现代电子技术2025，Vol.48Issue(13)：1-10,10.DOI:10.16652/j.issn.1004-373x.2025.13.001
基于FNM-Net的轻量级遥感目标检测算法
Lightweight remote sensing object detection algorithm based on FNM-Net
摘要
Abstract
In view of the current challenges in remote sensing object detection,such as low accuracy,slow processing speeds and large quantity of model parameters,an FNM-Net,a lightweight remote sensing object detection network based on an improved YOLOv7-tiny architecture,is proposed.Firstly,a lightweight feature extraction network Faster-Net is introduced to substitute the original one,so as to prevent the network's excessive overlap of feature maps.Secondly,the focal modulation module is introduced and the spatial information integration module(SIIM)is proposed to construct a new path aggregation network that mitigates issues of the information redundancy and the overlook of intra-layer features during feature fusion.And then,the multi-fine-grained detection heads are designed to avoid the large scale variations of remote sensing objects.Finally,a pruning method utilizing a layer adaptive amplitude pruning(LAMP)score is employed to eliminate connections with minor weights,thereby reducing parameter number and computational burden and increasing the detection speed.This method is validated by the public RSOD dataset.The results show that the proposed method achieves a 51.2%reduction in parameter number,a 55.2%decrease in FLOPs,a 6.5 f/s increase in detection speed,and a 2.1%improvement in mAP(mean average precision)in comparison with those of the baseline model.Additionally,its generalizability is confirmed on the NWPU VHR-10 dataset.关键词
遥感目标检测/FNM-Net/轻量级/剪枝/改进YOLOv7-tiny/SIIMKey words
remote sensing object detection/FNM-Net/light weight/pruning/improved YOLOv7-tiny/SIIM分类
电子信息工程引用本文复制引用
文斌,张俊,王浚银,王子豪,丁弈夫..基于FNM-Net的轻量级遥感目标检测算法[J].现代电子技术,2025,48(13):1-10,10.基金项目
国家自然科学基金项目(62273200) （62273200）
湖北省输电线路工程技术研究中心研究基金(2022KXL03) （2022KXL03）
湖北省自然科学基金联合基金项目(2024AFD409) （2024AFD409）
