Dehazing Algorithm for UAV Image Based on Smooth Dilated Convolution
Juan Wang, Guanhai Chen, Sheng Wang, Hao Yang, Ye Cao, Yonggang Ye · 2023
Uncrewed Aerial Vehicles (UAVs) are widely used for tasks such as remote sensing mapping, ground target detection, and environmental information extraction. Their advantages lie in the broad aerial perspective, versatile shooting angles, and suitability for different scenarios. In regions characterized by frequent rain and fog, the presence of significant moisture significantly affects the image quality of UAV aerial imagery. This results in reduced contrast, changes in tonality, and obscured details, severely limiting the ability to derive meaningful information from these images. Traditional deep learning-based image-dehazing algorithms rely too much on atmospheric scattering models and are prone to color distortion and detail loss. The end-to-end image-dehazing algorithm presented in this paper uses a smooth dilation convolution approach. Unlike conventional methods based on the atmospheric scattering model, this technique achieves a direct end-to-end dehazing process, seamlessly transforming the input image into a clear output. The color distortion phenomenon can be effectively solved through experiments.