Wheat Monitoring Image Segmentation Technology Using Variety Of Color Spaces

In agricultural,vegetation ground coverage is directly related to many ecological processes:such as crop growth, soil water balance, photosynthesis of canopy and circulation of materials and so on. Therefore, how to quickly and effectively segment green vegetation from agricultural images(especially field monitoring images) is significantly important for automatic crop monitoring & management. As is known to all, wheat is a very important grain crop in China. There is obvious difference between the colour of wheat plants and background soil, selecting the appropriate color space can be helpful to effective segmentation. At present, although there are a variety of color space(such as RGB, HSV, YCb Cr, L*a*b*) for color image processing, as a result, each color space has its own advantages and disadvantages, and there is no unique color model that can be applied to all color image processing. Hence, the choice of an optimal color space is a problem of the present stage of color image segmentation. Based on massive number of wheat monitoring images from November 2013 to March 2014, this study goes through a preliminary comparative analysis on many common color models, for instance, the single color channel image: H component of HSV, Cr component of YCb Cr, a* component of L*a*b*, G-R of RGB. Different values of each component are selected for segmentation based on thresholding, and results are statistically analyzed: for each color component segmentation average accuracy and standard deviation. The final result shows: a* component of L*a*b* color space model is the most accurate and stable method: 34 different points of the crop monitoring image, image segmentation, the average accuracy achieves 90%, standard deviation of accuracy is smaller than 0.08. The conclusion of the study will provide an important basis for the research on color image segmentation,more importantly, to provide technical support in remote wheat monitoring(crop growth identification) and farmland remote information management applications.