Research On Technology Of Solid Wood Floor Sorting Based On Machine Vision

Continued increases in the cost of materials and labor make it imperative for Wood products enterprises to control costs by improved quality and increased productivity. Detecting and classifying surface by machine vision are required to grade and sort wood floor. Sorting system can Improve production efficiency and reduce long-term cost. The improved quality create value-added opportunities for some wood floor to be used for high performance applications. The existing algorithms are difficult to meet the requirements of online detection. This paper aimed at researching on the key technology of an inspection and sorting system.Image segmentation is the primary problem of defects and texture discrimination and classification. In this paper, a fast superpixel combination algorithm is proposed, HSLIC algorithm is used to get super-pixel segmentation image, the improvement is mainly from speed and adaptive threshold. At the same time, using MeanShift algorithm to get segmentation image, and segmentation results using different algorithm are analyzed. The result shows that, Three kinds of defects were segmented with good results. The algorithm has low complexity, it is Suitable to use in the online system.The color, shape and texture feature of defect image is extracted, using OOB error to calculate the feature importance, analysis the results and the extraction time. Through the comparison of the overall performance, Tamura texture and color histogram are the optimal parameter. Random forest algorithm is improved from of speed and the generalization error, the color histogram and Tamura texture are combined input, and The high similarity decision tree were merged. The improved Random Forest has higher classification accuracy and faster prediction speed, it could meet the requirement of Online sorting system.In order to realize the plate surface texture classification, Double LBP algorithm is proposed to extract the edge information of wood texture. Using the HOG operator to describe direction feature of large scale texture. Because of the sample unbalanced, this paper analyzed the small sample by using three order polynomial interpolation to get 200 sample incloud original sample. The OOB errors of interpolation sample are reduced greatly, so that the overall OOB error will also reduce. The experimental results show that the improved random forest algorithm has higher classification accuracy, it could reach 96.77%. Using this algorithm, it only cost 0.40s to complet the sorting process from image segmentation to texture classification, it can fully meet the requirements of online system.In summary,the application of improved HSLIC superpixel segmentation algorithm in wood surface defect images segmentation is better than other similar algorithm.Based on computer vision technology, using solid wood floor surface texture, color and geometrical characteristics as input information, the random forest classifier can be improved with higher precision and faster prediction speed, it could basically meet the requirement of online sorting system.