The reason of frequency band derangement and the rule of frequency band division of wavelet packet transform were analyzed. The correct frequency band sequence of wavelet packet transform (WPT) was found. Then in this study, because of the texture feature information is on the middle frequency band, according to the conclusion, we select the correct frequency band sequence of WPT to classify tea categories, the images of tea were captured by multi-spectral imager (3CCD). Firstly, image was transformed by WPT, the decomposed lever was 3, so 8 nodes were obtained, and the 8 nodes were rearranged. The node (2,3) and node (2,7) were selected as the input texture feature set of training and prediction using by least squares-support vector machine (LS-SVM). The nodes which were selected by the correct frequency band is satisfactory as evident from the experiments; up to 100% accuracy is obtained using LS-SVM
[1]
Yang Ya-jing.
Model of recognition system of underwater acoustic signal using high frequency wavelet-pocket energy method
,
2006
.
[2]
Soo-Hyung Kim,et al.
Two-dimensional wavelet-packet-based feature selection method for image recognition
,
2000,
IS&T/SPIE Electronic Imaging.
[3]
Kuen-Tsair Lay,et al.
Image coding based on energy-sorted wavelet packets
,
1995,
Other Conferences.
[4]
Gangbing Song,et al.
Damage diagnosis of framework structure based on wavelet packet analysis and neural network
,
2004,
SPIE Smart Structures and Materials + Nondestructive Evaluation and Health Monitoring.
[5]
S. Mallat.
A wavelet tour of signal processing
,
1998
.