Research Improved SURF Algorithm And Neural Network Identification Of Fruits

Computer vision has been widely studied for the identification of fruits in recent years, but less appli cation used in superm arkets and farmers market of agricultural products on sale. W ith the developm ent of people's life and economy, farmers markets and supermarkets sell a variety of fruits and vegetables are m ore and more, and its shape, texture and size are not the sa me. One of the aim s at this study is to extract feature points o f how fast the fru its and vegetables and how to correctly identify the classification of furits and vegetables. According to the above questions, this paper adopts kinds of furits for quick and automatic recognition of experimental research: The first image acquisition and the establishment of the database, then the pretreatment, and then the im iproved SURF algorithm for feature extraction, finally using the BP neural network classification. Exprimental results show that the method for a wide variety of fruits and vegatables can classify effectively, its recognition rate is 96%.This paper uses kinds of fruits as the research object model.This collection of a variety of fruits image, establish a database of fruits.In the im age feature extraction, there are many methods, such as tem plate method and geom etric method tec.But in th is article w e have used the shape parameters of fruits as the featu re extraction, instead of using the im poved SURF algorithm to extract different features of fruits and vegetables.The experimetal results show that the fruits placed in any, different light noise and o ther complex environment, the improved SURF algorithm to improve the feature point detection and matching time, higher recog nition rate of fruits.Classifier selection and application, this paper based on previous papers and liter ature researcher reference without the use of very ear ly pattern recognition m ethod, instead of using the BP neual network classificatioin for fruits.BP neural network is a widely used network model, which is com posed of input layer, hidden layer and output layer.In order to improve the convergence speed of BP neural network, the weights and thresholds of the regulation, thus greatly improve the recognition rate of fruit images.