A denoising representation framework for underwater acoustic signal recognition.

Abstract

To suppress the noise interference in underwater acoustic signals for recognition, a practical denoising representation and recognition method is proposed. This algorithm first generates the multi-images between marine noise and target signal by correlation and "dropout" processing, adaptively. Second, a convolutional denoising autoencoder is designed to train the segmented multi-images in parallel to acquire denoising features. Finally, to improve the classification accuracy of random forest (RF), the weight fusion is exploited to initialize parallel RF classifier. Numerical experiments are shown that demonstrate superiority to three other methods in feature denoising and classification under underwater acoustic scenes.