Combined LOFAR and DEMON Spectrums for Simultaneous Underwater Acoustic Object Counting and F0 Estimation

In a typical underwater acoustic target detection mission, we have to estimate the number (N), perform source separation when N>1, and consequently predict motion parameters such as fundamental frequency (F0) from separated noises for each target. Although deep learning methods been adopted in task, their successes strongly depend on feed-in features. this paper, evaluate several time-frequency features propose universal feature extraction strategy object counting F0 estimation simultaneously, with convolutional recurrent neural network (CRNN) backbone. On one hand, LOFAR DEMON are feasible low-speed high-speed analysis, respectively, combined (LOFAR + DEMON) cope full-condition estimation. other comb filter (COMB) is designed applied spectrum harmonicity enhancement, which will be further streamed into CRNN prediction. Experiments show that (1) feeding filtered COMB) achieves an accuracy of 98% lake trial dataset, superior COMB (83%) or (94%) alone, demonstrating combination plausible. (2) prediction DEMON, included excluded) comparable state-of-the-art simulation dataset dominates rest indicating can used common both tasks. (3) The inclusion accelerates convergence speed however, it penalizes task by depression 13% average, partly due merging effects brought broadband filtering COMB.