Research For Sense Of Citrus Grading Based On Computer Vision

The classification technology is always a great weak link for our country’s fruit industry that directly affects the performance of orange in market. Orange is some of the largest fruits output, and it’s also important in the foreign trade fruits. People are plant Satsuma mandarin, especially in the South. Tampering with the brand’s position and reducing competiveness because of the orange on the market is at present colloid measure the good and bad are intermingled by the poor classification technology is used after orange are picked. However, labor costs are now rising in our country, and it also in manual grading costs. Labor are low work efficiency in the manual grading who are influenced so easily because they tires by long works. The accuracy of the manual grading is less than guaranteed by artificial errors.This paper use a combination of technology, image processing, mode recognition, software engineering to research the method of classification technology for orange that is based on computer vision technology.A vast amount of research work have been done with the largest output of Satsuma mandarin as the object of study. Image collection machine is designed. Acquisition system is selected as it is non-real-time and under a specific environment. Depending on the ideas of relevant experts, we collected132samples of the all kinds of image.Image preprocessing methods of analysis and research samples. The image of collected are converted from one color space to others. We improve image enhancement algorithms, and take the measures of image filter, image of a smooth process to undo noise. It analyzes and extracts features. Be selected according to the relevant national standards according to Satsuma geometric shape, size, color, fruit spot, and scars, the extracted feature value of each feature normalization processing.Main Innovation Points of Thesis:1. We gets the edge of image and extracted features, remove the background. In the process of removing image background we use global thresholding method in that converted from RGB color space to HIS color space which is more suitable for human visual perception, and it makes result better than the traditional way. 2. It researches target classify methods which is based on support vector machine and perfect linear and optimal hyperplane algorithm. Resolve linearly inseparable problem by constructed variable and called Lagrange coefficient. It avoid calling kernel function according to traditional methodology. It also designed classifier for less categories and resolve inseparable problem.