Research On Bamboo Strip Defects Recognition Technology Based On Computer Vision

Bamboo strips must be picked before gluing,removed the defective bamboo strips and graded by their color,to improve the quality of bamboo glulam.The defects of bamboo strips mainly include wormhole,mildew,rotten,remaining outter skin,remaining inner skin,cracking and other defects.At present,the task is done by labors in most glulam lines.As a result of that,its distinction of defects is inaccuracy,low efficiency and high cost.Based on the computer vision technology,the features and recognition method of bamboo defects are studied in this works,to realize automatically sorting and grading bamboo strips,accelerate the automation process of bamboo glulam production,improve the quality of bamboo glulam,and reduce the production cost.An image acquisition system was set with an industrial camera for capturing photos of the bamboo strips surface.After a large number of bamboo strips images were captured,they were filtered and equalized,to eliminate the image noise and reduce the error of features.After the characteristics of various type bamboo strips were compared,the defect features of bamboo strips were defined and analyzed from four aspects including the color difference,the textile,the frequency region amplitude spectrum and the mathematical morphology.After the differences between the features of various type bamboo strip images were analyzed,some features that were not significantly different were screened off,and 25 features were obtained,which were the first moment,the second moment and the third moment of the color difference;the first moment and the second moment of the gray-level distribution;the mean and the contrast of the gray-level difference matrix;the contrast,the energy,the second moment of the gray-level co-occurred matrix;the mean,the variance and the entropy of the frequency amplitude spectrums in range R3C3-50 and R3-50C3;and the maximum,the minimum and the mean of width,length,area,perimeter,circularity of the connected region;and the width and length of the connected region with the largest area.A feature vector for recognizing the defects of the bamboo strips was constructed with above 25 features.A BP neural network was established to determine whether the bamboo piece is defective,its input was the feature vector,and the output was defective or not.After the neural network was trained,the accuracy judging it was defective or not was over 98%.After that,a BP neural network was constructed to recognize the type of the bamboo defects,the input was the feature vector,but the output was the defect type coded with 8-bit binary.After the neural network was trained,the recognition rate of bamboo strip defects was over 91%.The results show that the feature vector can be used for judging and recognizing bamboo strip defects effectively.