改进YOLOv8的建筑物毁伤效果评估算法  

Improved YOLOv8 Algorithm for Building Damage Effect Evaluation

作　　者：沈先耿 王鑫 刘晓阳 SHEN Xiangeng;WANG Xin;LIU Xiaoyang(Officers College of PAP,Chengdu 610213,China)

机构地区：[1]武警警官学院,四川成都610213

出　　处：《信息工程大学学报》2025年第2期154-160,共7页Journal of Information Engineering University

基　　金：武警部队军事理论课题(WJJY24JL0141)。

摘　　要：针对现代战争中战场环境复杂和传统评估方法过于依赖主观经验的问题,提出一种改进YOLOv8的建筑物毁伤效果评估算法。首先,在样本输入端采用Mosaic-9进行图像预处理,提升网络模型的泛化能力。其次,在骨干网络中引入高效多尺度注意力机制(EMA),同时利用新构造的快速跨阶段局部网络融合模块(FC2f)模块,提升网络模型的特征提取能力和运行效率。最后,采用Scylla交并比损失函数(SIoULoss)对网络损失函数进行优化,进一步提升网络模型的检测评估精度和运行效率。实验结果表明,该算法检测评估精度较高,运算速度较快,具有一定的军事应用价值。Aiming at the problems of complex battlefield environment and excessive reliance on subjective experience in traditional evaluation methods in modern war,an improved YOLOv8 algorithm for building damage effect evaluation is proposed.Firstly,the Mosaic-9 is used for image preprocessing at the sample input to improve the generalization ability of the network model.Secondly,the efficient multi-scale attention(EMA)mechanism is introduced into the backbone network,and the newly constructed fast cross stage partial network fusion(FC2f)is used to improve the feature extraction ability and operation efficiency of the network model.Finally,the Scylla intersection over union loss(SIoU Loss)function is used to optimize the network loss function to further improve the detection and evaluation accuracy and operation efficiency of the network model.Experimental results show that this algorithm has high detection and evaluation accuracy,fast operational speed,and certain military application value.

关 键 词：改进YOLOv8 建筑物 毁伤评估 图像预处理 损失函数优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]