Defect Detection On Colorful Image Of Veneer Based On Morphology

The main defect of timber in two ways:Firstly, a variety of defects would produced during the growth of process(knots, wormholes and corrosion); secondly is generated in the production process of manufacturing defects (degumming, deformation, etc.)A variety of defects(knots, wormholes, corrosion, etc.) are prone to appear in the growth process of wooden boards. These defects directly affect the quality of veneer, thus indirectly affecting the intensity and level of man-made sheet. With the development of the timber industry, traditional manual processing are not able to meet the needs of modern industry in terms of yield and quality. The application of computer vision technology on board defect detection overcome not only the problems of high labor-intensity, low productivity and poor economic in traditional production processes, but also effectively improve the accuracy and speed, reduce human error subjective in defect detection, it’s even able to classify defects on board surface.This study starts from the characteristics of object images such as geometry shape and burr of defects on veneer images surface which leads to superfluous information. Therefore, mathematical morphology filtering technology combines threshold segmentation and cluster analysis suit this study well. Combined with characteristics of the target image, the study comes up with two new method of board image defect detection. After that, a new BP neural network is designed to classify defects in different parts. At last we construct a rapidly detect and classify systems of veneer defect images. The main research contents are as follows:Firstly, the article summarizes today’s mainstream methods on image segmentation, based on methods above, a new veneer color image defect detection system was designed, including lighting systems, frame grabbers and software components.Then, this article detects the full range of veneer images with an improved filter of morphological,which is designed with multi-angle structural elements, without blind angle, furthermore, proposes two defect segmentation method that are based on the theory above. One is based on three separate elements in color space of HSI:turn images into the HSI space from RGB, detect edge of each elements anting the characteristics of the veneer, reintegration and filling, and finally extract the edge. Another one is based on the K-means clustering algorithm, clustering analysis the veneer images in the Euclid space, selected the optimal image segmentation results in multiple images in different clusters. Experimental results show that the color space of the two methods for detecting a defect in the board can make full use of the image information carried by a color image.Finally, the article uses the detected images as HSI sample libraries, trains a BP neural network for classify images with geometric characteristics, selects four eigenvalues as the network input, divides the defects area into knots, wormholes and corrosion. Experimental results are satisfactory.