| Popis: |
The research on image defogging is a long-running topic. With the wave of artificial intelligence and deep learning sweeping the world, the research on image defogging based on deep learning has gradually reached a new height. However, the following problems are still emerging, such as severely distorted image color after defogging, image jagging, and image blurring and so on. In response to these problems, this paper proposes an improved image defogging network based on the traditional CycleGAN. Different from the single feature extraction of traditional CycleGAN, this paper adopts a multi-scale feature extraction method. Combining small-scale convolution and large-scale convolution with residual network for in-depth feature extraction, and at the same time, the extracted features are weighted and fused through the attention module to make the extracted network information more complete. Finally, a clear and fog-free image is reconstructed through upsampling. Compared with the traditional defogging network, the improved network can better avoid the problems of image color distortion and blurring. |
