Research On Feature Fusion And Solid Wood Floor Defects Detection Method Based On Compressed Sensing

In forestry processing products, wood flooring has a vast sales market, but the floor surface defects is an important factor affecting their sales results. Traditional artificial vision detection is not efficient but costs high, when floor grading subjectivity, does not apply to the development requirements of modern production. Based on compressed sensing feature fusion and wood flooring defect detection method to improve the level of automation of wood flooring detection method and enrich ways of defect detection and recognition, and has significant theoretical and practical value.In this paper, wood flooring surface defects live knots, dead knots and crack these three major defects research for the study. First, get a certain amount of wood flooring image surface defects through the floor image acquisition system, providing research materials for future study. Through image scaling, image enhancement and image gray median filter smoothing processing amount of image data pre-processing operations to reduce and weaken the influence of wood flooring surface texture noise. By morphological segmentation algorithm will segment the defective part. Extracting geometry and regional characteristics, texture features and invariant moments feature three categories of 25 specific features, according to principal component analysis feature fusion method and linear dimension reduction dimensionality reduction method. Finally, based on compressive sensing theory, design compressed sensing classifier for defect classification. The compressive sensing classifier and SOM neural network classifiers were compared to verify the reasonableness of the classifier design. Fusion classification and feature dimensionality reduction and feature classification experiment does not deal with the characteristics of classification according to the variance of the experimental treatment classification experiments comparing test compressed sensing classifier performance by feature actual classification. In the classification performance test, select the live section 20, section 20 dead,10 crack defect categories altogether 50 sample images to learn, to build the dictionary feature; Similarly, selecting 20 articulated,20 dead knots,10 three pieces of crack defects were a total of 50 samples of image classification experiments. Feature fusion defect classification time is 46.267ms, classification accuracy was 92%; defect feature dimension reduction method of classification time is 44.955ms, classification accuracy was 94%, the results of classification are good.