Research And Application Of Bamboo Defect Detection System Based On Machine Vision

During the production process of bamboo pieces,there may be appear defects such as broken shape,dirty bamboo,bamboo sections and other defects.Up to now,the defect detection of bamboo pieces is checked by manual visual inspection.A great number of bamboo products manufacturers are eager to utilize machine vision to detect bamboo pieces defects on the spot to replace manual labor and save testing expense.According to the bamboo shape defects and bamboo texture characteristics,this article is mainly based on machine vision technology to classify the bamboo pieces.This article designed a bamboo shape defect determination algorithm.The global threshold method and the maximum stable extreme region(MSER)method are used to compare the segmentation effect of bamboo with considering the gray feature information of the image comprehensively.The maximum stable extreme value region segmentation method has fewer burrs and projections on the edge of the bamboo segment than the global threshold segmentation method,and the background segmentation of the bamboo segment and the bamboo segment is completely clear.Meanwhile,the four arc regions after the differential operation of the maximum inner rectangle of the extraction region and the region are subjected to ellipse fitting.The length of the ellipse and the length of the minor axis are used as a judgment basis for the shape defect.Author proposed a LAWS-based Local Binary Pattern texture features extraction algorithm to achieve bamboo section judging.In this paper,texture features extraction methods based on histogram statistical moment,gray level co-occurrence matrix,LAWS Algorithms and LBP algorithm are studied.Based on this,an advanced texture features extraction algorithm of LBP based on LAWS is proposed.Firstly,the LAWS texture filtering is applied to the texture image to obtain the LAWS texture energy map.Then,the LBP texture features description is performed on the LAWS texture energy map,and the statistical histogram of the LBP profile is extracted as a features vector for neural network classification.The experimental results show that the new algorithm has a classification accuracy of 99% for bamboo sections and bamboo pieces.A complete visual inspection hardware and software system was designed and implemented.The system has undergone long-term testing of large batches of data at the customer's production site.A total of 100,000 bamboo pieces were tested in batches,and the tested bamboo pieces were review manually.According to the results of the field,the detection rate of the qualified bamboo in the visual inspection system was99.5%,the detection rate of bamboo was 99%,the defect of shape was 99.3%,the maximum running time was 20 ms,real-time detection of 4 channels was realized,and the detection speed was 1200 pieces/minute.The system acquired good comments from customers.