Using wavelet transform and multi-class least square support vector machine in multi-spectral imaging classification of Chinese famous tea

This article presented an intelligent method for recognition of different types of Chinese famous tea based on multi-spectral imaging technique. Two kinds of feature extraction methods including gray level co-occurrence matrix and wavelet transform (WT) were adopted for mining characteristic of multi-spectral image. Then multi-class least square support vector machine models were adopted for classification of multi-spectral image, which has little been used in this domain. Meanwhile the receiver operating characteristic (ROC) curve analysis was used to evaluate the performance of multi-spectral imaging classifier. To explore the structure of the wavelet textural features (WTFs), principal component analysis (PCA) was performed based on all the WTFs, and the most important features were detected through loading weight analysis of PCA. In experiments, the potential of WTFs was confirmed for extraction of characteristic from multi-spectral image with high recognition accuracy of 96.82%. And 18 WTFs were detected as the most important features for recognition by PCA. Furthermore, it can be found that the 18 features were the textural features of “contrast” of wavelet sub-space images. This finding may give great help for later research about multi-spectral image classification. The experimental results indicate that the proposed method is effective for recognition of multi-spectral image of different types of Chinese famous tea, the WT is an effective method for mining knowledge from mass multi-spectral imaging information, and PCA can be used to clear the structure of the WTFs.

► This article put forward an intelligent multi-spectral imaging technique to classify Chinese famous tea.
► The feature extraction way combined by gray level co-occurrence matrix and wavelet transform was proved as an effective approach for revealing characteristic information from multi-spectral image.
► Principal component analysis was demonstrated as a useful tool for exposing the inherent structure of wavelet texture feature, and discovering the most important features.
► This article proved the potential of multi-class least square support vector machine in multi-spectral imaging classification.