Seabed sediment classification using multibeam backscatter data based on the selecting optimal random forest model

摘要

Seabed sediment classification using acoustic remote sensing technique is an attractive approach due to its high coverage capabilities and limited costs compared to taking samples of the seafloor. This paper focuses on backscatter intensity correction, sonar image quality improvement, and classifier construction, which aims to improve the accuracy of seabed sediment classification. The details are as follows. 1) A series of multibeam echosounder backscatter intensity correction model is constructed, including time-varying gains (TVG), transmission loss, actual area of insonification, source level, transmitting and receiving beam patterns, specular area correction, etc., to obtain accurate intensity values that accurately reflect seabed sediment types. 2) The pulse coupled neural network (PCNN) image enhancement model is established to improve the quality of sonar images, and 40 dimensional features are included to enrich the intensity description. 3) Selecting optimal random forest (SORF) seabed sediment automatic classification models that can select the input feature vectors and optimize the model parameters automatically are established. 4) Taking multibeam backscatter intensity data collected in Jiaozhou Bay as an example, the effectiveness and advantages of SORF are verified by comparing with support vector machine (SVM) and random forest (RF) classifiers. (C) 2020 Elsevier Ltd. All rights reserved.