杨棉绒, 牛丽平. 基于LGBM的Zernike特征选取及红外图像目标识别方法[J]. 红外与激光工程, 2022, 51(4): 20210309. DOI: 10.3788/IRLA20210309
引用本文: 杨棉绒, 牛丽平. 基于LGBM的Zernike特征选取及红外图像目标识别方法[J]. 红外与激光工程, 2022, 51(4): 20210309. DOI: 10.3788/IRLA20210309
Yang Mianrong, Niu Liping. Zernike’s feature selection based on LGBM and identification methods of infrared image target[J]. Infrared and Laser Engineering, 2022, 51(4): 20210309. DOI: 10.3788/IRLA20210309
Citation: Yang Mianrong, Niu Liping. Zernike’s feature selection based on LGBM and identification methods of infrared image target[J]. Infrared and Laser Engineering, 2022, 51(4): 20210309. DOI: 10.3788/IRLA20210309

基于LGBM的Zernike特征选取及红外图像目标识别方法

Zernike’s feature selection based on LGBM and identification methods of infrared image target

  • 摘要: 红外传感技术有效解决了夜间观测的难题,成为现代战场侦察的重要手段之一。不断提升基于红外图像的目标识别能力是实施精确打击、态势感知的有力途径。针对红外图像识别问题,提出基于轻量级梯度提升机(Light Gradient Boosting Machine, LGBM)的Zernike特征选取算法,并结合稀疏表示分类器(Sparse Representation-based Classification, SRC)完成目标类别确认。首先,基于红外图像中的目标区域提取多阶Zernike矩特征,表征待识别目标的本质特性;其次,采用LGBM特征选择算法对多阶矩特征进行二次筛选,减少冗余的同时提高特征的针对性;最后,基于SRC对最终选择的Zernike矩特征矢量进行分类。该方法通过LGBM的特征选择有效提高了最终特征的有效性,同时降低了分类的计算复杂度,有利于提高整体识别性能。采用公开的中波红外目标图像数据集(MWIR)开展验证实验,对10类典型军事目标进行区分识别。实验分别在原始样本、噪声干扰样本以及部分缺失样本三种条件下进行并与几类现有红外目标识别方法进行对比讨论。结果表明:所提方法可取得更优性能,证明其有效性。

     

    Abstract: Infrared sensing technology effectively handle the problem of night observation, which is becoming an important measure for battlefield reconnaissance. Continuously improving the ability of target recognition based on infrared images was a powerful way to implement precision strikes and situational awareness. Aiming at the problem of infrared image recognition, a Zernike feature selection algorithm based on Light Gradient Boosting Machine (LGBM) was proposed, combined with Sparse Representation-based Classification (SRC) to complete the target category confirmation. Firstly, based on the target area in the infrared image, multi-order Zernike moment features were extracted to characterize the essential characteristics of the target to be recognized; Secondly, the LGBM feature selection algorithm was used to screen the multi-order moment features twice to reduce redundancy and improve the pertinence of features; Finally, the final selected Zernike moment feature vector was classified based on SRC. The method effectively improves the effectiveness of the final features through the feature selection of LGBM, at the same time reduces the computational complexity of classification, which was beneficial to improve the overall recognition performance. The publicly available mid-wave infrared target image data set was used to carry out verification experiments to distinguish and identify 10 types of typical military targets. The experiment was carried out under the three conditions of original samples, noise interference samples and partially missing samples, and compared with several types of existing infrared target recognition methods. The results show that the proposed method can achieve better performance and prove its effectiveness.

     

/

返回文章
返回