Cross-Layer Cascading Chain-of-Thought Based Object Detection Algorithm for Aerial Images in Foggy Weather

In response to the challenges posed by fog obstruction and small object in the detection of aerial images under foggy conditions, this paper presents an image restoration network (CLC-CPA) based on the cross-layer cascading chain-of-thought prompt mechanism that can be integrated with detection algorithms. The proposed network executes the restoration strategy step by step, with each step based on the results of the previous step (i.e., chain-of-thought mode), gradually restoring foggy images to a level conducive to detection. Additionally, it combines deep and shallow information via cross-layer cascading to preserve the features of small objects. Furthermore, the HorNet is employed to enhance the backbone network of the detection algorithm, extending feature extraction into higher-order spaces, which enables the network to capture more distinguishable spatial relationships and improves the detection accuracy of small objects in aerial images. Finally, utilizing the VisDrone dataset, a foggy scene aerial image dataset (FoggyVisDrone) is constructed based on the atmospheric scattering model. Experimental results indicate that, in the test set of aerial images with varying fog concentrations, the mAP50 of the improved algorithm exceeds that of the baseline algorithm by more than 4.4%.