Identification Of Dangshan Pears Surface Defects Based On Machine Vision

With the improvement of economic conditions,people have higher and higher requirements for the quality of fruits.The quality of fruits will directly affect the interests of businesses and the desire of consumers to buy,and whether the surface defects of fruits directly affect the division of fruit grade.Judging by the naked eye not only has a certain degree of subjectivity,but also it is inefficient.Therefore,it is of great significance to study the automatic sorting system for fruit defects based on machine vision.In this thesis,taking Dangshan pear as the research object,through the establishment of the experimental hardware system,we collected the image of the Dangshan pear,and used the image processing technology to study the automatic recognition of its surface defects.The three kinds of defects on the surface of Dangshan pear,brown spots,bruises and The rot were found,as well as the peduncle area that is confused with the defect area.These four patterns were identified and classified.The research content includes image collection,defect area extraction,defect feature extraction,and classification and identification.Focus on the following four areas:(1)Segmentation of defect area.For Dangshan pear,the pixel gray value of the defect and peduncle area is significantly different from the intact area.Therefore,the defect area pixels are detected based on the color information,and the surface defect area of Dangshan pear is extracted through threshold segmentation,edge extraction and morphological processing.(2)Feature extraction of suspicious areas.In order to extract the texture features of the suspicious area better and solve the problem that the geometric features and color features of the scratches and peduncle areas are similar and difficult to distinguish,this paper proposes a new algorithm based on Gabor filter to extract texture features of Dangshan pear surface defects.For each suspected defect image,40 detailed images of five dimensions and eight directions are obtained through filter decomposition.The gray average and gray variance are calculated for each detail image to form an 80-dimensional texture feature vector.(3)The dimensionality reduction of feature vectors.Because the dimension of the feature vector is high,and the features are related to each other,this paper uses principal component analysis(PCA)to reduce the dimension of the feature matrix,and then combines the geometric features and color features to form the final feature matrix.Compared with the gray level co-occurrence matrix algorithm,the experimental results show that the proposed texture extraction can more effectively classify the surface defects of Dangshan pear.(4)Classification of defects.In this paper,BP neural network and support vector machine(SVM)classifiers were used to identify the surface defects of Dangshan pear.Through the comparison of the final experimental results,the recognition accuracy of SVM was higher than that of BP neural network classifier.The average recognition accuracy rate is 91.5%.Therefore,the algorithm proposed in this paper can accurately and effectively classify and recognize the surface defects of Dangshan pear.