Research On Fruit Segmentation Of The Same Color System Based On Optimized Cluster Analysis

Whether in the orchard production test or the picking robot operation,the target recognition of green apples of the same color has become a new challenge.Because the target fruit has a color representation similar to that of the background,and the effects of complex lighting changes and occlusion of branches and leaves in the orchard,the existing same-color fruit segmentation schemes cannot meet the needs of fast and accurate identification tasks.In order to better realize the recognition of green apples,this paper proposes a method for segmenting Same-color target fruits that optimizes cluster analysis.The new algorithm calculates the local density of green apple image data points and the distance from high-density points based on the definition of clustering centers of peak density.The double-sort rule is used to automatically find the cluster center;the local density of the data points is obtained by kernel density estimation,and the superpixel region representation of the image is obtained by the SLIC algorithm,which greatly reduces the amount of calculation.The main research contents and conclusions are as follows:(1)The SLIC algorithm is used to process high-resolution green apple images.The approximate pixels in a small area are assembled to form a superpixel area.The basic calculation unit of the vision algorithm is changed from a pixel to a superpixel area,which reduces the data complexity.The contour coincidence index was used to evaluate the degree of retention of real target contour information by the superpixel boundary.The average was 85.49%.The SLIC algorithm was used to simplify the image data while avoiding the damage to the true contour of the green apple.(2)Study the effective characteristics of the green apple target,analyze the performance characteristics of the target and the background under the R,G,and B color channels,study the data distribution differences between the green apple and the background foliage under the RB channel and the GB channel,and construct an RB region based on superpixels.The twodimensional feature components of the mean and GB region mean are used to establish a green apple color feature space for cluster analysis.Compared with the traditional green apple feature space and clustering and segmentation feature space data distribution and segmentation effect,the green apple color feature space proposed in this paper can achieve good segmentation results and more specific and clear data structure description.(3)Aiming at clustering optimization research of green apple images of the same color system,analyzing the data distribution law of the feature space of green apple images,clustering segmentation using density peak clustering algorithm,analyzing local and local principle characteristics,and applying kernel density estimation instead of traditional cut-off How local density is calculated,For clustering analysis and optimization of green apple images of the same color,analyze the data distribution law of the feature space of green apple images,use the density peak clustering algorithm to perform clustering segmentation,analyze the principle characteristics of local density,and apply kernel density estimation instead of cutoff distance Calculations,Through experiments with traditional local density calculation methods and calculation methods based on Gaussian kernel functions,kernel density estimation is effective in determining the local density,ensuring the clear and accurate expression of local density in different complex scenarios.The object characteristics of the green apple image were analyzed,and the clustering centers were automatically captured using a double sorting rule.The experimental analysis of the cluster distribution of the real clustering results proves the feasibility of the dual sorting rule.Through experimental analysis and verification,the segmentation efficiency,segmentation validity,false positive,and false negatives of the proposed algorithm on 86 orchard high-resolution green apple images are 87.69%,88.53%,2.32%,and 14.66%,respectively.The same-color green apple target can be accurately and clearly segmented under complex lighting conditions and occlusion conditions.And explored the application of the same color fruit and vegetable segmentation.Lay the foundation for the subsequent target recognition of fruits and vegetables of the same color.