Underwater Acoustic Image Classification Based on Image Denoising and Transfer Learning

With the potential shown by artificial intelligence in recent years, deep learning has been widely used in the field of marine environment monitoring such as underwater image classification. It will be disturbed by the marine environment in the acquisition process of underwater acoustic images, which will lead to the acquisition of images containing a lot of noise information. In addition, it is very difficult to collect a large amount of underwater acoustic image data because it is expensive and time-consuming to do so. Both of these factors will reduce the accuracy of underwater acoustic image classification. In order to overcome the influence of the above factors, we proposed an underwater acoustic image classification method based on image denoising and transfer learning. Firstly, we use curvature filtering for image denoising to eliminate the noise contained in the image, which can effectively remove noise while maintaining the edge information of the image. Then we utilize the transfer learning technique to improve the accuracy of underwater acoustic image classification. The experimental results show that the curvature-filtered images have better image evaluation indexes than other methods. Moreover, the classification methods based on transfer learning has better classification accuracy than the ones without transfer learning.