Research On Identification Of Infield Crops/Intra-Row Weeds Based On Weeding Robot

With the rampantly growth of weeds and the disease of weeds, it takes severe influences on crops'growth. For present, spraying herbicide has always been the major treatment to weed, but massive use of herbicide will not only result in a wasteful use of resources, but also cause environment pollution. Aimed at this case, to adapt to the request and need of precision agriculture, identification of infield weeds, as a new research area, has been researched and discussed in this article.With the development of computer technology and information processing technology, great importance has attached to the research of automatic navigation of self-propelled agricultural machine,machine vision and agricultural robot. This paper taking corn 2-3 leaves as the research objects, investigates and proposes a infield weeds identification software program which applies to weeding robot or intelligent sprayer, the main research contents are as follows:Many complex background of corn pictures were taken by image collect equipment. Noise was produced during this process, so the original RGB images were filtered before image processing. By the comparative tests with the domain mean filter and the vector median filter, domain mean filter was chosen for its better features; Using different kinds of feather factors as image grayness methods deal with those pictures in RGB,HSI and rgb color space, then confirm the optimum feather factor; Study on Image Binarization Algorithm, after comparing several kinds of methods, an appropriate method suits for our experiment is selected. Salt noises appears in the binary images, this paper disposes it with median filter or mathematical morphological method; This paper mainly embodies the innovation points in:To identify the intra-row weeds, position information is first proposed to analyze regular rows of corn. Pixels in binary image are gathered as rows, then we analyze and process the statistics of pixels, at last, weeds are identified. Aiming at complicated situations such as with single corn missed or leaves overlapped, different from ordinary methods, this article presents a process scheme of changing images to signal and figuring out weeds with the algorithm of cross-correlation.Based on related basic knowledge such as digital image processing technology and digital signal processing technology, this paper analyses and processes infield corn images, and has a better performance for identifying intra-row weeds, which offers theory and technology support for the further design of intelligent weeds identification system.