Borehole Image Classification Using NSCT, Hu moment invariants and SVM based APBCS System

The role played by Axial View Borehole Camera System (AVBCS) in acquiring borehole images and in analyzing geographical structures opens new possibilities in geographical exploration. Considering the current situations, where the borehole images are interpreted by humans, a borehole image classification method based on NSCT and Hu moment invariants is proposed to classify them into three main classes of border, fracture and intact rock mass images. Firstly, the borehole images are decomposed into one low frequency subband and multiple band pass directional subband coefficients by NSCT, to obtain the mean and variance of these subband coefficients. In the subsequent phase, 18 features are acquired with NSCT which are combined with 7 Hu moment invariant features of source images to form the image feature vector. Finally, after the feature optimization based on variance ratio test, the optimal feature vector is furnished as input to the SVM to obtain the image classification results. Experiments show that the proposed classification method can effectively classify three major borehole images with the accuracy of 94.44% in the training samples and 93.33% in the testing samples.