Multi- feature fusion in weed recognition based on Dempster-Shafer´s theory

As accurate identification of weeds from crops is the prerequisite for precise herbicides spraying, this paper proposes a multi-feature fusion method based on neutral network and D-S evidential theory to improve the accuracy of weed recognition. Firstly, three kinds of single features such as color, shape and texture are extracted from the weed and crop leaves after a series of image processing. Secondly, the leaves are classified with each kind of feature by neutral network and the output of each sub-network are made as an independent evidences to construct the basic belief assignment. Finally, using D-S combination rule of evidence to achieve the decision and giving final recognition results by classification rules. The experimental results have shown that the multi-feature fusion method has good performance on accuracy compared to the single feature-based method in weed recognition.