No abstract

Abstract. Under the influence of light refraction and particle scattering, the underwater image has a low contrast and color attenuation. This paper proposes an underwater image enhancement method based on Generative Adversative Networks (GANs). The main principle is to apply the GANs to learn the pixel transformation between images through confrontation training, so as to achieve the purpose of image enhancement. In this paper, we select the real underwater image as the data set and degrade it before inputting the image into the neural network. Then we input the corresponding degenerate image, so that the neural network can learn the mapping from input image to real image and realize the color reduction and enhancement of underwater image. Experiments show that this method has strong robustness and can be applied to underwater shooting and underwater navigation.

In order to solve the problem of abnormal detection of household appliances and realize the extraction of abnormal signal features of household appliances and the positioning of the time, this paper proposes an abnormal processing method based on wavelet denoising and DAG-SVM. The noise is eliminated by the wavelet threshold denoising method, which reduces the interference of the noise to the fault signal. Thus, the real signal can be reduced from the signal of strong noise. The abnormal signal of household electrical appliances was extracted after noise reduction. These characteristics are input into DAG-SVM for training and modeling. The validity of this method is verified by simulation and experiment.