Study On Maize Weed Recognition System Based On DaVinci Video Processing Technology

The growing environment of land crops in Northeast China is good,but the overgrowth of weeds also causes insufficient nutrition for crop growth,resulting in the loss of crop fruits and the decline of crop yield.Traditional chemical weeding can effectively control the rapid propagation of weeds,but extensive spraying operations bring about serious pesticide residues,soil and water pollution and poisoning of the applicators.With the development of science and technology,weed identification has become a hot research topic,but its real-time and accuracy are not optimistic.Under the background of the requirement of green,accurate and efficient agricultural modernization,this paper proposes a weed identification research system based on DaVinci video real-time processing technology.Through DaVinci technology,the efficiency and accuracy of weed identification can be realized,and the two requirements of faster and more accurate weed identification can be achieved,which provides a theoretical basis for modern weeding equipment.This research mainly includes the following four parts: Firstly,collecting field seedling and weed video,transforming the video into image,extracting a frame of image per second to ensure the continuity of the image,and preparing for the later accurate recognition.Secondly,we collect weed images as standard samples of image processing,and preprocess them,including grayscale,binarization and morphological filtering,in order to extract accurate feature parameters,find the most suitable feature values for weeds,and improve the recognition accuracy.Thirdly,through edge extraction,several characteristic values of weed sample images are obtained,which are all in pixels.Through comparison,the rectangular degree and the first invariant moment are the most suitable dimensionless feature parameters,which can accurately determine weed characteristics.Finally,the weed characteristic parameters are used to validate the image extracted from the field seedling and weed video.The conclusion is that the recognition accuracy is 95.27%,and the average processing speed of each picture is 1.082 seconds per frame.