Research On Apple Size Measurement And Classification Based On Machine Vision

Apple sorting technology classifies apples into different classes based on specific characteristics such as apple size,shape,color,and quality.Applying machine vision technology to the sorting technology,the camera collects apple images on the conveyor in real time,image processing obtains its image feature information,calculates visual parameters such as size,color and texture,etc.Improving sorting efficiency and accuracy is of great significance.This paper uses Apple sorting technology as the starting point,and uses machine vision,image processing,texture description,and machine learning technologies or methods as media to conduct research on apple size measurement and variety classification.The main research work is as follows:According to the selection points of the optical system,the camera,lens and light source were selected,and a machine vision system based on Apple image acquisition was established.A distortion model consisting of radial distortion,tangential distortion,and thin lens distortion is described.Used Zhang Zhengyou’s calibration method to achieve distortion correction to get camera internal parameters,which are used to convert Apple pixel size to physical size.Apple edge extraction and size measurement are realized.The apple’s RGB image is grayed out,and the grayscale image is de-dried using a bilateral filter.Based on this gradient information,the edge of the apple is detected using the Canny operator.The seed filling method is used to fill the interior of the closed edge,and then the morphological opening operation removes the noise of the background image.Finally,the Sobel operator extracts the edges of the binary image.After obtaining the apple edge point coordinate information,the improved Huff transform is used to detect and the circle with the highest apple fit.Estimate the volume of the apple based on the size of the circle and the physical parameters after calibration.An apple damage detection and variety classification method based on Gabor texture feature SVM classification and Res Net neural network is proposed.Based on the RGB color components,the uniformity and contrast characteristics of the Gobar texture are extracted,and the scale parameters and orientation parameters are(3,3).Mark the samples of "Red Fuji","Marshal Huang" and "Surface-loss" apples,reasonably allocate the proportions of the training set,verification set,and test set according to the number of samples,and use the edge extraction method in Chapter 4 to remove interference.Use SVM classifier and Res Net network for classification training respectively.Figure 32;Table 4;Reference 52.