An Airborne Target Recognition Model Based on SPD, PConv and LADH Detection Heads

This paper addresses the need for ground or aerial unmanned systems to reconnoitre aerial targets and provide guidance information by vision sensors at the end of the guidance. To solve the problem of difficult aerial target recognition due to low resolution and large scale variation of moving targets, we propose a target recognition model based on Space-to-Depth (SPD) Conv, Patial-Conv (PConv), Multi-Level Asymmetry Detector Head (LADH) and Yolov8 Baseline. The target recognition model we developed is for small multi-scale targets in aerial low-resolution scenes. And we have conducted the comparative performance experiments to verify that it has better detection accuracy and recall than other SOTA-level detectors, and ablation experiments are conducted to verify that the combination of our improvements can significantly improve the detection performance of the network and minimize the number of additional parameters. Based on the proposed airborne target recognition model and AppDesigner, we also developed a software interface that can display the information of the identified airborne targets in real time.