Research On The Species Identification Of Vegetable In Food Safety Traceability System

With the food safety problems occur frequently,many countries have established the food safety traceability system which accords with the situation of national food safety traceability system.Further,in the process of vegetable traceability system construction,there some problems about the collection of vegetable species information,such as the accuracy of the vegetable species information,the standardization of vegetable species name and the commercialization of the vegetable species identification.In order to solve these problems which exist in the establish process of vegetable species information archives by the artificial in the traceability system,presents a research on vegetable species identification algorithms.This paper used the twenty kinds of vegetables in our common life as the research object,based on the theory of machine vision technology,researching the image preprocessing,segmentation,feature extraction,classifier design of the vegetable images.The research on the vegetable images,include the image preprocessing,segmentation,feature extraction,classifier design.Research works in this paper are summarized as follows:(1)According to the research requirement,a test system of vegetable categories recognition is designed,including hardware system and software system.The hardware system is composed of lighting system and image acquisition system,and the software system is mainly composed of four modules: image acquisition,image processing and analysis,training and identification.(2)In respect to the preprocessing of vegetables image,the first step is denoising for the sample image by bilateral filter,and then by comparing the enhancement of adaptive histogram equalization algorithm and contrast-limited adaptive histogram equalization algorithm,choose the latter for its enhancement processing;when segmenting image,owing to complex background of vegetable image,we propose a new segment algorithm that is the combination of salient area detection FT and OTUS,which makes a satisfactory result.(3)In accordance with the own attributes for all kinds of vegetables,This paper has carried on the corresponding feature extraction based on color,shape and texture.In the aspect of color feature,two new color feature extraction algorithms is proposed,including color fingerprint and HS-MD color correlation method;In the aspect of shape feature,extracted eight characteristic values based on area and edge of images;In the aspect of texture feature,energy,entropy,contrast and correlation of four eigenvalues are extracted by using the algorithm which combined gray level co-occurrence matrix with local binary pattern.(4)In terms of classifier design,this article did not use the common pattern recognition algorithm,but for all characteristic values,select the different kernel function to carry on the training,and then uses the weight of each kernel to carry on the weighted fusion,finally uses the SVM to carry on the classification training,its recognition rate achieves the satisfied effect.