Research On Single Image Dehazing Algorithm Based On Deep Learning

As a weather phenomenon,haze not only affects the atmospheric environment,but also troubles the safety of people’s transportation,the imaging quality of outdoor imaging equipment,and scientific research tasks based on computer vision.Image dehazing technology,as a fundamental research content,aims to restore clear and haze-free images from hazy images,improve the overall quality of images,and have a positive impact on practical life and scientific research.In recent years,with the iterative updating of related technologies,significant progress has been made in the study of single image dehazing.However,many problems still exist,such as the presence of artifacts in the image and the overall quality of the image being still poor.To alleviate these problems,this paper proposes two single image dehazing algorithms based on Encoder-Decoder structure,and applies the proposed dehazing algorithm to traffic object detection tasks in hazy weather,Improve its detection effect.By selecting objective evaluation indicators such as Peak Signal-to-Noise Ratio and Structural Similarity Index Measure on the RESIDE dataset,the model is tested and compared,as well as the real-world dataset Foggy_driving and UADETRAC,indicating that the single image dehazing algorithms HFGAN and HF-ViT proposed in this paper can maintain good dehazing performance in indoor and outdoor scenes and various complex hazy weather backgrounds.The main work of this paper is as follows:1.Aiming at the problems of artifacts,poor detail effect and weak generalization ability of dehazing model in the dehazing image,an single image dehazing model HFGAN(Haze Free GAN)based on generative adversarial network is proposed,which uses parallel dilated convolution in the generator,it can expand the receptive field of the model,learn global feature information and enhance the feature expression ability without adding additional computation.At the same time,an improved composite loss function is introduced in the model to constrain the boundary information of the generated image and improve the detailed features of the image after dehazing.Finally,the experimental verification of multiangle subjective and objective methods proves the effectiveness of the HFGAN algorithm.2.Currently,the objective evaluation indicators of Peak Signal-to-Noise Ratio and structural similarity of images processed by image dehazing algorithms need to be further improved.At the same time,there is room for improvement in the comprehensive quality and visual sensory effects of images after dehazing.Regarding the above issues,a multi-head self attention module with linear time complexity has been designed based on Vision Transformer,Based on this,a image dehazing network HF-ViT(Haze Free Vision Transformer)is constructed.This model restores high-quality haze-free images through the strong representation ability of the Transformer,and introduces a feature attention mechanism to further improve the quality of the dehazing image.In addition,by designing a self-attention mechanism module with linear complexity,the overall computational complexity of the model is reduced while improving the quality of the dehazing image.The ablation experiment verifies the overall effectiveness and good dehazing ability of the algorithm HF-ViT.3.At present,the image dehazing algorithm needs further application in practical life,This paper applies the two proposed dehazing algorithms to the detection task of traffic targets in hazy weather.Firstly,the proposed algorithm is used to dehaze images in specific scenes,and then sent to the target detection model for traffic target detection.Experiments have shown that the quality of object detection in images processed by the algorithm in this paper has been further improved.4.This paper designs and implements an image dehazing system,integrating the two dehazing algorithms and various comparison algorithms mentioned into the system interface.Users can select the image dehazing algorithm to achieve image dehazing function and object detection function.Through system testing,it has been shown that the system operates stably and reliably.