Correction of roll-caused stripe noise in side scan sonar images

Ensuring high-quality images obtained using side-scan sonar is crucial for enhancing the effectiveness of underwater research, as distortions such as striping noise can complicate data analysis. The aim of this paper was to investigate the nature of striping noise, determine the correlation between image intensity and the tilt of the sonar, and develop a new method to improve the quality of sonar images. The study employed a statistical correction method based on calculating a horizontal moving average for intensity correction, as well as a machine learning model using a threelayer neural network to predict the horizontal moving average considering the beam’s incidence angle, the sonar’s height above the seafloor, and the initial line intensity. Statistical methods and machine learning techniques were applied to correct the striping noise caused by tilting in sonar images, significantly enhancing their quality. The statistical approach, which uses the mean value of the horizontal sway, effectively reduced noise while preserving critical details and improving overall clarity. The machine learning model incorporated additional parameters, enhancing intensity prediction accuracy and improving adaptability to various sonar positioning conditions. Moreover, the new method accounts for varying environmental conditions, making it flexible and effective for real-world underwater research. These results provide valuable insights for improving sonar image processing methods, paving the way for more efficient underwater exploration and improving the accuracy of object detection on the seafloor