GFLNet: A lightweight real-time detector with knowledge-guided compression in remote sensing images
- 1Faculty of Computer Science and Information Technology, Universiti Putra Malaysia, Serdang, Malaysia.
- 2Department of General Education, Anhui Vocational College of City Management, Hefei, China.
- 3Whole vehicle development-Material technology, Volkswagen China Technology Company, Hefei, Anhui, China.
Abstract
Remote sensing images usually exhibits densely distributed targets and scale disparities. In such scenarios, it is still challenging to preserve high detection precision while controlling costs and inference latency. To address this issue, we propose GFLNet, a lightweight single-stage detector, which is based on collaborative feature modeling and knowledge-guided compression. Specifically, Ghost-Shuffle dual fusion convolution (GSD-Fuse) is designed to achieve efficient spatial-channel interaction while reducing computational cost; an adaptive receptive-field selection module inspired by LSKNet is introduced into the deeper backbone and neck stages to adaptively adjust contextual responses for small-object and multi-scale detection; finally, LB-Compress combines layer-adaptive magnitude pruning with Bidirectional Cross-task Knowledge Distillation (BCKD) to form a compact network suitable for practical deployment. Evaluated on the DIOR and UCAS-AOD datasets, GFLNet improves [email protected] over the baseline by 3.9% and 3.4%, respectively. Moreover, the model contains only 2.35M parameters, with a calculation of 5.6 GFLOPs, and an inference speed of 261 FPS on the GPU, thereby delivering a favorable accuracy-efficiency synergy.
Related
- A lightweight deep-learning detector for the real-time monitoring of the invasive frogs Rhinella marina and Polypedates leucomystax
- CAS-YOLO: A lightweight and efficient object detector for infrared UAV imagery
- Lightweight YOLOv8n marigold corolla detection and picking-point localization method based on ECR progressive channel compression and dual attention mechanisms
- A temperature-adaptive logit-standardized knowledge distillation for lightweight network intrusion detection
- Bridging Local Ecological Knowledge and Remote Sensing for Wildlife Assessment and Conservation: Insights from Sebitoli, Kibale National Park, Uganda
BibTeX
@article{zhang2026gflnet,
title = {{GFLNet: A lightweight real-time detector with knowledge-guided compression in remote sensing images}},
author = {Jing Zhang and Mas Rina Mustaffa and Fatimah Khalid and Zainal Abdul Kahar and Tao Zhang},
journal = {PLOS One},
year = {2026},
month = oct,
doi = {10.1371/journal.pone.0361806},
url = {https://doi.org/10.1371/journal.pone.0361806},
note = {CC BY},
}
RIS
TY - JOUR
TI - GFLNet: A lightweight real-time detector with knowledge-guided compression in remote sensing images
AU - Jing Zhang
AU - Mas Rina Mustaffa
AU - Fatimah Khalid
AU - Zainal Abdul Kahar
AU - Tao Zhang
PY - 2026
DA - 2026/10/30
JO - PLOS One
DO - 10.1371/journal.pone.0361806
UR - https://doi.org/10.1371/journal.pone.0361806
AB - Remote sensing images usually exhibits densely distributed targets and scale disparities. In such scenarios, it is still challenging to preserve high detection precision while controlling costs and inference latency. To address this issue, we propose GFLNet, a lightweight single-stage detector, which is based on collaborative feature modeling and knowledge-guided compression. Specifically, Ghost-Shuffle dual fusion convolution (GSD-Fuse) is designed to achieve efficient spatial-channel interaction while reducing computational cost; an adaptive receptive-field selection module inspired by LSKNet is introduced into the deeper backbone and neck stages to adaptively adjust contextual responses for small-object and multi-scale detection; finally, LB-Compress combines layer-adaptive magnitude pruning with Bidirectional Cross-task Knowledge Distillation (BCKD) to form a compact network suitable for practical deployment. Evaluated on the DIOR and UCAS-AOD datasets, GFLNet improves [email protected] over the baseline by 3.9% and 3.4%, respectively. Moreover, the model contains only 2.35M parameters, with a calculation of 5.6 GFLOPs, and an inference speed of 261 FPS on the GPU, thereby delivering a favorable accuracy-efficiency synergy.
SN - 1932-6203
ER -