Hyperspectral RX Anomaly Detection Algorithm with Visual Attention Mechanism
-
摘要: 视觉注意机制具有快速引导关注到重点区域的特性，将其引入高光谱图像异常检测中具有可行性。本文从采样方式、波段选取、融入局部光谱特征3方面构建更适用于计算高光谱图像显著性的视觉注意机制模型。针对经典的基于高斯统计分布假设的RX异常检测算法在背景参数估计中易受潜在异常干扰的问题，利用视觉显著性结果对原图像进行高斯加权，在加权后图像中进行背景均值与协方差的重新估算，进而使用更精确的背景参数对原图像进行RX异常检测。在5个经典数据上的实验结果表明，本文方法有效地表现了潜在的异常目标，改进的RX异常检测算法具有更高的检测精度与更低的虚警率。Abstract: A visual attention mechanism (VAM) can quickly highlight region-of-interest targets; therefore, it is reasonable to introduce visual attention into hyperspectral image (HSI) anomaly detection tasks. By adjusting a bottom-up VAM model in three aspects, namely sampling method, band selection, and local spectral features, a more applicable VAM model for calculating the saliency of hyperspectral images was constructed. The resulting VAM is called bottom-up hyperspectral saliency map (BUHS). To solve the problem of background parameter estimation in the RX(Reed-Xiaoli) algorithm, which is susceptible to interference, BUHS was used as a Gaussian weighting parameter for the original image, in which new parameters of the RX anomaly method were calculated. Compared to the traditional RX, the background parameters are more accurate. The experimental results on five HSI datasets show that the proposed method can effectively identify potential anomaly targets and improve the RX algorithm with a higher detection accuracy and lower false alarm rate.
-
Key words:
- hyperspectral image /
- anomaly detection /
- visual attention mechanism /
- saliency /
- RX
-
图 1 融合视觉注意机制的高光谱RX异常检测算法流程
Figure 1. Hyperspectral RX anomaly detection algorithm with visual attention mechanism flowchart
图 2 高光谱影像与其对应异常目标分布
Figure 2. Hyperspectral images and their anomaly targets distribution
表 1 LCVF模拟异常目标
Table 1. LCVF simulates anomalous targets
Column Number of targets/cells Proportion of target Proportion of background 1 5/20 0.8 0.2 2 5/20 0.6 0.4 3 5/5 0.4 0.6 4 6/6 0.2 0.8 
下载: 导出CSV
表 2 异常检测数据介绍
Table 2. Anomaly detection data introduction
Data Sensor Resolution/m Size Targert type Number of targets/cells LCVF AVIRIS 20 120×160×190 Simulated anomalous targets 21/51 airport4 AVIRIS 3.4 100×100×205 Airport 3/60 beach4 ROSIS-03 1.3 150×150×102 Car and bare soil 7/68 urban1 AVIRIS 17.2 100×100×204 Building 9/67 urban3 AVIRIS 3.5 100×100×191 Ship 11/52
下载: 导出CSV
表 3 参数实验结果
Table 3. Parameters experimental results
Parameter Size AUC Parameter Size AUC Number of bands 3 0.9759 Number of pyramid layers 3 0.9081 6 0.9621 5 0.9759 9 0.9681 7 0.9709 12 0.9777 9 0.9624 15 0.9779 18 0.9772 Spectral feature window size 3×3 0.9769 21 0.9765 5×5 0.9759 24 0.9746 7×7 0.975 27 0.9726 30 0.9725
下载: 导出CSV
表 4 消融实验
Table 4. Ablation experiment
model Sampling mode Band selection Spectral feature AUC A 0.9625 B √ 0.9690 C √ 0.9788 D √ 0.9759 E √ √ 0.9846 F √ √ 0.9831 G √ √ √ 0.9862
下载: 导出CSV
表 5 AUC统计结果
Table 5. AUC statistics
RX ITTIRX HITTIRX LCVF 0.8333 0.8753 0.9999 Airport4 0.9526 0.8768 0.9894 Beach4 0.9588 0.9302 0.9847 Urban1 0.9907 0.9881 0.9983 Urban3 0.9513 0.9625 0.9862
下载: 导出CSV
-
