The Research On Cow Image Recognition System Based On Mulit-feature Fusion

Cow Recognition is an important part of the management of the cows, theprevious identification of cows artificial recognition, but the efficiency is low, and therecognition result is susceptible to human factors. With the development of scienceand technology, the electronic tag identification is also used in dairy cow recognition,but the use of high cost. Cows image by image recognition technology, automaticidentification, and other relevant information timely and accurate collection, themethod has a low error rate, low cost, fast, interfere with the advantages of the lessaffected by human factors.This paper is based on the shape of cows image invariance cows imagerecognition, double-feature information LHA (Local Hu Alogirthm,) algorithms andintegration of multiple color space-based scale-invariant feature transform SIFT (ScaleInvariant Feature Transform) algorithm, and the final two algorithms proposed dairyrecognition algorithm based on multi-feature. The LHA algorithm as a rough sentence,the SIFT algorithm as fine-contracting, the better recognition results. In this papercompleted the following work:1) Cow images pre-processing work. In order to remove the the cowsbackground and light cows image itself, cows image identification preprocessing work.This paper studies the cows image recognition effect take into account, thus usephotoshop software cows manual segmentation of the target and background image,then use the homomorphic filter to remove the effects of light on the image of cows.2) The LHA algorithm based on double eigenvalue. In this paper, the Humoments algorithm based on adding binarization processing and secondcharacteristic information, proposed LHA algorithm. The first cows image isconverted to binary image, threshold selection using the Otsu algorithm, then extractthis image7Hu moments characteristic value as the first characteristic value,followed by the number of black pixels of the extracted image as the value of thesecond feature. These two characteristic values to be saved at the same time, as thefeature template. Minimum Euclidean distance of the first eigenvalue template and recognition, first calculate the image to be recognition of the first eigenvaluecharacteristics, the results as to be identified. Then calculate the minimum Euclideandistance between the second feature value, and if it is less than the selectedthreshold, then the results as a recognition result of the first characteristic value,otherwise it would be subject to a recognition result of the second eigenvalue.3) The proposed algorithm SIFT features multiple color space. The SIFTalgorithm is applied to the cow image recognition, and experiments in multiple colorcomponents, the final choice of the RGB and LUV space as a result of the to beidentified, will eventually SIFT feature point matching the most points as arecognition result. Experimental results show that the SIFT features multiple colorspace algorithm can effectively identify cows image, the average recognition rate canreach more than90%.4) Multi-feature fusion cows recognition algorithm, the the LHA algorithm toidentify results as a result of the rough sentenced, the integration of multi-colorspace SIFT algorithm recognition results as a result of the fine-contracting, theexperimental results found that the fusion of two algorithms, has good recognitioneffect.5) Development of a complete cow identification system. Based on thewindows platform, the development of the dairy cattle identification system basedon OpenCV. The system mainly consists of the binarization processing module, theillumination of the processing module, the identification module, characterized in theprocessing module. Each module is easy to operate, portable system windowsplatform, the experiments prove that this system is easy to operate, easy to carry outthe research and application.