Data correction for visualisation and classification of sidescan SONAR imagery

Of all the remote sensing modalities available for underwater applications, acoustic methods, covering frequency ranges from a few Hz to several MHz, are by far the most flexible and widely used. The authors propose a method for preprocessing sidescan sonar data for visualisation, detection and classification purposes. Sidescan imagery is highly sensor specific and is typically affected by factors that have either a range dependency or an angular dependency. Each of these is altered in a different way given variations in sensor altitude over the seabed. Working from the physics and geometry of the sonar process, the proposed method estimates separate correction factors for range and angular dependencies directly from the image data. Once calculated, these factors can be applied over large data sets to provide radiometric correction over the entire survey area. Simpler image processing algorithms are more effective because the image statistics are improved with more stable means and variances across the sonar swath. The method requires a good bottom detection algorithm for estimation of sensor altitude at each transmission time and incorporates a resampling scheme for the calculation and application of the angular-dependency correction factors. Results showing improved classification performance for two large area surveys are presented. The method proposed provides a more complete solution than previously reported resampling schemes and offers significant improvements in terms of accuracy, robustness, usability and execution times.