基于深度学习及模糊层次分析的毁伤评估算法

Damage assessment algorithm based on deep learning and fuzzy analytic hierarchy process

针对军事目标的毁伤检测和毁伤评估问题,提出一种基于深度学习及模糊层次分析法的双阶段毁伤评估方法。首先在带有坐标注意力机制的Yolov5双层目标检测子系统中,通过关键区域提取机制,利用基于交并比和匈牙利线性匹配和决策树相结合的毁伤部件分类器,进行毁伤部件的分类和毁伤程度的量化处理。而后在基于三角模糊层次分析的毁伤评估子系统中,通过设计多种毁伤评估权重体系和毁伤树判据,对前一阶段提取到的毁伤特征和类别进行综合考虑,实现对目标的在线实时毁伤评估,实验结果表明,在多种干扰因素的仿真数据集下,带有注意力机制的Yolov5双层目标检测子系统相较于经典目标细粒度识别算法多种部件和毁伤的平均检测准确率提升了3.6%以上,在目标关键区域提取、毁伤分类和评估表现出了更好的性能,为目标毁伤评估和军事作战决策提供了有力的支持和参考。

To address the detection and assessment of military target damage,this paper proposes a two-stage damage assessment method based on deep learning and fuzzy analytic hierarchy process.Firstly,in the Yolov5-based dual-stage target detection subsystem with coordinate attention mechanism,a key region extraction mechanism is employed to extract damage components,and a damage component classifier based on the combination of intersection over union,Hungarian linear matching,and decision tree is used for classification and quantification of damage severity.Then,in the damage assessment subsystem based on the triangular fuzzy analytic hierarchy process,multiple damage assessment weight systems and damage tree criteria are designed to comprehensively consider the damage features and categories extracted from the previous stage,achieving online real-time damage assessment for targets.Experimental results show that on various simulated datasets with multiple interference factors,the Yolov5-based dualstage target detection subsystem with the attention mechanism achieves an average improvement in detection accuracy of over 3.6% when compared to classical target fine-grained recognition algorithms.It demonstrates better performance in target key region extraction,and damage classification and assessment,providing strong support and reference for target damage assessment and military operation decision making.