Fast defogging algorithm based on improved Ghost module

Recently, end-to-end learning algorithms based on convolutional neural networks have made progress in defogging, but most of them use networks that are too deep and too large to fit foggy image data, resulting in slow defogging speed. To solve this problem, an efficient defogging model is proposed in this paper, which can be used to quickly defog a single image. The model is based on the improved GHOST module to build a lightweight neural network.Pyramid pooling and attention mechanism are combined to obtain the context information of different areas of the image, thus improving the network's ability to obtain the global information of foggy images. The network RESIDE in Donatella's outdoor data set for training and testing. Experimental results show that, compared with DCP, MSCNN, AODNet and other advanced algorithms, the proposed method has satisfactory defogging quality and speed.