基于深度学习的小目标森林火灾识别方法  

Small Target Forest Fire Recognition Method Based on Deep Learning

作　　者：罗明明 黄建华[1,2] 孙希延[1,2] 万逸轩 LUO Ming-ming;HUANG Jiang-hua;SUN Xi-yan;WAN Yi-xuan(Guanxi Key Laboratory of Precision Navigation Technology and Application,Guilin University of Electronic Technology,Guilin Guangxi 541004,China;National&Local Joint Engineering Research Center of Satellite Navigation and Location Service,Guilin University of Electronic Technology,Guilin Guangxi 541004,China)

机构地区：[1]桂林电子科技大学广西精密导航技术与应用重点实验室,广西桂林541004 [2]桂林电子科技大学卫星导航与位置服务国家与地方联合工程研究中心,广西桂林541004

出　　处：《计算机仿真》2024年第9期167-172,共6页Computer Simulation

基　　金：国家自然科学基金(61861008,62061010,62161007);桂林市科技局桂林市国家可持续发展议程创新示范区建设重点项目(20190219-1)。

摘　　要：由于在森林火灾初期火灾目标的尺寸较小,导致模型对小目标的森林火灾产生漏检的问题,提出一种基于YOLOv5的小目标检测的森林火灾识别方法。为了提高检测算法对小型火焰的检测效果,提出了一个新的模块并将其命名为CSP-SPPFP(Cross Stage Partial-Spatial Pyramid Pooling Fast Plus),将YOLOv5主干网络中的SPP替换为CSP-SPPFP;为了增强火焰特征的表达,提出一种基于CBAM注意力的模块;为减少小目标的上采样过程中的信息损失,将原有的上采样方法替换为转置卷积。设计1组实验来确定选择使用哪种CBAM注意力模块、1组消融实验以及1组对比实验来验证所提出的算法的有效性。实验结果表明,算法与原YOLOv5算法相比,mAP@50提升了2.12%,针对小目标森林火灾目标能够准确检出,有效提升了森林火灾防范能力。Due to the small size of fire targets in the early stage of forest fires,which leads to the problem that the model misses the detection of forest fires with small targets,this paper proposes a forest fire identification method based on YOLOv5 for small target detection.In order to improve the detection algorithm for small flames,a new module is proposed and named CSP-SPPFP(Cross Stage Partial-Spatial Pyramid Pooling Fast Plus),replacing SPP in the YOLOV5 backbone network with CSP-SPPFP.To enhance the representation of flame features,a module based on CBAM attention is proposed.To reduce the information loss during the upsampling of small targets,the original upsampling method is replaced with transposed convolution.In this paper,one set of experiments is designed to determine which CBAM attention module to use,one set of ablation experiments,and one set of comparison experiments to verify the effectiveness of the proposed algorithm.The experimental results show that the algorithm in this paper improves the mAP@50 by 2.12% compared with the original YOLOv5 algorithm,and can accurately detect forest fire targets for small targets,which effectively improves forest fire prevention capability.

关 键 词：森林火灾 注意力机制 小目标检测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]