Research On Feature Analysis Of Fruit And Vegetable Automatic Categorization Technology Based On Computer Vision

Fruit and vegetable supermarket sales have become the main form in China, a big fruit and vegetable production country. But in the sale process, the categorization of fruit and vegetable namely the price still mainly relies on artificial memory or query, which consumes a lot of manpower, material resources as well as financial resources. In order to solve the existing problems, this paper presents the feature analysis of automatic fruit and vegetable categorization technology based on computer vision, and has carried out the in-depth study of fruit and vegetable categorization system. Besides, it analyzes the development status at home and abroad along with technical problems. This paper puts forward different categorization systems and algorithm frameworks according to the single and multi body, and makes a deep research on preprocess algorithm, the improvement of the speed of Gabor feature extraction and dimension reduction, the optimization of multi-feature and harmonic adaptive fusion. Finally, multi body fruit and vegetable categorization algorithm is improved and implemented on ARM.The main work and results of this thesis are as follows:(1) the establishment of fruit and vegetable image database. In order to make the database close to the actual situation, almost all of real life situations are taken into account when shooting image.(2) the study of the preprocess method. Preprocess methods are improved respectively according to the single and multi body, including denoising, segmentation, filling, normalization. Image database is established finally, preparing for feature extraction.(3) the in-depth study of feature extraction method of single fruit image and improvement of its speed, dimension reduction and its improvement as well as the classifier. Based on the statistical overall characteristics, this thesis puts forward a method based on Gabor energy mean and variance feature. It first preprocesses the image, getting the normalized image (32x32). Then, feature is extracted for SVM categorization by five scales eight directions Gabor filtering and calculating mean and variance of its energy.(4) the in-depth research on multi-feature extraction method of multi body fruit image. Considering extraction of multi-feature, feature fusion and distance function improvement, this thesis presents a method based on multi-feature extraction including GCH_HSV, LBP and BIC from three aspects of color, texture and shape along with harmonic adaptive feature fusion. Experiments are carried out to verify the correctness and relevance of the algorithm.(5)the hardware implementation of algorithm for categorization of multi body image and its improvement. Through the code transplant, cross compiling, the algorithm is implemented on the hardware. The hardware and software platforms of the research system are builded respectively.By programming, debugging and simulation, algorithm is well implemented. The results on the ARM board are similar to the MATLAB simulation results, but there still exists the gap of the actual application.In spite of this, it puts forward an effective implementation of the practical application of automatic categorization for fruit and vegetable.