Abstract:

To address the problem of low accuracy exhibited by lightweight models for UAV remote sensing target detection in low-power hardware environments, a lightweight detection model based on improved GhostNetv3 is proposed. Taking GhostNetv3 as the backbone network, a dual-branch convolution module is introduced to improve feature representation capability, and depthwise separable convolution down sampling (DSConv-Down) is used to further reduce computational overhead. Through simplified SPPF, target features at multiple scales are aggregated. In the feature fusion stage, a multi-scale feature pyramid is constructed to fully integrate multi-level features. In the detection stage, a consistent dual-assignment detection head is used to avoid the computational overhead caused by the non-maximum suppression (NMS) algorithm. Experimental results show that the proposed model outperforms current mainstream lightweight models in detection accuracy across different datasets, and demonstrates good generalization ability. On the test hardware platform, it also achieves fast detection inference.

Key words: UAV remote sensing, lightweight target detection, GhostNetv3, DualConv, multi-scale feature pyramid, DLA Head