Research On Infield Weeds Detection Technique Based On Machine Vision

The weed is one of the important factors which affect our country agricultural product quality and output. At present using the herbicide spraying, especially extensive spraying to remove weeds. This kind spraying method not only enhances the agricultural costs, but also undermines the land quality, pollutes the environment, and does not favor the agricultural sustainable development.In response, using computer vision technology to identify weeds and to determine the location and types of weeds was studied in this paper, and it provided theoretical and technical support for accuracy of automated spraying herbicides for the crops field.Main contents are as follows:(1) Introduced the development of weed detection at home and abroad; proposed the research necessity, feasibility and methods.(2) Discussed the pretreatment methods of filtering out the image noise; compared the mean filtering method with median filtering method, the latter was chosen in this experiment.(3) Studied the method which used color characteristic of color image to segment green plants and soil background in view of weed detection in complex background. This paper mainly elaborated segmentation under the RGB and HIS color models. Considered the real time operation, the study used color space and color components, as a characterization of the image color information threshold parameters. Using automatic threshold segmentation realized the segmentation between the weed area and background area. The experimental results showed that the classified statistics the different color components obtained color space and the color characteristic components suited green plants and soil background segmentation. It was practical and effective that threshold parameters expressing images colors information divided weeds from the background area.(4) In view of the drilling crops, considered the system execution the speed, has first used the position characteristic law recognition between the lines weed; Regarding the line in the weed, according to the crops and weed's shape characteristic, the author proposed uses widely long compares, the round fullness, roundness, the rectangle, the skeleton area to compare, the skeleton perimeter to compare six zero dimension shape characteristic parameter to take the pattern recognition input eigen vector/feature vector/proper vector, establishes BP neural network. The experimental results showed that the system has good effective for weed images.The research of weed identification system had practical significance for suitable applying herbicide, enhancing automatic level of agriculture and protecting environment.