Research On Inspection Technology Of Vegetable’s Fresh Levels Based On Computer Vision

Currently, many countries have using computer vision techniques on testing agricultural product quality to achieve the crop identification, detection and management. As a new non-destructive testing technique, computer vision is an inevitable trend to achieve automatic visual inspection of agricultural products.There has much research on the quality of primary products. We select three grades leafy vegetables and research the identification of leafy vegetables by computer vision. Based on computer vision and pattern recognition theory, we obtain a white background vegetable leaf image. At first, the image has been processed and analyzed by MATLAB. Then, using Fisher discriminant method combined with principal component analysis to achieve feature extraction and discriminant model building.13feature parameters will be integrated into four new variables by PCA. In the research using Fischer discriminant recognizes and grades the leafy vegetables. At last the sample’s recognition accuracy rate is84%.Firstly, a computer vision system has been developed to detect the leafy vegetables’ surface feature and to grade the leafy vegetables.Secondly, algorithms of image pretreatment have been studied. During the background segmentation, the images have been extracted in R, G, and B channels by adaptive median filtering and OSTU. In the research we found B channel is the best. Contrasted with the selected threshold binary image and the original color image, it has been achieved to extract the background color image.Thirdly, feature extraction:color features, shape features, texture features. Color parameters have been denoted by HSI color model and RGB color model. After image processed, it gets the edge image and the target area, in order to obtain the shape features of the leafy vegetable. Different degrees of vegetables have different texture features. In this paper, we use GLCM to get leafy vegetables’texture features.Fourthly, establish model. After feature extraction, each sample consists of13features. Principal component analysis makes the data dimensionality reduced. Then build the discriminant function with low-dimensional parameters. Test analysis shows that the recognition accuracy rate of the discriminant function for the sample is84%.Finally, a leafy vegetable grading system for surface features has been designed with MATLAB and.NET. The software has friendly interface, which can finish background segment and smooth of the image and extract leafy vegetable and figure out the feature parameters.