Weed recognition by color image processing

Water pollution has more and more bad effects on the health of human beings. The research developed in this paper is concerned with the reduction of chemical herbicides which will be only sprayed where there are weeds. We take into account the texture and the color during the image processing in order to classify the pixels into 2 classes : the soil and the vegetation. An EM algorithm is performed for doing this. Then, the recognition phase consists in recognizing the weed leaves from the cereal leaves using a local feature. Finally, the weed density is very easily computed. The proposed method also works on complex images where the leaves are overlapping each other.