[1] 赵春晖, 李晓慧, 王玉磊. 高光谱图像异常目标检测研究进展[J]. 电子测量与仪器学报, 2014, 28(8): 803-811. doi: 10.13382/j.jemi.2014.08.001ZHAO Chunhui, LI Xiaohui, WANG Yulei. Research progress of hyperspectral image anomaly target detection[J]. Journal of Electronic Measurement and Instrumentation, 2014, 28(8): 803-811. doi: 10.13382/j.jemi.2014.08.001 [2] Reed I S, YU X. Adaptive multiple-band CFAR detection of an optical pattern with unknown spectral distribution[J]. IEEE Transactions on Acoustics Speech & Signal Processing, 1990, 38(10): 1760-1770. [3] Ashton E A, Schaum A. Algorithms for the detection of sub-pixel targets in multispectral imagery[J]. Photogrammetric Engineering & Remote Sensing, 1998, 64(7): 723-731. [4] Taitano Y, Geier B, Bauer K. A locally adaptable iterative RX detector[J]. EURASIP Journal on Advances in Signal Processing, 2010, 2010(1): 341908. doi: 10.1155/2010/341908 [5] Molero J M, Garzón E M, García I, et al. Analysis and optimizations of global and local versions of the RX algorithm for anomaly detection in hyperspectral data[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2013, 6(2): 801-814. doi: 10.1109/JSTARS.2013.2238609 [6] Kwon H, Nasrabadi N M. Kernel RX-algorithm: a nonlinear anomaly detector for hyperspectral imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2005, 43(2): 388-397. doi: 10.1109/TGRS.2004.841487 [7] 徐超, 詹天明. 基于低秩全变差正则化的高光谱异常检测方法[J]. 计算机科学与探索, 2020, 14(12): 2140-2149. doi: 10.3778/j.issn.1673-9418.2002003XU Chao, ZHAN Tianming. Hyperspectral anomaly detection method based on low-rank full-variance regularization[J]. Computer Science and Exploration, 2020, 14(12): 2140-2149. doi: 10.3778/j.issn.1673-9418.2002003 [8] CHEN Yi, Nasrabadi N M, Tran T D. Sparse representation for target detection in hyperspectral imagery[J]. IEEE Journal of Selected Topics in Signal Processing, 2011, 5(3): 629-640. doi: 10.1109/JSTSP.2011.2113170 [9] SUN Weiwei, LIU Chun, LI Jialin, et al. Low-rank and sparse matrix decomposition-based anomaly detection for hyperspectral imagery[J]. Journal of Applied Remote Sensing, 2014, 8(1): 083641. doi: 10.1117/1.JRS.8.083641 [10] XU Yang, WU Zebin, LI Jun, et al. Anomaly detection in hyperspectral images based on low-rank and sparse representation[J]. IEEE Transactions on Geoscience & Remote Sensing, 2016, 54(4): 1990-2000. [11] LI Wei , DU Qian. Collaborative representation for hyperspectral anomaly detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(3): 1463-1474. doi: 10.1109/TGRS.2014.2343955 [12] 沈旭, 程小辉, 王新政. 结合视觉注意力机制基于尺度自适应局部对比度增强的红外弱小目标检测算法[J]. 红外技术, 2019, 41(8): 764-771. http://hwjs.nvir.cn/article/id/hwjs201908012SHEN Xu, CHENG Xiaohui, WANG Xinzheng. Combined visual attention mechanism based on scale adaptive local contrast enhancement for infrared weak target detection algorithm[J]. Infrared Technology, 2019, 41(8): 764-771. http://hwjs.nvir.cn/article/id/hwjs201908012 [13] Itti L, Koch C, Niebur E. A model of saliency-based visual attention for rapid scene analysis[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1998, 20(11): 1254-1259. doi: 10.1109/34.730558 [14] LIANG Jie, ZHOU Jun, BAI Xiao, et al. Salient object detection in hyperspectral imagery[C]//IEEE International Conference on Image Processing, 2014: 2393-2397. [15] CAO Yan, ZHANG Jing, TIAN Qi, et al. Salient target detection in hyperspectral images using spectral saliency[C]//IEEE China Summit and International Conference on Signal and Information Processing, 2015: 1086-1090. [16] 刘嘉诚, 王爽, 刘伟华, 等. 显著性权重RX高光谱异常点检测[J]. 遥感学报, 2019, 23(3): 418-430. