A Study On Vegetables Recognition Algorithm

Food safety is the basic condition and guarantee to maintain human survival, growth and development. Traceability system is the Quality Assurance System to store all relevant information in the process of supply of the product, which is an effective method to achieve food quality and safety management.Some relevant laws have proposed the establishment of food traceability requirements, and develop a number of relevant standards and guidelines. Some provinces and cities set up a food traceability information system and network switching platform. Traceability scale is the entire system of data collection terminals, it can achieve food information input, transmission and other functions. However, compared to the ordinary scale, traceability scale operations is cumbersome, and businessmen can not be effectively used.Thus, the manager can’t get the real data, and it is difficult to trace the food information.In this paper, because of the actual demand, we focus on vegetables recognition algorithm, that is the automatic identification of vegetable species, in order to improve the traceability of the scale function. The main works are as following:First, we sort by using crushed leafy vegetables and non-shredded leafy vegetables. In Spring and Autumn conditions at 5:00 pm outdoor illumination changes, we build an image library, which includes 34 kinds of vegetables,40 per kind and a total of 1360 images of vegetables.Second, image segmentation. Comparison of threshold segmentation algorithm, edge detection segmentation algorithm, and using K-means clustering segmentation algorithm make an segmentation on the vegetables images in RGB, Lab, HSV color space. The result shows it gets a better result in the HSV color space.Third, feature extraction. The vegetable has some features, like color, shape and texture. This article describes the vegetable by GCH and CLBP to find their differences.The result shows the GCH and the CLBP can express vegetables’difference, which is the basement of pattern recognition.Fourth, vegetable recognition. Focusing on KNN, LDA and SVM classifiers, we design a two-level classify, the first level uses color, and the second uses texture or shape, to improve the rate of vegetable recognition. The result shows using SVM+GCH as the first level, and SVM+CLBP as the second level will get the best rate to 97.94%.As is described above, the study proposed an effective vegetable recognition algorithm. It can be used to improve the traceability of scale, which simplify the use of traceable scales and improve the credibility of traceability system. By this method, it provides protection for food safety traceability system.