Research On Wood Floor Image Fusion Segmentation And BP-SOM Networks Identification Method

The surface defects of solid wood floor seriously affect the floor’s quality and level. Traditional artificial visual floor surface defects detection method, which is the main means to identify and rank floors, has high production costs and low efficiency and this method has also been unable to meet the needs of modern production. Detection technology based on computer vision defects can detect surface defects without damaging the wood, providing new ideas of defect detection and having a very broad research and application value. In this paper, Sound knots, dead knots, pin knots and cracks are the main subject of this research.Firstly, we study and build a solid wood floor image acquisition and level sorting system, and get a certain amount of floor surface images. Then choose the proper image preprocessing algorithm such as image gray, image enhancement, noise reduction and image sharpening. To solving the problems of ineffective of traditional Otsu segmentation algorithm, spending lots of time of purely morphological segmentation algorithm, and poor efficiency of the classic region growing algorithm, we design the defects segmentation algorithm based image fusion. This algorithm uses image scaling to position defects quickly; uses creating taboo table, design tabu search rules and dual-threshold growth low to finish defect segmentation. For the feature extraction and selection, we get21features. We determine the minimum variance within the class, the largest inter-class variance as feature selection principle, and13features are chosen as input features in pattern recognition finally. At last, the neural network with a strong ability of generalization is chosen as defect identification model, and we research and design BP and SOM network model. To give full play to the BP network’s data compression capabilities and SOM network’s pattern clustering function, we design BP-SOM hybrid neural network model. The recognition accuracy rates of BP network, SOM network and BP-SOM hybrid networks are:87.5%,90%and95%, and a single sample recognition time are:2.365ms,0.172ms and2.497ms.In this paper, all tests are completed by MATLAB2011b, and the computer is clocked at2GHZ. The experiment reveals that:the defects segmentation based on image fusion algorithm designed in this paper can segment defects faster and more accurate; BP-SOM hybrid network model can effectively improve the accuracy of defect identification, while the elapsed time can meet the requirements of online sorting.