Improvement Mean Shift-based Image Segmentation Approach for Automatic Agriculture Vehicle

Mean Shift algorithm, a statistic iterative procedure, is robust when applied to farmland image segmentation. It can effectively overcome the influence of shadow, weeds or illumination changes, etc. However, the Mean Shift procedure has relatively high time complexity and can not meet the requirements of real-time processing. Based on pyramid algorithm, we can obtain a low resolution representation of the images being processed. Then, run Mean shift algorithm on a set of seed points that selected in the low resolution image. Through this method, the time consumption is significantly lower than the original Mean Shift Procedure. The objects in farmland images are large and there are only two major types of structure in it, so the examination accuracy of proposed method is changed little. At the same time based on spatial structure and color distribution of farmland image, Mean Shift Kernel radius in the spatial and range domain is selected. In addition, according to different seasons, crops show different colors. In this case, the equations which convert color image into a grayscale image are discussed.