Research On Single Image Dehazing Algorithm Based On Convolution And Deep Neural Network

Single-image dehazing research has garnered considerable attention as a fundamental problem in the field of computer vision and has become one of the challenging research directions.In hazy weather conditions,atmospheric particles suspended in the air cause severe light scattering,resulting in image degradation,color attenuation,and reduced contrast,significantly diminishing image quality.Therefore,research on image dehazing holds significant practical significance.In this paper,we focus on the problem of single-image dehazing and conduct research in several aspects:Firstly,addressing the color distortion and haze residue issues present in existing image dehazing algorithms employing prior-based and deep learning approaches,we propose an attention-based detail recovery algorithm for image dehazing.We introduce an improved convolutional network-based attention module into the network model and design attention basic blocks,which are encapsulated into grouped blocks.Additionally,to enhance information interaction within the grouped blocks,we incorporate dense connection residual blocks between them.Finally,we design a detail recovery module to restore the details in the dehazed image and further alleviate the influence of haze residue.Numerical simulation experiments demonstrate that the proposed algorithm achieves higher image evaluation metrics compared to mainstream dehazing algorithms on the RESIDE(Realistic Single Image Dehazing)dataset,while also yielding superior visual results on real-world images.Furthermore,to explore a more superior dehazing model while reducing network parameters,we propose a single-image dehazing algorithm based on a network architecture that combines Transformer modules and a decoder structure.We introduce Transformer multi-head channel attention blocks into the network model to enhance the model’s ability to extract global features.Moreover,based on the residual dense connection blocks,we introduce channel attention and spatial attention to design an improved residual dense connection block that extracts important features and promotes information interaction among them.The network architecture is designed as an encoder-decoder structure,incorporating up-sampling and down-sampling modules to extract rich features at different resolutions.Experimental results demonstrate that the proposed algorithm achieves favorable image evaluation metrics on the RESIDE dataset,while also delivering excellent visual effects.In summary,this paper addresses the existing issues in single-image dehazing and proposes two single-image dehazing algorithms combining attention mechanisms and convolutional neural networks.Compared to mainstream dehazing algorithms,the proposed algorithms achieve impressive dehazing effects in both objective evaluation metrics and visual results,laying a foundation for further practical applications and demonstrating practical value.