Forage Recognition Based Wavelet Transformation And Improved Local Binary Pattern

The grasslands of our country with an area of4million square kilometres, has the richest grassland types and the grass species resources in the world. But the grassland ecosystem has seriously degraded owing to unreasonable utilization for a long time and the influence of climate or other factors. Digital grass industry technology is now technical support for effective management of grassland resources, preventing grasslands further degradation and sustainable development of grass industry in our country.And at the present, digital grassland at a lower level, especially grass classification and identification work need to be done manually which has the disadvantages such as low efficiency, subjectivity and low identification precision. The forage automatic identification method is carefully studied in this thesis.It may improve the efficiency of forage identification and recognition accuracy and enrich the grass digital research framework.Identification of plant leaves is the most effective and simple way to identify plants from the perspective of plant classification.Leaves contain rich and stable texture information that does not change with leaves color. Therefore, the focus of this research is to extract the leaf texture information, and try to combine wavelet technology with LBP to extract the texture feature of grass leaves. Nearest neighbor classifier and ant colony algorithm are used to be classifier to match feature to realize the grass recognition based on the leaves. Specific content as follows:(1) The images collecting stage.The camera of samsung mobile phone GT-I8150is used to get grass leaf images, which makes access more convenient and expands the application range of pasture recognition.(2)The stage of image pre-processing. The leaf images are preprocessed traditionally such as image gray processing, image segmentation, eliminating noise. By contrast, The collected images are preprocessed through the wavelet transformation. The low-frequency sub-picture obtained by dbl wavelet transform can be extracted the effective texture information, reduce the recognition time and improve the recognition accuracy.(3) Feature extraction phase. Local binary pattern (LBP) is one of the more heated texture feature extraction algorithm. In order to improve the representation precision of LBP, bilinear interpolation of LBP improves upon three interpolation of LBP. It can make the feature vector data more evenly distribute and increase the distance between the classes. Finally the recognition rate is improved.(4)Feature matching phase. Besides common method of nearest neighbor classifier, the paper also uses the ant colony algorithm to classify. It can achieve a better recognition effect.