Acoustic sediment classification methods are important for protecting and studying the ecological environment of rivers and seas. Single-beam echosounder, multi-beam echosounder and side-scan sonar are the main measuring means for sediment classification. Recently, neural networks have been applied to sediment classification to improve its accuracy.
Sediment classification using single-beam echosounders has the longest history, and many methods have been proposed for sediment classification using single-beam acoustic sounders.
Jong Kum Chol, a researcher at the Faculty of Naval Architecture and Ocean Engineering, proposed a new MLP-KNN model, a combination of recently widely-used MLP neural network and KNN model to improve sediment classification accuracy in shallow waters.
In shallow waters, the accuracy of sediment classification is reduced because echoes are overlapping. Hence, he first designed an MLP neural network where six feature parameters from the overlapping echo are taken as an input layer and the area of the overlapping part as an output layer, and estimated the area of the overlapping part. Then, based on the estimation results, he calculated E1 and E2 derived from the first and second echoes, respectively, and classified sediment using the KNN model.
The comparison results with the sediment classification method by BP neural network showed that the proposed method classifies sediment with higher accuracy in shallow waters.
For more information, please refer to his paper “Acoustic sediment classification using MLP-KNN model on single-beam echosounder data from shallow water” in “Marine Systems & Ocean Technology” (EI).