Abstract:First of all, the framework of sediment classification system based on echo intensity is introduced, The paper then analyzes storage structure and content of ALL format multibeam sonic data with multibeam sonar principles. It can decode ALL format data,and extract all kinds of data packets from ALL raw file by means of software developing.By means of a series of data preprocessing, it can create high-resolution sonic images based on a footprint envelope grid from these sonic data. In the end,the sediment classification is implemented with improved BP neural network in the sonic image.The experiment proves that it can obtain good effects for sediment classification with sonic image, and provide an accurate and reliable basis for submarine microgeomorphology recognition.