LEARNING METHOD AND INFERENCE METHOD BASED ON CONVOLUTIONAL NEURAL NETWORK FOR TONAL FREQUENCY ANALYSIS

摘要

The present invention provides a convolutional neural network-based learning and inference method for tonal frequency analysis in a sonar gram generated by performing frequency analysis on acoustic signal data received using a sonar sensor in water and on the surface of the water. In contrast, the steps of extracting positive sampling and negative sampling at a predetermined ratio based on the window size to generate training data, and using the convolutional layer in multiple layers, the kernel size of a long vertical rectangle (Kernel Size) forming a convolutional neural network structure and learning the convolutional neural network structure using the training data, and for a phonogram, a region as much as a predetermined window size for a phonogram. extracting from each frequency domain to generate a batch, and inputting the batch into the learned convolutional neural network structure to infer the existence and change of tonal frequencies.