Methods of Introducing Reference Objects Into Images Obtained Using Hydroacoustic Systems of Computer Vision
Aleksandr Saenko, Andrey Mironov, Ekaterina Fomina · 2024
The purpose of the presented work is the development of neural networks to solve the problem of introducing an object with given geometric parameters in the images obtained using side (sector) scan sonars (SSS). As part of the study, there was conducted an experimental assessment of the effectiveness of applying the proposed comprehensive algorithm for the introduction of objects that puts into effect the use of the “Copy and Paste” method together with the neural network method of “Deep Painterly Harmonization”. In the course of the study, methods of visual subjective qualitative assessment and quantitative analysis were used to determine the effectiveness of the result of introducing of objects in the image. The evaluation criteria were the realism of the images obtained and the retention of the geometric parameters of the introduced reference objects. The results obtained allow for making a conclusion on the advantages and disadvantages of applying the proposed comprehensive algorithm with the use of each of the considered methods to solve the problem of introducing objects into sonar images. In the conclusion, on the basis of the conducted analysis, recommendations are proposed for selecting the most suitable method for introducing objects with specified parameters to the image of the side scan locator, taking into account specific needs and tasks.