Algorithm for analysis of multispectral aerial images from UAV for identification of water pollution using analytical methods and neural network approaches

The article is devoted to the development of algorithms for the analysis of pollution on the surface of water bodies based on visual information obtained using a multispectral camera mounted on the body of a UAV. The structure of the algorithmic complex for the analysis of multispectral aerial photographs is proposed. Within the framework of the developed approach, each of the analyzed images undergoes a preprocessing procedure that ensures the alignment and alignment of its spectral channels into a single multidimensional raster. The developed analytical algorithm makes it possible to process and convolve the channels of a multispectral image using three mathematical operators - bandpass filtering, contrast change, and brightness change. At the same time, the choice of parameters for identifying pollution on the surface of water bodies is based on a preliminary stage associated with maximizing the contrast excess index for the reference area. The proposed neural network pollution analysis algorithm is based on the application of the sliding window method in combination with the convolutional architecture of the neural network classifier for the analysis of image fragments located on a rectangular grid. The software implementation of these algorithms, as well as the development of a graphical user interface, made it possible to confirm the assumption about the effectiveness of each of the considered approaches. Experimental studies have shown that the neural network algorithm wins in accuracy, and the analytical approach is easier to interpret from the point of view of an expert.