结合特征增强注意力的混合卷积去雾网络OA

Fog can cause severe visual degradation to images,and affect their details and contrast.Furthermore,it will impact the readability of the images and the performance of subsequent processing tasks.In view of the incomplete feature extraction,loss of image details,and poor dehazing effect on non-uniform hazy images found in existing image dehazing algorithms,a mixed convolutional dehazing network integrating feature enhancement attention is proposed.Differential convolution is combined with original convolution to form a mixed convolution layer,expanding the feature information extraction range.The feature enhancement attention module formed by pixel attention mechanism and convolutional block attention module is used to improve the detail processing ability of the network.The feature information of channel,space and pixel is fused to make the network focus on the differences of fog distribution.Experimental results show that the proposed network can extract features comprehensively,produce detailed and clear dehazed images,and achieve thorough dehazing.It performs well on both objective indicators and subjective visual assessments,and has good dehazing effect while maintaining strong robustness and generalization ability.