Improved Density Peak Clustering Algorithm And Its Application In Purple Soil Image Segmentation

The recognition of purple soil by machine vision technology is of great significance.The background of a color image of purple soil may be very complex,often containing impurities such as green vegetation and surface soil,which will seriously interfere with the identification of purple soil.Therefore,the background interference can be avoided by extracting the object-soil region from the soil image.The main work for the extraction of purple soil region is as follows.(1)According to the better aggregation of the soil region in the color image of purple soil in a component in the Lab color space,the density peak clustering algorithm is improved,and the improved is applied to the initial segmentation of the color image of purple soil.Firstly,the density and distance of each gray value are defined by the frequency and gray value of each pixel in a component.Then,according to the principle that the density of clustering center is large and the distance between each other is long,we establish the optimization model,getting the clustering centers by the maximum inter-class variance.Finally,we make the image of purple soil classified according to the determined clustering center,and the closest one to the foreground region is selected to complete the initial segmentation of the purple soil color image.(2)Aiming at the problems of a large number of isolated points,scattered clods and voids in the initial segmentation results,we conducts post-processing operations on the initial segmentation results.The image post-processing algorithm consists of two steps.The first step is boundary extraction of soil region,and the boundary extraction of soil region is realized by searching for the connected boundary with the maximum number of boundary points in the initial segmentation image.And the second step is region filling of soil image.The filling of the soil image area starts from the center of the image,and then we traverses the elements of the top,right,bottom and left positions of the point.We then iterates over the top,right,bottom,and left of the already accessed elements until all points in the boundary are accessed.Finally,the result of the region filling and the original image are made hadamar product to complete the final segmentation of the purple soil color image.(3)An API of purple soil color image segmentation algorithm is developed based on improved peak density clustering.By analyzing the functions of each part of the image segmentation algorithm of purple color soil,every function and class is partitioned.And every function is implemented and tested.Finally,the algorithm is encapsulated into an algorithm API which is a lib file.