Research On Classification Of Fruit Surface Level Based On Machine Vision

Binocular vision is an essential component of machine vision,three-dimensional coordinates of a point in the world coordinate system can be calculated by the optical parallax of two images which come from the cooperation of two cameras,binocular vision is a simple and reliable way to simulate the spatial information of human's eyes,and it has extremely strong application value in fruit quality automatic identification of fruit quality.Different from the traditional sorting method of mechanical recognition and artificial sorting,the working mode of machine vision is more active,not only is standard of quality judgment objective and concrete,but also executability is strong,and the non contact way makes it widely used.Based on the reference of a great deal of related literature and documents,it was found that the previous research on identification and detection of fruit surface quality mainly relied on two-dimensional feature information,while there was rarely involved in fruit height information extraction;furthermore,research on surface defects of fruits showed that the morphological characteristics of the defects were less studied.The main feature was to calculate the overall characteristics of the entire sample image,so as to identify whether the sample had defects.It took the common fruit,apples and pears as research object,a fruit defect recognition system of support vector machine based on different kernel functions and automatic recognition system of flawless fruit surface quality based on particle swarm optimization BP neural network were established.The thesis carried out the following sections of work:(1)The hardware requirements of machine vision system were automatically identified by analyzing the surface quality of fruit,the selection of camera,the production of calibration board and background board,and the arrangement of light source were completed.The image processing system was replaced by MATLAB 2016 a software and a notebook computer.(2)Based on the principle of binocular stereo vision acquisition of disparity maps and three-dimensional coordinates,the calibration of single and double-headed camera was investigated,and the calibration results were analyzed.(3)The pre-processing method for the selected topic was explored.By comparing the advantages and disadvantages of various methods,the image edge extraction and image segmentation were realized.(4)The common color features,texture features and morphological features were screened,the advantages and disadvantages of different stereo matching methods were analyzed,the image reconstruction based on linear stereo matching was realized with the calibrated parameters,and the height features of samples were obtained by the histogram.(5)A support vector machine defect classification model suitable for small sample classification was established through segmenting the defective binary image.The color features,texture features and morphological features of fragrant fruits including height features were extracted,a 12-dimensional vector that retained 95% of the features after principal component analysis was selected as input,and a BP neural network based on particle swarm optimization was established.The recognition of fruit surface quality and comparison of recognition rates and working methods of different BP improved algorithms.