摘要
针对遥感影像中森林火焰烟雾检测任务存在的多尺度特征差异显著、复杂背景干扰严重以及小目标漏检等问题,本文提出一种基于改进RT-DETR的森林火焰烟雾目标检测方法.该方法引入特征调制融合模块,强化多尺度跨层级特征的有效融合;设计轻量化瓶颈结构,实现空间语义特征与局部细节特征之间的信息交互;同时,添加P2小目标检测层,增强模型对小目标火焰图像局部特征信息的关注程度.实验结果表明,本文算法参数量降低7.40%、精确率提升1.07%、召回率提升3.58%、平均精度均值mAP50、mAP50-95分别提升3.49%、1.12%,同时,F1分数从0.799 3提升至0.824 0,能更好满足森林火焰、烟雾等复杂场景下小目标的检测定位需求.
To address the issues in the task of forest fire and smoke detection in remote sensing images,such as significant multi-scale feature differences,severe interference from complex backgrounds,and missed detections of small targets,this paper proposes a forest fire and smoke object detection method based on an improved RT-DETR.This method introduces a feature modulation and fusion module to strengthen the effective fusion of multi-scale cross-level features;designs a lightweight bottleneck structure to realize the information interaction between spatial semantic features and local detail features;meanwhile,adds a P2 small target detection layer to enhance the model s attention to the local feature information of small-target fire images.Experimental results show that the algorithm proposed in this paper reduces the parameter amount by 7.40%,improves the precision by 1.07%and recall by 3.58%,and increases the mean average precision(mAP)mAP50 and mAP50-95 by 3.49%and 1.12%,respectively.Meanwhile,the F1-score rises from 0.7993 to 0.8240,which makes it more capable of meeting the requirements for detecting and locating small targets in complex scenarios such as forest fires and smoke.
作者
龚德燕
赵晨萌
孙云洲
袁淑婷
蒋雨婷
张卡
Gong Deyan;Zhao Chenmeng;Sun Yunzhou;Yuan Shuting;Jiang Yuting;Zhang Ka(State Key Laboratory of Climate System Prediction and Risk Management,Nanjing Normal University,Nanjing 210023,China;School of Geography,Nanjing Normal University,Nanjing 210023,China;Key Laboratory of Virtual Geographic Environment,Nanjing Normal University,Ministry of Education,Nanjing 210023,China;Jiangsu Center for Collaborative Innovation in Geographical Information Resource Development and Application,Nanjing 210023,China)
出处
《南京师大学报(自然科学版)》
2026年第1期115-124,共10页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金项目(42271342)。
关键词
遥感影像
火焰烟雾检测
多尺度特征
分组重排卷积
特征调制融合
RT-DETR
remote sensing images
smoke and fire detection
multi-scale features
group shuffle convolution
feature modulation fusion
RT-DETR
