基于知识图谱的红外目标部件识别

打开文本图片集

关键词：红外目标识别;知识图谱；整体识别;全局向量模型；注意力模块;学习率控制中图分类号：TP391.4文献标识码：Adoi：10.37188/OPE.20253309.1446 CSTR：32169.14.OPE.20253309.1446

Knowledge graph-based method for infrared target component recognition

LIUHaiyi，LIZhengzhou*，LIAoran，LIUHaitao （School ofMicroelectronics and Communication Engineering，Chongqing University， Chongqing 401331， China） * Corresponding author， E -mail： lizhengzhou@cqu. edu. cn

Abstract：To address challenges in component recognition，such as self-occlusion，unclear visual features，and substantial feature variation due to distance，a novel infrared target component recognition method based on a knowledge graph is proposed. This method utilizes a whole-to-component prediction （WCP） strategy to sequentially recognize target components. Initially，the overalltarget is detected，followed by an expansion of the target region to high resolution to enhance signal details. Subsequently，a component-related attention module （CAM） integrated with the knowledge graph exploits structural relationships among parts to infer visible interconnections and employs attention mechanisms to improve recognition performance，thereby mitigating issues arising from ambiguous visual features. For components affected by self-occlusion，a Self-Occlusion Learning Rate Decay（SLD） control strategy，based on self-removal capability，is introduced to strengthen the model's capacity to learn from occlusions and facilitate convergence. Validation is conducted using an indoor target equivalence scaling system，employing roombased models across various orientations and distances，with aircraft tested under diverse conditions， achieving an average precision of 92.2% . Experimental results demonstrate that the proposed method surpasses existing approaches in component recognition accuracy and recall，markedly enhancing both precision and recall metrics.

Key words： infrared target recognition；knowledge graph；whole to component predict； atention module；self-occlusion learning rate decay

1引言

红外图像中目标部件识别能够更精确地理解和分析其结构，有利于精准定姿和定位，在工业领域和军事应用中具有重要作用[1]。（剩余12352字）