LVMDet: locally enhanced state space network for aerial transmission line foreign object detection

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

To address the challenges of complex background interference and multi-scale object detection in aerial transmission line scenarios, LVMDet, a foreign object detection network based on a state-space model, is proposed. Specifically, an efficient dynamic local perception (EDLP) block is incorporated into the VMamba framework to construct the Local VMamba backbone, enhancing the representation of fine-grained features such as edges and textures and improving foreground–background discrimination. In addition, an adaptive dilated depthwise fusion module is embedded within EDLP, where dynamic receptive field modulation enables effective integration of local details and contextual information. Furthermore, a hierarchical adaptive feature fusion network is developed by combining a bidirectional feature pyramid with a group-aware adaptive spatial feature fusion module. Through pixel-wise dynamic weighting, multi-scale feature alignment and adaptive fusion are achieved, enhancing the model’s adaptability to objects of different scales. Experimental results on the self-built TLOI-5 dataset and the public InsPLAD-det dataset demonstrate that LVMDet achieves mAP@0.5 scores of 91.8% and 88.8%, respectively, outperforming YOLOv12m by 2.9% and 2.4%, which verifies its effectiveness and generalization ability in complex transmission line scenarios.