Multi-scale clustering method based on diffusion equation

The invention relates to the technical field of clustering, in particular to a multi-scale clustering method based on a diffusion equation. Randomly selecting k points in a data set which can be divided into k clusters as central points of the k clusters; dividing other points into k clusters; calculating the concentration of the whole system; selecting the maximum Euclidean distance between sample points of the data set as an initial diffusion scale, and generating a Gaussian distribution random number with the expectation of 0 and the variance of 1 by using a Gaussian random number generation function; according to the generated Gaussian distribution random number and the diffusion scale, generating a new center point coordinate according to a center point motion formula, namely performing cluster diffusion; the process is repeated until a preset circulation threshold value is reached; halving the diffusion scale until the diffusion scale is smaller than a preset threshold value; outputting an optimal solut