Research On Method Of Measuring Apple Bodybased On Machine Vision

Real-time measurement of apple body is valuable for evaluating the apple’s quality and apple grading.Aiming at the existing problems of apple orchards in our country,it is proposed to solve the problem of apple non-contact estimation of apple body to guide production.In this paper,smart phones are used as image acquisition tools,machine vision and small size calibration plates are used to explore the measurement method of apple body parameters(length,width,mass and volume).The technology of apple image correction,apple body detection and establishment of estimation model in the measurement process were studied,and body parameter measurement system based on mobile terminal was established.The main research are as follows:(1)Image acquisition and preprocessing.Apple images are collected by using mobile phones.Each apple takes two side-view images.In the preprocessing stage of the image,according to the structural characteristics of the smartphone camera,the calibration model are used to correct the acquired image.A segmentation method based on local background region is used to obtain the ROI region of the image.(2)Extraction of body parameters.In the stage of apple shape feature extraction,except length,width,projected area and apple shape,apple projection mesh parameters were extracted.The mesh parameter mentioned in this paper is to draw 6 equally divided virtual grids on the ROI region of the image.The length of these virtual grids is called grid parameters.A small size checkerboard calibration board is used to calculate the scale between the actual value and the pixel value in the image collected by the smart phone,so as to calculate the length of the appearance feature in the three-dimensional world coordinates,which makes the result more absolute.The R~2 values of the length and width regression results were 0.918 and0.958,The RMSE values were 0.218cm and 0.202cm.And in accordance with the apple shape characteristics(length,width,projection area,the grid parameters),combined with the actual measurement for the apple quality,volume,using PCA to shape dimensionality reduction characteristic parameters,and compare multiple model,finally using neural network to establish estimate model of apple quality and volume,volume and quality estimation error are 5.620%and 2.440%,respectively.(3)The construction of Interactive system.Combined with the Apple body measure model,Django was used to build the website,and the development of the front end and back end was finally completed.The front end realized image uploading and display,result display and other functions,while the back end realized image preprocessing,ROI segmentation,apple body measure calculation and information storage and other functions.In addition,field test of the model showed that the absolute error between the calculated value of the length,width,mass and volume of the body parameters and the measured value was less than 2.870%,2.845%,5.614%and 2.438%,respectively.The average time from uploading the image to giving the body size measurement results was 15.324s.The experimental results showed that the model has high accuracy and good stability,and played a positive role in improving the efficiency of apple sorting and reducing the labor cost in the field.It provides a new tool and method for the farmers to record,collect and analyze the fruit growth process.