Adaptive Sparse Wavelet Denoising and DenseNet121 Framework for Underwater Acoustic Signal Enhancement

The word proposed shines light on an innovative approach for the underwater acoustic signal denoising using sparse TQWT in combination with the wavelet threshold. The focus of the work is to improve the signal quality of underwater acoustic signals that often gets downgraded by noise, which can lower the quality of the analysis accurately and its classifications. To mitigate this, we develop a model based on a. streamline web application, which allows the processing and helps to visualize the acoustic signal in an interactive UI. The original audio signal, is represented in formats such as .MP3 and WAV, which undergo preprocessing steps which includes the removal of noise. the preprocessed signal then undergoes advanced feature extraction technique, which is improved DTQWT to find the key indicators from the preprocessed signal. The data is then divided in to practice and sequences data in the percentage of 80% and 20% for the classification used in the DPL models such as DenseNet121 and GNN. These models performance are then measured in terms of accuracy, SSIM, PSNR and error rate which are used to verify the efficiency of the techniques.