基于DT-CWT和LS-SVM的苹果果梗/花萼和缺陷识别

Apple stem/calyx and defect discrimination using DT-CWT and LS-SVM

该文提出了一种基于双树复小波变换(DT-CWT)和最小二乘支持向量机(LS-SVM)区分苹果的果梗/花萼和缺陷的方法。对苹果图像使用DT-CWT分解,使用变换后得到的高频子带系数的均值和方差构造特征向量,然后使用最小支持二乘向量机作为分类器进行分类。对180幅苹果图像进行了试验。讨论了DT-CWT分解层数以及目标图像大小对分类正确率的影响。试验结果显示,使用3层DT-CWT对大小为64×64子图像进行小波分解提取纹理特征,能达到最好的分类效果,分类正确率可以达到95.6%。

This paper proposed a method for apple stem/calyx and defects discrimination by integrating the Dual Tree Complex Wavelet Transform （DT-CWT） and Least Squares Support Vector Machines （LS-SVM） method. The DT-CWT was used to decompose the apple images, and the feature vectors were generated by computing mean and standard deviation from the coefficients of individual wavelet subbands and the LS-SVM was used for classification. 85 apple images were tested, in which there were 25 stem and calyx images respectively and 35 defect images. Moreover, the influence of the DT-CWT decomposition levels on the classification rate was analyzed. The result showed that with 3-level DT-CWT the best classification result could be obtained, and an overall detection rate of 97.1% was achieved.