Research And Application Of The Beta Drugs Residue Detection System Based On Machine Vision

In recent years, along with the widely application of colloidal goldimmunochromatographic assay (colloidal-gold strip) in food, medicines, water,health, and the increasing high demand of all types of drug residue rapid quantitativedetection, traditional detection methods have high detection precision, but thedetection equipment is expensive, time-consuming, high technical requirementsshortcomings, while it can’t be applied in the rapid quantitative detection field, so ithas an important significance to realize the rapid drug residues quantitative detectionsystem based on the colloidal-gold strip. In order to provide an important basis for theapplication of the quantitative detection system, in this paper we make an intensivestudy of the module design of this detection system, the image processing of thecolloidal-gold strip, and found the concentration predict model, this article is mainlyelaborated from the following several aspects:1、First of all, according to the requirements of on-site rapid quantitativedetection we designed a detection system based on computer vision technology.Completed the test strip image acquisition device structure design based on machinevision and get the test strip image.2、According to the characteristics of the test card image itself, we presented acolloidal-gold strip image detection based on image edge detection and gray integralprojection. This method avoids the influence of the maximum possible test line cardshell image of a lighter color feature value, and compared with the traditionaldetection methods are more accurate.3、According to the characters of the colloidal-gold strip image, the gray-scaleimage integral projection curve is used for the image segmentation, but the curve cannot be used directly, it need to go through the median processing, facilitate for thenext phase of the test strip image characteristic value extraction.4、The neural network has the advantage of nonlinear approximation, and canapproximate almost any type of non-linear continuous function. This paper utilizedleast squares support vector machine (LSSVM) algorithm’s advantage of nonlinearapproximation to detect the sample concentration of ractopamine, this algorithm was improved by genetic algorithm and simulated annealing algorithm. The results showthat the detection accuracy of the algorithm significantly higher than the standardLSSVM and BP neural network prediction model.After the pilot test of the system to achieve a low-cost, simple, on-site rapiddetection of ractopamine requirements, the system user interface is clear and simple,small error and high reliability.