The Research Of Apples Detection Algorithm Based On Machine Vision

The study of apple grading on-line detection is based on machine vision, which is great significant to improve competitiveness of our country apple market. At present, the apple graduation method that exists shortcomings of low stage efficiency, high miscarriage rate, slow speed and so on doesn't meet requirements of real-time classification. Based on this, the algorithm study of apple detection based on machine vision is focused on apple's image processing algorithms and apple's size, shape, color, defect classification algorithms.In this paper, first of all, the apple's images that are collected will be changed for 24-bit GDI bitmap by A/D, the apple's images converted are processed by smoothing, segmentation, gray, sharpening and so on. The classifications of apple are realized by the size, shape, color and texture features parameters of apple images processed which are extracted. The classifications of apple size and shape which use method of diameter discriminance can be divided into four classification levels, the accuracy rate of classification is over 90%. And the classification of apple shape which uses the difference value method of reference shape extreme track can effectively divide fruits into round fruits and malformation fruits. The BP neural networks as a classifier is proposed in this paper for the classifications of apple color and defect, the color grading uses BP neural networks optimized particle swarm optimization, the characteristic parameters hue frequency of apple color that is input of BP neural networks can train BP neural networks optimized to achieve apple color grading, the grading accuracy rate is not less than 96%. The classification of apple defect uses BP neural networks optimized genetic algorithm, the characteristic parameter average gradients of apple defect that is input of BP neural networks can train BP neural networks optimized as a classifier to achieve apple defect classification, the rate of classification recognition is more than 89%. At last, the apple's classification detection system is designed, the classifications of apple size, shape, color and defect are realized by experiments.As a result, the apple detection classification system is designed in this paper, and the experiments verify that this system meets requirements of real-time, at the same time, feasibility and correctness of this paper's the theory study and design.