Applicability Analysis of VGG Network for Marine Mammal Calls Recognition

Deep learning method has been widely used in Marine mammal calls (MMC) recognition. In this paper, we studied the influence of parameter changes on accuracy for VGG of MMC recognition. We present the pre-training VGG model, the self-training VGG model, and the VGG model for changing the size of the convolutional kernel. The effects of different combinations of convolutional neural network activation function, loss function, and optimization algorithm on the performance of used models are investigated. And then MMC classification experiments are carried out. Finally, deconvolution method is utilized to analyze the recognition results. The experimental results show that the pre-training model uses the ReLU activation function, Categorical Cross entropy and Adagrad algorithm with a learning rate of 0.01 has the best accuracy.