Key points are not available for this paper at this time.
Intelligent detection of small targets on drones is a key task in the field of computer vision and has a wide range of applications, including intelligent monitoring, security, disaster response, etc. Traditional methods often have difficulty in achieving good detection results in complex backgrounds and multi-target scenes, resulting in missed detections or false detections. To address these challenges, we propose a model that combines sparse convolution-enhanced feature extraction, multi-level attention mechanism dynamically weighted spatial and channel information, and edge enhancement filters to improve target boundary capture. The addition of intelligent detection algorithms further enhances the adaptability and accuracy of the model in real-time scenarios. Experiments on the VisDrone2019 dataset show that the proposed method has superior performance in terms of precision, recall, and mAP compared to classical methods. Ablation studies highlight the importance of each module, especially sparse convolution, in improving detection accuracy.
Wu et al. (Wed,) studied this question.