Research On Influencing Factors Of Image Segmentation For Crop And Weed Identification

The development of information technology is important for modern society. It can improve agriculture productivity, help weeds control, reduce the labor intensity and cost, protect ecological environment, realize sustainable development in agriculture with auto identifying weeds using digital image processing. In this way, digital image processing has attracted a good deal of attention step by step. Existing identification of crops and weeds has poor adaptation because the natural environment is changing in field, and the real-time processing is not satisfying. To resolve these problems, this thesis did some basical research focused on image segmentation effect of young plant of green pepper, mature pepper and weeds from nightshade. The contents of the study could be briefly summarized as follows:1. It introduces the characteristics of reflection spectrum of plant and soil, and their physical and biological mechanism, and influencing factors such as light intensity, sunny exposure, picture taking angle and content of soil moisture that effect color image segmentation are analyzed in theory.2. Hardware of machine vision system is constructed, experiment plan of how to obtain crop and weed images in natural lighting is designed based on the experiment environment. Crop and weed images are gained in outdoor natural environment.3. According to the problem that image segmentation is effected by factors of light intensity, sunny exposure, picture taking angle and content of soil moisture, orthogonal experiment was applied, the relationship among the four factors and priority of each factors are obtained, and main influencing factors that effect the results of color image segmentation based on extra-green feature were studied.4. Carrying out further study of effecting factors that effect the results of color image segmentation based on the foregoing orthogonal experiment, and relation of segmentation threshold and the main effecting factor is achieved, further more, corresponding modified segmentation threshold model is constructed.This research improves the adaptability of crop and weed image segmentation according to the change of natural environment without time expense. It is important for using digital image processing technology of machine vision to identify crop and weed in outdoor natural environment real-time.