UAV Aerial Image Target Detection Based on Visual Attention Combined with Dual-weight Adaboost

UAV Aerial Image Target Detection Based on Visual Attention Combined with Dual-weight Adaboost

For the problem that UAV (Unmanned Aerial Vehicle) aerial image targets are small and easily interfered by background and illumination changes. This paper proposes a UAV aerial image target detection algorithm combining visual attention with dual-weight adaboost (Dw-adaboost). First of all, a visual attention combined with Dw-adaboost image saliency detection framework is proposed, we selected the brightness, direction, regional contrast, and spatial location features as the main feature channels to generate the saliency map; Secondly, a Dw-adaboost classification algorithm is proposed to determine the optimal weight of the main feature channel; Finally, we use high-efficiency sub-window search on the saliency map to achieve target detection in aerial images. Experiments show that the method in this paper improves the problem that UAV aerial image targets are small and easily interfered by background and illumination changes. It can achieve more accurate detection of UAV aerial image targets in complex scenes.