Research On Apple Online Grading System Platform Based On Machine Vision

Apple is one of the common fruits grown in large areas in China,at present,the quality inspection of apples mainly depends on manual work.However human beings can cause misjudgments,slow speeds,and high costs.To address this issue,this study uses machine vision technology to independently design and develop an online quality grading system based on machine vision,this system is of great significance for improving apple quality detection and grading.The main research content is:(1)Analyzed the research progress and methods of fruit quality testing at home and abroad,proposed the research content and technical route of this study,according to the research content,the hardware platform of the machine vision system is independently designed and built.(2)The basic principle and imaging model of camera calibration were analyzed,based on this,using grid image,using OpenCV camera calibration program to achieve CCD industrial camera calibration,completion of the CCD industrial camera internal reference matrix and its correction.Finally use the machine vision system to capture the apple image,each frame of image contains the front and two sides of the same apple.(3)Researched apple quality grading method based on external features.Firstly,the median image filtering method is used to preprocess the apple image and the three regions of apple's region of interest in a frame of image are extracted,and then divided into three independent images for separate processing,make a background segmentation of the apples in both sides of the mirror image.Then,size and fruit shape features are extracted from the apple front image information,and color and defect features are extracted from both sides of the apple image information.Finally,the extracted characteristic parameters were compared with the national apple quality evaluation criteria to generate a grading result.By testing 400 Hanfu apples,the results show that the overall accuracy rate of apple quality classification based on external features is 76.75 %.(4)Research on apple quality grading method based on convolutional neural network.Firstly,a 12-layer CNN model was designed,including one input layer,five convolution layers,three pooled layers,two fully connected layers,and one output layer.In the Caffe framework,the training parameters of the model are set,and the CNN model is trained using 1 600 training sets and 400 verification sets.Finally,the training accuracy of the model is increased to 89.75 %.In this model,400 Hanfu apples were tested.The results showed that the overall accuracy rate of apple quality grading method based on convolutional neural network reached 83.25 %.(5)According to the apple quality grading method described above,select VS2010 development environment and using OpenCV + MFC + Caffe,the development of a host computer application program based on machine vision apple online grading system.