Research on Small Target Detection Method for Aerial Images Based on Improved Yolov8

In order to solve the problems of insufficient accuracy and high model complexity faced by UAV aerial image target detection algorithms, the Yolo-Tpralgorithm is proposed in this paper. The core of the algorithm is to use RFAConvBlock to replace the bottleneck structure in C2f, so as to enhance the feature extraction ability of the convolutional neural network. In addition, we have introduced the Triplet Attention module after the SPPF module to enhance the feature representation force and the robustness of the model. At the same time, we have added a new detection head, P2, which not only improves the detection accuracy and robustness, but also effectively reduces the computational burden. Experimental results on the VisDrone2019 dataset show that the Yolo-Tpralgorithm significantly accelerates the detection speed while the detection accuracy. The mAP50 indicator jumped from 32.2% to 38.5%, an increase of 6.3 percentage points compared with the original model, and performed well in aerial small target detection.