The Image Segmentation Of Purple Soil Based On H Color Component Distribution

The purple soil identification based on the machine vision is helpful for the scientists and farmers to distinguish the soil type,which has important significance for the agricultural production.Because the purple soil images for identification are collected in the field,it is inevitable that some of the plants,lichens,moss,weeds and something else in the background would impair the precision of the identification.Therefore,the image segmentation of the purple soil under the effect of cluttered background is one of the fundamental problems for further analysis of the purple soil image.In this thesis,we focus on the distribution of H component in the purple soil image,and aim to extract the area of purple soil for identification effectively in the real world.The main contents and contributions are as follows:(1)We analyze the aggregation characteristics of the purple soil in the HSV color space.The distribution of purple soil in each channel of HSV is compared by experiments.Experimental results demonstrate that the H component of the purple soil is aggregated obviously,and the S component and V component is not.(2)We have carried out some research into the normal distribution characteristic of the H domain in the purple soil image,and we apply the threshold of the H domain to extract the purple soil region.The algorithms of recognizing and extracting boundary and filling blank hole are designed to optimization the preliminary segmentation results.In the boundary recognition algorithm,a new template of recognizing boundary is designed in this thesis,which can effectively identify boundaries and eliminate outliers.The boundary extraction and region filling algorithm are realized by stack technology,and it improves the filling effect of the algorithm.(3)The distribution of H component in the purple soil image is further studied with generalized Gauss distribution,mixed Gauss distribution and mixed generalized Gauss distribution.According to the study of the generalized Gauss distribution fitting of H component in purple soil,we apply the nonlinear function which is convenient for derivation to fit the complex function of the shape parameter ?.Experimental results demonstrate that we can solve the ? parameters more effective.In addition,we present a more accurate H threshold segmentation algorithm based on the mixed generalized Gauss distribution.At last,we incorporate the piecewise distribution and the piecewise H threshold to overcome the 0 point in the H distribution.(4)In order to verify the accuracy and robustness of our algorithms,we make some experiment to compare with the traditional segmentation algorithm based on extreme value segmentation and the traditional hole elimination algorithm based on dilation and erosion.Experimental results demonstrate that the segmentation algorithm of H threshold based on the approximate normal distribution is more accurate and more robust than the threshold segmentation algorithm based on extreme value.The improved segmentation algorithm based on the mixed generalized Gauss distribution is more robust than the algorithm based on the approximate normal distribution.The boundary recognition and extraction,and the algorithms of filling hole in this thesis are more effective than the algorithm of filling hole based on dilation and erosion.