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXB201903005.htmLIU Jiacheng, WANG Shuang, LIU Weihua, et al. Significance weighting RX hyperspectral anomaly detection[J]. Journal of Remote Sensing, 2019, 23(3): 418-430. https://www.cnki.com.cn/Article/CJFDTOTAL-YGXB201903005.htm [17] ZHAO Minghua, YUE Liqin, HU Jing, et al. Salient target detection in hyperspectral image based on visual attention[J/OL]. IET Image Processing, 2021, 15(11): https://doi.org/10.1049/ipr2.12197 .[18] XIANG Pei, SONG Jiangluqi, QIN Hanlin, et al. Visual attention and background subtraction with adaptive weight for hyperspectral anomaly detection[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 2270-2283. doi: 10.1109/JSTARS.2021.3052968 [19] ZHU Wei, LIU Jian, ZHU Mingyue, et al. Research on improved algorithm of DR image enhancement based on Gauss-Laplacian pyramid[J]. Chinese Journal of Medical Instrumentation, 2019, 43(1): 10-13. doi: 10.3969/j.issn.1671-7104.2019.01.003 [20] 纪磊, 张欣, 张丽梅, 等. 基于空谱加权近邻的高光谱图像分类算法[J]. 激光与光电子学进展, 2020, 57(6): 165-172. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202006017.htmJI Lei, ZHANG Xin, ZHANG Limei, et al. Hyperspectral image classification algorithm based on null-spectrum weighted nearest neighbor[J]. Advances in Lasers and Optoelectronics, 2020, 57(6): 165-172. https://www.cnki.com.cn/Article/CJFDTOTAL-JGDJ202006017.htm [21] WANG Qi, ZHANG Fahong, LI Xuelong. Optimal clustering framework for hyperspectral band selection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(10): 5910-5922. [22] 叶珍, 白璘, 何明一. 高光谱图像空谱特征提取综述[J]. 中国图象图形学报, 2021, 26(8): 1737-1763. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB202108003.htmYE Zhen, BAI Lin, HE Meiyi. A review of hyperspectral image null spectrum feature extraction[J]. Chinese Journal of Graphics, 2021, 26(8): 1737-1763. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGTB202108003.htm [23] 汪家宝, 陈树新, 吴昊, 等. 采用平方根高斯核积分滤波的目标跟踪算法[J]. 西安交通大学学报, 2022, 56(4): 157-164. https://www.cnki.com.cn/Article/CJFDTOTAL-XAJT202204017.htmWANG Jiabao, CHEN Shuxin, WU Hao, et al. A target tracking algorithm using square root Gaussian kernel integral filtering[J]. Journal of Xi'an Jiaotong University, 2022, 56(4): 157-164. https://www.cnki.com.cn/Article/CJFDTOTAL-XAJT202204017.htm [24] 黄远程, 钟燕飞, 赵野鹤, 等. 联合盲分解与稀疏表达的高光谱图像异常目标检测[J]. 武汉大学学报: 信息科学版, 2015, 40(9): 1144-1150. https://www.cnki.com.cn/Article/CJFDTOTAL-WHCH201509002.htmHUANG Yuancheng, ZHONG Yanfei, ZHAO Yehe, et al. Joint blind decomposition and sparse expression for hyperspectral image anomaly target detection[J]. Journal of Wuhan University: Information Science Edition, 2015, 40(9): 1144-1150. https://www.cnki.com.cn/Article/CJFDTOTAL-WHCH201509002.htm [25] KANG Xudong, ZHANG Xiangping, LI Shutao , et al. Hyperspectral anomaly detection with attribute and edge-preserving filters[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017(10): 1-12. [26] 成宝芝, 杨桂花, 王凤嫔, 等. 基于低秩张量分解的高光谱RX异常目标检测算法[J]. 光学技术, 2022, 48(3): 379-384. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJS202203020.htmCHENG Baozhi, YANG Guihua, WANG Fengpin, et al. Hyperspectral RX anomaly target detection algorithm based on low-rank tensor decomposition[J]. Optical Technology, 2022, 48(3): 379-384. https://www.cnki.com.cn/Article/CJFDTOTAL-GXJS202203020.htm -
点击查看大图
图(4) / 表(5)
计量
- 文章访问数: 167
- HTML全文浏览量: 39
- PDF下载量: 40
- 被引次数: 0
