Research on Target Detection Algorithm of UAV

In the UAV aerial photography scene, aerial images have the characteristics of high density, small target size and far-reaching operational area, which create a domino effect toward the problems of false detection and missing detection within current target detection models. For the purpose of effectively increase the accuracy of target recognition, we introduce an improved target detection model for YOLOv8. Specifically, first, we introduce DConv2 and integrate it into the C2F module in the backbone network to extend the receptive sides of the image, thus improving the detection efficiency of tiny targets; Secondly, in the neck network, we have adopted Fusion Block module to enhance the network depth and increase the network feature syncretism capability; Then, by adding the taking sample operations, we enable the tiny target detection layer and discard the large target detection layer, thus further improving the detection accuracy of tiny targets; Finally, we applied the ECIoU loss function for the model to replace the original CIoU loss function, which improved the accuracy of target detection and accelerated the convergence speed of the model. The improved model has been verified by the wellknowed dataset, and the experiments show that the improved YOLOv8s_UAV model has significantly improved its performance. On the premise for meeting the requirements of UAV calculating speed, the model we improved can detect small target more accurately in the process photography set.