PTCDet: advanced UAV imagery target detection

Abstract

Abstract Object detection in drone aerial images is challenging due to the difficulty of small object detection and complex backgrounds. To address these issues, this paper proposes an improved object detection model, perception and target capture detector (PTCDet), to increase detection accuracy and robustness in complex scenes. Specifically, the proposed multiple feature extraction attention (MFEA) module significantly enhances the ability of the model to detect small objects through multidimensional feature map augmentation. The weighted perceptive field augmentation (WPFA) module is designed to improve the contextual awareness and feature representation of the model, optimizing detection accuracy for small objects. Based on the multiscale feature fusion structure, an enhanced scale fusion detection (ESFD) module is used to improve small object detection by generating larger scale feature maps. Ultimately, the inner focaler IoU loss (INFL) function effectively accelerates the regression of detection bounding boxes, enhancing the generalization ability and overall detection performance of the model. The experimental results on three public datasets demonstrate that PTCDet outperforms other detection algorithms. For example, on the VisDrone dataset, compared with the baseline model YOLOv8, [email protected] and [email protected]:0.95 are improved by 6.21% and 4.21%, respectively. PTCDet exhibits excellent performance in addressing complex backgrounds and small object detection, providing an effective and robust solution for object detection tasks in drone aerial images.

Keywords

• Object detection
• YOLOv8
• Small target
• Loss function