The application of support vector machine in veed classification

As accurate weed identification is the for precise herbicides spraying, this presents a weed recognition method based on support vector machine (SVM). At first, five kinds of weeds are segmented from the background images and their shape and texture parameters are extracted. Then, according to the distribution of the feature data, the most effective combination of feature data are selected and inputted into SVM classifier for classification training. As SVM has advantages of high-dimensional and nonlinear processing capabilities, in this paper, the effective character parameters are selected by analyzing the distribution of feature data, which reduced the complexity of the algorithm. At the same time, the reliability of classification are ensured by the cross-validation classification and training. The experimental results show that the accuracy of weed recognition in proposed method is 93.3% and the classification time is 1.18s. This is an effective classification method and will find wide application in order aspects.