Texture and Materials Image Classification Based on Wavelet Pooling Layer in CNN

Convolutional Neural Networks (CNNs) have recently been proposed as a solution in texture and material classification computer vision. However, inside CNNs, the internal layers of pooling often cause loss information and, therefore, is detrimental to learning architecture. Moreover, when considering images with repetitive essential patterns, this affects performance subsequent stages, such feature extraction analysis. In paper, solve problem, we propose system new method called Discrete Wavelet Transform Pooling (DWTP). This based on image decomposition into sub-bands, which first level sub-band considered its output. The objective obtain approximation detail information. As result, can be concatenated different combinations. addition, wavelet uses wavelets reduce size map. Combining these methods provides acceptable for three databases (CIFAR-10, DTD, FMD). We argue that helps eliminate overfitting graphs reflect datasets show generalization. Therefore, our results indicate analysis feasible classification. some cases, it outperforms traditional methods.