Wavelet analysis and neural networks for automatic classification and reconstruction of buried sedimentary features

The use of acoustic methods for geomorphologic reconstruction has been well established over several decades. However, the manual description and interpretation of large acoustic datasets can require a long time for the mapping and reconstruction of the morphological features revealed by the acoustic surveys. In order to reduce the processing time of the traditional descriptive analyses of acoustic echograms, we implemented a new algorithm of sub-bottom sediment classification based on wavelet analysis and neural networks. The optimized algorithm consists of an automatic echo shape parameters classification procedure, dedicated to extract the morphological features buried in the sediments up to the first 5-6 m beneath the bottom surface. We developed and tested a classification algorithm containing a limited set of wavelet transformation parameters as input to a Self-organizing Neural Network. This algorithm was applied on 580 km of acoustic data echograms acquired in the Venice Lagoon with a 30 kHz single-beam echosounder without any special pre-requirement on the data, which were collected over several years. The algorithm was successfully tested against the descriptive analysis allowing in a very short time a 2D mapping of the buried features of the area, distinguishing between different types of palaeochannels, buried creeks and no channel areas. The test was supported by the information about the sedimentary facies of 11 cores sampled in the survey area allowing a detailed palaeo-environmental reconstruction.