Features Extraction and Classification of Wood Defect Based on Hu Invariant Moment and Wavelet Moment and BP Neural Network

Abstract

Wood defect will reduce wood properties, wood quality and use value, so it is of great practical significance to detect wood defect[1]. The key to feature extraction of target defect image is target recognition and classification. Moment feature is a common feature descriptor in defect extraction algorithm. Aiming at the problem that the seven feature components of Hu moments differ greatly in magnitude and are affected by scale factor, based on the principle and characteristics of invariant moments and wavelet energy, a feature extraction algorithm based on wavelet moments is proposed and applied to the feature extraction of wood defects. Finally, the experiment collects the actual wood defect image, decomposes the preprocessed image into three sub-images by wavelet transform, calculates the modified Hu moment invariants for the sub-images, takes the moment invariants as the feature variables, and obtains the recognition results by using the minimum neighborhood distance classification. The experimental results show that the feature extracted by this method has the invariance of translation, rotation and scale, and can reflect the important and original attributes of the target image. Compared with the traditional Hu moment, the recognition rate is significantly improved, and the expected goal is achieved.