This project woking towards the Shell XPRIZE Ocean Discovery Competition. And I helped our team (Blue Devil Ocean Engineering) getting into the top 10 which will be proceeding to the final round! Click here to see the poster.
And the general goal is to create a high-resolution (5 meters in horizontal direction and 0.5 meters in vertical direction) ocean floor map of a 500 square kilometer area during the competition.
Currently we accomplished identifying different underwater topography, such as hills and valleys, by applying Mel-frequency cepstral coefficients (MFCCs) for feature extraction and Convolutional Neural Network for classifying.
The simulated sonar data are generated through Bellhop software and feed as input of the MFCC and CNN. The confusion matrix and ROC curve shows that the neural network has been well-trained and can classify input data effectively.
Please check out our poster for more details and future plans.