基于 SSA-GRNN 的非接触式目标表面红外激光物性反演方法  

Non-contact infrared laser physical property inversion method for target surface based on SSA-GRNN

作　　者：李荣华[1,2] 周心晨 翁传欣 薛豪鹏 吴锦龙 林宸宇 LI Ronghua;ZHOU Xinchen;WENG Chuanxin;XUE Haopeng;WU Jinlong;LIN Chenyu(Institute of Mechanical Engineering,Dalian Jiaotong University,Dalian 116028,China;Advanced Robot Sensing and Control Technology Innovation Center,Dalian 116028,China;Beijing Institute of Environmental Characteristics,Beijing 100854,China)

机构地区：[1]大连交通大学机械工程学院,辽宁大连116028 [2]大连市先进机器人感知与控制技术创新中心,辽宁大连116028 [3]北京环境特性研究所,北京100854

出　　处：《红外与激光工程》2024年第10期161-172,共12页Infrared and Laser Engineering

基　　金：大连市高层次人才创新支持计划项目(2022RJ03)。

摘　　要：在目标物性反演时,接触式测量方法在复杂环境下进行存在困难,而非接触式测量方法,由于测量数据相比接触式测量存在一定的误差,导致反演准确率受到影响。针对以上问题,提出了一种基于红外激光回波的非接触式目标表面物性反演方法。首先,测量不同目标表面的激光回波强度信息,采用麻雀搜索算法,优化并训练广义回归神经网络,建立红外激光回波强度预测模型;其次,分析测量距离、测量角度对激光回波强度的影响,建立材料表面激光回波强度数据库;最后,采集未知目标在四种距离下的表面激光回波强度信息,赋予材料种类编号,输入到回波强度预测模型中,计算预测值与实际值的相对误差,反演未知目标表面材料物性。实验结果表明:在反演目标相同的情况下,回波强度预测结果的均方根误差从传统网络的11.337降低到了优化后的2.482。优化后的神经网络模型的相对反演准确率可达88.89%以上,与传统方法相比,平均反演准确率提高了45.83%,文中所提方法具有较高的准确性和推广性,为武器系统非合作目标的探测、材料反演提供方法,提高了目标识别能力。Objective Different targets have different material parameters on their surfaces.In the physical property inversion of targets,the contact measurement method is difficult to be carried out in complex environments,while the non-contact measurement method,due to certain errors in the measurement data compared with the contact measurement,causes the inversion accuracy to be affected.Therefore,it is necessary to propose a surface physical property inversion method for non-contact targets.Methods In this paper,a non-contact target surface physical property inversion method of infrared laser echo is proposed(Fig.1).The laser echo intensity measurement system is built(Fig.7).First,six materials(Fig.4)and seven measurement distances were selected.Through the 4.6μm infrared laser transmitter,the laser is launched to the material at a certain distance away,and after the reflection of the material surface,the laser echo intensity information is collected by the receiver to establish a database of the laser echo intensity on the material surface;second,the SSA-GRNN neural network is used to obtain the prediction model of the laser echo intensity on the material surface;lastly,the echo intensity information of the unknown material is measured,and by assigning the material Finally,the echo intensity information of the unknown material is measured and input into the prediction model by assigning the material type,calculating the error between the predicted echo intensity value and the real value,and obtaining the material number with the smallest error to invert the material properties of the unknown target surface.Results and Discussions The measured echo intensity data were used to train the SSA-GRNN echo intensity prediction model,and the model established by the SSA-GRNN generalized regression neural network not only has strong generalizability,but also has high accuracy.The echo intensity data on the surface of the unknown target at five distances are measured(Tab.3),and the predicted values of echo intensity as well as t

关 键 词：红外激光 回波强度 SSA-GRNN神经网络 物性反演 

分 类 号：TN249[电子电信—物理电子学]