Application of neural networks in system of recognition of commercial aquatic organisms during increased fluctuation

The article focuses on using artificial neural networks for the recognition of commercial hydrobionts in conditions of increased fluctuation. Research has not been conducted in the field of recognition of aquaculture objects for reasons of solving the applied problems and achieving the required indicators. There is described the development of software for recognition of aquaculture objects based on the YOLOv5 model, as well as the proper methodology for the semi-automatic formation of effective visual training samples. There has been analyzed the model operation scheme and substantiated the importance of the training sample and its influence on the quality of pattern recognition by a neural network. Effective identification of aquatic organisms requires the methods for dividing it into zones according to various distinctive features and separating the required objects not only from the background, but also taking into account fluctuations above the water surface. There has been developed a method for eliminating fluctuations in order to increase the efficiency of recognition of aquatic organisms by a neural network. There is described the process of testing software for the detection of commercial hydrobionts trained on a sample by the proposed method. It has been inferred that the recognition rate directly depends not only on the quality of the training sample and the volume of images, but also on the technical characteristics of the electronic computer.