Research On Improved Adaptive Clustering Method Based On Density Peak

With the continuous development of information technology,the massive data generated by it has become a key resource,and it is very critical to find useful and implicit content from the massive data.Clustering can discover the internal relations of data without classification marks.The principle of Clustering by fast search and find of density peaks is easy to understand,easy to operate,and has high clustering accuracy.However,the DPC algorithm still has certain limitations.Therefore,two improved methods of DPC algorithm were proposed in this paper,and the corresponding clustering effect is verified.Firstly,the clustering principle of DPC algorithm and its advantages and disadvantages were introduced in detail.The local density value of the clustering center defined by the DPC algorithm is larger and away from the samples with higher density than that point.However,the important parameter cutoff distance of the algorithm needs to be set by the user according to the empirical value,and the clustering center needs to be manually selected,which brings great instability to the clustering result.Secondly,an adaptive density peak clustering algorithm based on K-nearest neighbor(KNN-DPC)was proposed.According to the algorithm principle of KNN,the calculation method for density and similarity with higher density points was redefined.At the same time,the clustering center was selected adaptively based on the idea that the product of the local density of the clustering center and the distance to the higher density point is larger than other samples.Thirdly,an adaptive density peak clustering algorithm based on self-adjusting step size fruit-fly optimization(SFO-DPC)was proposed.The SFO algorithm was used to calculate the cut-off distance,an important parameter of the DPC algorithm.The S-FOA algorithm effectively improves the ability of the fruit-fly optimization algorithm(FOA)to jump out of the local optimum and accelerates the computational efficiency of the FOA algorithm.The SFO algorithm was used in DPC parameter calculation to effectively solve the problem of DPC parameter setting and improve accuracy.Finally,the 5 artificial and 5 real standard data sets commonly used in the evaluation of clustering algorithms were selected for comparative experimental testing of KNN-DPC and SFO-DPC.A 3D asphalt pavement crack detection system based on SFO-DPC was established,and the SFO-DPC algorithm was used to perform cluster analysis on the three-dimensional data of the asphalt pavement collected by the Gocator3100 intelligent scanner.The experimental results show that the KNN-DPC and SFO-DPC proposed in this paper improve the efficiency and accuracy of the algorithm while solving the defects of DPC algorithm parameter setting and determining the center point.This also proves that the SFODPC crack recognition system proposed in this paper can detect cracks in asphalt pavement accurately and efficiently.