Study On Recognition Method Of Fruit Surface Defects Based On Machine Vision

In recent years, machine vision technology with its high speed, large information quantity,multiple functions and other features has been widely applied in the quality detection ofagricultural products, greatly improved the accuracy and efficiency on quality of agriculturalproducts detection, to become the research focus of agricultural products’ automatic detectionand sorting technology. This paper based on machine vision technology for fruit surface defectdetection and recognition method research, by experiments on acquisitioned daily fruit images,determine the image low level processing and segmentation method, eventually extracted thefruit surface defect color, gray scale, texture and geometry and other characteristics of thevarious information from the defect picture, and identify the defects based on the BP neuralnetwork. The main contents of the paper are:1. Taking the common fruit citrus (pure fruit), Dangshan pear (spotted fruit) as the researchobject, analysis of fruit defect detection machine vision system design requirements, completionof the machine vision system hardware (light box, camera, camera lens, computer, conveyor beltand manipulator) selection and construction.2. Study on the low level image processing operation of the preprocessing method suitablefor the subject of fruit image, according to differences of the experimental environment,equipment and processing requirements. Including spatial domain image enhancement,frequency domain image enhancement, image gray scale and color transfer, morphologicaloperation, and so on.3. Study on the fruit surface defect feature extraction method. For the miscarriage of justiceon fruit edge and spots of spotted fruit, using image synthesis (i.e. to edge) and morphology(remove spots) processing, thus complete the fruit surface defect detection. Then save the defectimage, as well as the gray and two value images. Extraction of defect image characteristic valuesincluding color, gray scale, texture, geometry.4. Study of the defects classification method based on BP neural network, design of12-10-6BP network structure. The input of neural network are defect area and fruit surface arearatio, the average gray, texture feature values, the output are the six types of defects of pure fruitor spotted fruit. Through the training and testing, network can complete the defect recognition.Then realize the classification based on the defect fruit.5. Design and development of the fruit surface defect recognition in the visual systemsoftware, convenient and practical.