The paper presents a study of the current state of water pollution detection systems. A formalization of the centroid map for a three-channel aerial photograph is proposed. An example of using the Kmeans algorithm for clustering terrain and water areas on test aerial photographs is considered. The visualization of the results of clustering of aerial photographs for a different number of centroids is given as well as the results of pollution segmentation. A block diagram of the clustering algorithm is presented. Its advantages and disadvantages are identified. The structure of the developed software using Python and cross-platform computer graphics libraries is described. An assessment of the accuracy of using the clustering algorithm using the F1-measure is performed. Preliminary experimental studies showed that the inclusion of an expert in the contour of decision-making allows increasing the flexibility of the program, due to the possibility of selecting a target area, choosing the number of clusters and segmentation accuracy.

K-means algorithm, clustering, image segmentation, pollution detection, F1-measure

Khasanov M.B.,  ,  HSE University

Diane S.A.K.,  ,  orcid.org/0000-0002-8690-6422,  ICS RAS