Wheat Varieties Identification Research Based On Sparse Representation

Wheat is the main crop and has many varieties.Different varieties of wheat have big differences between quality, and directly affect the subsequent processing quality of wheat.So researching the wheat varieties of categories has very important significance.Applying computer vision technology in wheat varieties classification has advantages of fast,nondestructive and high recognition rate and so on.The sparse representation method is used for image feature extraction,in order to get better recognition result, need a lot of images for training, in turn, led to a large amount of calculation, classification efficiency is greatly reduced.Based on conventional sparse representation, on the basis of theoretical research, introduction of dictionary learning improved sparse representation method, and applied to wheat varieties of image recognition,finally the effectiveness of the proposed method is verified by MATLAB.This article selects four kinds of wheat varieties in henan(zhengmai 103, kaimai 21,zhoumai 20 and yubao1) to go on classification research.Firstly,extract the characteristics of the color of the wheat grain(R, G, B, H, I, S), morphology parameters(perimeter, area, circle,rectangle degrees, stretched length) and the texture feature parameters(energy, entropy,contrast, correlation) to go on normalized processing, build the dictionary for sparse representation method.Secondly, puts forward the improved sparse representation method based on dictionary to learn.The method using the K-SVD algorithm dictionary learning,under the premise that don’t change recognition rate, a small dictionary that has less atomic number is obtained by training can effectively represent the original dictionary, obtain better dictionary library of the power of expression, and then classified according to the residual.Finally,using sparse representation method, sparse representation method based on dictionary learning and neural network method is used to identify the classification of the above four kinds of wheat varieties.Results show that when the size of the dictionary is the same, the sparse representation method based on dictionary learning image classification efficiency increased significantly, with recognition time of 81.02 seconds, reduced to 21.77 seconds.When the recognition rate is the same or close to, sparse representation method based on dictionary learning needs of atomic number is far less than the regular number of dictionary atoms needed for the sparse representation method.Sparse representation method is compared with the neural network method not only recognition rate is higher, and the recognition time is shorter.