别率达到72.0%,为该试验选出的最优方法。
In this experiment, 240 near-infrared spectral hyperspectral images from 900 to 1,700 nm were collected from 240 samples of “Qin Guan” apple water and fruits, and then the apple region in the hyperspectral image was selected as the region of interest (ROI) Their average spectra were discriminated respectively by four kinds of feature selection methods and three kinds of kernel function support vector machines (SVM) classifiers to explore the feasibility of using near infrared hyperspectral imaging to distinguish watermelon from apple. The results show that the two feature selection methods based on chi-square test and support vector machine recursive elimination (SVM-RFE) are better than those based on F-test and decision tree. The accuracy of SVM classifier in discriminating watery heart from 1 to 200 bands was 48.6% ~ 70.2%, 48.6% ~ 72.0%, 33.3% ~ 71.8%, 47.2% ~ 70.8%. Based on the feature selection based on SVM-RFE test, the correct recognition rate of SVM to water-heart disease was 72.0%, which was the optimal method selected for this experiment.