The paper proposes the feature fusion attention network (FFAN) algorithm to remove fog effects from images. The input of FFA-Net is a fuzzy image. It is transformed into a sparse feature map and fed into group architectures (N) with multiple skip connections. The output features of N are merged through the feature attention module. The final features will then be passed on to the reconstruction and the remaining global learning structure. Therefore, it is getting a smoke-free output blind. Furthermore, every group architecture combines (B) basic blocks with local residuals. The basic block combines bypass connections and feature attention (FA) modules. FA is an attention mechanism structure consisting of channel-wise and pixel-wise attention. The fog effect is a common problem in image processing. It can reduce the quality of the image and make it difficult to analyze and identify objects. The results show that the algorithm achieves high accuracy with a PSNR coefficient of up to 36.39 and SSIM of up to 0.992 which is feasible for practical application. This result reveals the proposed approaches’ readiness to be applicable in real-world applications, especially for the intelligent transportation system (ITS) services in urban cities like Hanoi and Ho Chi Minh City.