Research On Identification Of Weed Images Based On Computer Vision

In traditional agriculture, the most common irrationalities are that crops suffer from weeds that grow rampantly and weeds disease. Without the knowledge of the distribution of crops and weeds, massive use of chemicals will not only result in waste of chemicals, but also cause environment pollution and health hazards. Therefore, it is important to identify the weeds against the background consisting of crops and soil, and to selectively spot-spray chemicals in calculated amounts.In this paper, computer vision and pattern identification technique are used to study the method to detect the weeds in wheat field.1. In order to solve the problem of illumination infection and adaptability of environment in weed identification at present, a segment algorithm based on color feature is presented in this paper. The algorithm separates the weed area from soil background according to the color eigenvalue, which is obtained by analyzing the color difference between the weed and soil in three color models RGB, rgb and HIS by using AOI (area of interest). The result shows that it can get very good result in segmentation by using color characteristic identification.2. According the appearance difference between the wheat and weeds, after pretreatment, dilation and erosion of morphology method and seed fills of computer figure subject are introduced in, making use of to set the fit structure element to separate the weeds area and wheat area, and statistics the density of the weeds in the images, which is the leading of spraying.3. On the basis of the texture difference between five familiar kinds of weeds in wheat cropland and the wheat images, the method of the gray matrix which based on the statistics characteristic is proposed in this paper, pick-up five eigenvalue of HIS gray matrix: energya contrast correlation energy homogeneity, which used to identification the weeds images.And adopt improved neighbors classification to carry on identifying experiment,the results show that it can obtain a good classification identify result.