Process the ShipsEar dataset, the dataset includes 90 audio files, includes 5 classes and 12 categories. We preprocessed the data by removing the blank signal and framed the original signal by 5s. Finally, 1956 labeled sound samples with a total of 12 categories of targets can be used.
Category Targets Frames Class A fishing boats, trawlers, mussel boats, tugboats, dredgers 98/28/ /95/23/52 Class B motorboats, pilot boats, sailboats 195/26/76 Class C passenger ferries 703 Class D ocean liners, ro-ro vessels 174/261 Class E background noise 225
The 'Preprocessing' achieve the function of framing, visiualization, add noise, feature selection The 'ShipsEar_Classification_CNN,LSTM,CRNN' verify the classification performance of different network models The 'ShipsEar_Classification_ResNet' is the transfer learning The 'ShipsEar_Classification_VGGish' Verify Google's VGGish network classification performance The 'ShipsEar_data_aug 5s/1s' and 'specaugment' verify the data augmentation with methods of google's specAugment and network structure with VGGish
The classification performance can reach 97%, but the division of training set and test set is not independent, it does not mean that the method is effective. In the future, will continue to verify.