Innovative Research on Image Dehazing by Integrating Multi-Dimensional Attention and Multi-Model Fusion MixdehazeNet

Due to the complexity of information loss and uneven noise distribution, single image dehazing has always been an important challenge in the field of computer vision. Existing dehazing techniques usually rely on the output of a single control layer. Although this method simplifies the model design to a certain extent, it also increases the complexity of training and significantly increases the demand for computing resources. To address this problem, this paper develops a multidimensional attention module (MDA), which combines the coordinate attention mechanism with a stronger channel attention mechanism, focusing on solving the uneven distribution of fog and the removal of deep haze, thereby improving the accuracy and robustness of the model. At the same time, this paper also innovatively adds 3 control models to the original model for fusion output. By integrating 3 different control models such as the improved atmospheric scattering model and the improved exponential model, and adding them to Mixdehazenet through the multidimensional attention module, the PSNR value is increased by 41.77db while the SSIM value remains unchanged.