Research And Implementation Of Key Techniques For Defect Detection Of Round-Like Fruits Based On Machine Vision

Fruits are favored by people because of the rich nutrition they have.However,during the transportation of fruits,many factors can damage their surface,such as environmental changes,diseases of pests,collision,friction and so on.Therefore,defect detection on fruits must be implemented before they are officially distributed on the market.For this problem,the key technologies of round-like fruit surface defect detection are studied,and different algorithm are appropriately improved based on analysis.The main work of this paper includes the following four aspects:Firstly,a defect detection method based on neighborhood color description features and sparse representation is proposed.Due to most defects are visually different in color from non-defects,this section not only analyzes the color similarity between central patch and its neighborhood patches and distant neighborhood patches,but also combines the number distribution ad spatial distribution of the central patch itself.Based on this,a reasonable and effective color feature extraction method is proposed.Secondly,based on sparse representation convolutional sparse representation,a dual-type dictionary and multivariate detection method are proposed.Due to the various colors and random texture on some fruits,in order to reduce the pressure of dictionary learning during training phase,this section divides the image patches by constructing the maximum response image and maximum response orientation image,and then use the dual type dictionary learning.For the problem that using only one single variable of reconstruction error in detection phase may leads to low reliability of the result,this section combine multivariate to construct low-dimensional detection features and judgement regions to determine the final defect region.Experiments show that the defect detection effect of this method is much better than the detection on reconstruction error.Thirdly,based on the low rank decomposition model,a defect detection method is proposed,which can not only fuse the high-level and low-level information of the image,but also has the sensitivity to detect insignificant defects.For the insufficient use of image information,the redundancy of atomic set matrix,which can easily cause similar characteristics of the data,and the existence of insignificant defects,which can lead to missed detection or false detection,this section will try to solve these problem on three parts: combining high-level image information,updating the atom set matrix dynamically and increasing the distance between vectors of atomic representation.Experiments show that the improved low rank representation has made decrease in both missed detection and false detection.Finally,contrast experiments are made on the four different defect detection algorithms proposed in this paper.Experiments show that the improved low rank detection algorithm is better than others in terms of detection time and speed.