Publications

Abstract : Underwater image augmentation is a technique for recovering low-resolution underwater photographs to produce equivalently high-resolution images. Deep learning-based techniques for enhancing photos commonly use paired data to train the model. Another crucial issue is how to effectively keep the fine details in the improved image. In order to address these problems, providing a revolutionary unpaired underwater picture-enhancing technique that fixes the underwater photos using a cycle generative adversarial network. At last test results on two datasets of unpaired underwater image datasets showed the recommended model’s utility by surpassing cutting-edge image enhancement methods. The CycleGAN generator includes a content loss regularizer that retains the important information of one low-resolution image in the corresponding clear image that is generated. The results are given based on mean square error and mean absolute error after the findings have been optimized using the ADAM optimizer. This architecture has a 4 × 4 kernel and a 2 × 2 stride, using Adam as the optimizer and a learning rate of 0.002.