Identification of plankton populations in the surface waters of the Azov Sea based on neural network structures of various architectures

Don State Technical University, 1, Gagarin sq., 344003 Rostov-on-Don, Russia

^* Corresponding author: cvv9@mail.ru

Abstract

In recent years, deep learning technology has been widely used to solve the problem of recognizing the boundaries of structures on the surface of a reservoir. This technology opens up significant opportunities for the use of aerospace methods of geoecological forecasting. The purpose of this work is to identify plankton populations on the surface of the Sea of Azov, using remote sensing data, including space images of the Earth. Machine learning algorithms for segmentation of these structures on the surface of the reservoir, based on convolutional neural networks with the following architectures, are built: U-Net, FCN32, SegNet, DelitedSegNet, U-Net (transposed convolutional). A comparison of the used neural network models based on the IoU (Intersection over Union) metric is carried out. The highest accuracy was demonstrated by the algorithm created on the basis of the U-Net neural network architecture. The results obtained can be used to create maps of the distribution of plankton populations and assess water quality.