Performance Estimation of Parameter Fusion Strategies for Improved Seafloor Classification for Mn-Crust Survey using AUV Data
Umesh Neettiyath, Harumi Sugimatsu, Blair Thornton · 2024
This paper compares the performance of seafloor classification methods based on data collected by autonomous robotic surveys of Cobalt-rich Manganese Crust deposits. Parameters were extracted from acoustic subbottom data using an autoencoder automatically, whereas visual data in the form of 3D color point clouds were analyzed to calculate physical parameters using mathematical equations. SVM classifiers were trained for each sensor modality, which showed that the acoustic classifier performed better than visual classifier. Middle fusion method, in which both parameter vectors are combined into a unified feature vector, performed better on almost every measure of performance than the previous two. Furthermore, it could prevent misclassification from both modalities.