Identification of defects on bamboo strip surfaces based on comprehensive features

The basic gluing unit of laminated bamboo timber (LBT) is a long and narrow bamboo strip. Each strip has a rectangular cross-section and is made from bamboo stalk by cutting, splitting, rough milling, caramelizing or bleaching, drying, and finish milling. Defects on the surface of the bamboo strips arise due to the invasion of insects, molds, and rotting fungi, or from improper processing techniques or equipment parameters. It is therefore important to identify and pick out defective bamboo strips. This article analyzes and compares the differences between defective and non-defective bamboo strip images from three perspectives: color, texture, and mathematical morphology. Obvious differences exist in color between defective and non-defective bamboo strips. Defective bamboo strips are darker than non-defective bamboo strips. In gray-level difference method (GLDM)-based and gray-level co-occurrence matrix (GLCM)-based texture features, the direction and offset affect the differences in the main characteristics between defective and non-defective bamboo strips. The main characteristic differences between defective and non-defective bamboo strips are more obvious when the GLDM array or GLCM matrix is generated by the offset of five pixels in the 90° direction. The length, width, perimeter, area, and circularity of various defective bamboo strips obtained by morphology operators are larger than those of non-defective bamboo strips because the faulty blocks in the defective bamboo strip images are usually large in length and width. Finally, an eigenvector is composed of features that have obvious differences between defective and non-defective bamboo strips. This serves as an input to the Back Propagation (BP) neural network to identify whether the bamboo strips are defective, and the recognition accuracy rate reaches 96.3%.