基于超像素注意力和孪生结构的半监督高光谱显著性目标检测

doi:10.12382/bgxb.2022.1162

摘要/Abstract

摘要：

高光谱显著性目标检测技术在伪装识别、异常检测等领域展现了惊人的潜力,并得到了广泛的关注。基于深度学习技术的神经网络模型克服了传统算法检测精度低、鲁棒性弱的缺点,但是数据标注成本限制了其进一步发展。为此提出了一种超像素注意力孪生半监督算法,使用少量全监督数据和大量弱监督数据进行训练,有效降低了标注成本。该算法由孪生预测模块和注意力辅助模块组成,其中孪生预测模块捕获弱标签隐式约束并生成显著性结果图,注意力辅助模块利用超像素级通道注意力机制优化预测结果。新提出的超像素注意力孪生半监督算法在高光谱数据集上实现了87%的检测精度,优于其他流行算法,在有效降低标注成本的同时具有优异的显著性检测性能。

关键词: 高光谱显著性目标检测, 半监督训练, 孪生结构, 超像素注意力机制, 深度学习

Abstract:

Hyperspectral salient object detection technology plays a key role in various fields, such as camouflage recognition and anomaly detection, thus having received extensive attention. The neural network model based on deep learning technology has improved issues such as low detection accuracy and weak robustness of traditional algorithms, but the cost of data labeling limits its further development. To this end, a superpixel attention siamese semi-supervised algorithm is proposed, which uses a small amount of fully supervised data and a large amount of weakly supervised data for training, effectively reducing annotation costs. The algorithm consists of a siamese prediction module and an attention assistance module. The siamese prediction module captures the implicit constraints of weak labels and generates a saliency result map, while the attention assistance module optimizes the prediction results with a superpixel-level channel attention mechanism. The newly proposed semi-supervised algorithm achieves a detection accuracy of 87% on hyperspectral datasets, outperforming other popular algorithms and demonstrating excellent saliency detection performance while effectively reducing annotation costs.

Key words: hyperspectral salient object detection, semi-supervised training, siamese structure, superpixel attention mechanism, deep learning

中图分类号:

TJ301

秦昊林, 许廷发, 李佳男. 基于超像素注意力和孪生结构的半监督高光谱显著性目标检测[J]. 兵工学报, 2023, 44(9): 2639-2649.

QIN Haolin, XU Tingfa, LI Jianan. Semi-supervised Hyperspectral Salient Object Detection Using Superpixel Attention and Siamese Structure[J]. Acta Armamentarii, 2023, 44(9): 2639-2649.

图/表 13

参考文献 32

[1]	WANG L, HUA G, SUKTHANKAR R, et al. Video object discovery and co-segmentation with extremely weak supervision[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(10): 2074-2088. doi: 10.1109/TPAMI.2016.2612187 pmid: 28113741
[2]	MENG F M, LI H L, LIU G, et al. Object co-segmentation based on shortest path algorithm and saliency model[J]. IEEE Transactions on Multimedia, 2012, 14:1429-1441. doi: 10.1109/TMM.2012.2197741 URL
[3]	HUNG T Y, VAIKUNDAM S, NATARAJAN V, et al. Phase fourier reconstruction for anomaly detection on metal surface using salient irregularity[C]//Proceedings of Conference on Multimedia Modeling.Berlin,Germany:Springer, 2017:290-302.
[4]	CAO Y, ZHANG J, TIAN Q, et al. Salient target detection in hyperspectral images using spectral saliency[C]//Proceedings of IEEE China Summit & International Conference on Signal & Information Processing. Chengdu, China:IEEE, 2015.
[5]	王成龙, 赵倩, 赵琰. 基于深度可分离卷积的实时遥感目标检测算法[J]. 电光与控制, 2022, 29(8): 45-49.
	WANG C L, ZHAO Q, ZHAO Y. Areal-time remote separable convolution[J]. Electronics Optics & Control, 2022, 29(8): 45-49. (in Chinese)
[6]	马梁, 苟于涛, 雷涛, 等. 基于多尺度特征融合的遥感图像小目标检测[J]. 光电工程, 2022, 49(4): 210363.
	MA L, GOU Y T, LEI T, et al. Small object detection based on multi-scale feature fusion using remote sensing images[J]. Opto-Electron Eng, 2022, 49(4): 210363. (in Chinese)
[7]	于博文, 吕明. 改进的YOLOv3算法及其在军事目标检测中的应用[J]. 兵工学报, 2022, 43(2): 345-354. doi: 10.3969/j.issn.1000-1093.2022.02.012
	YU B W, LÜ M. Improved YOLOv3 algorithm and its application in military target detection[J]. Acta Armamentarii, 2022, 43(2): 345-354. (in Chinese) doi: 10.3969/j.issn.1000-1093.2022.02.012
[8]	WANG W G, ZHAO S Y, SHEN J B, et al. Salient object detection with pyramid attention and salient edges[C]// Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Long Beach, CA, US:IEEE, 2019.
[9]	BEARMAN A L, RUSSAKOVSKY O, FERRARI V, et al. What's the point: semantic segmentation with point supervision:arXiv:1506.02106[R]. Ithaca,NY, US: Cornell University, 2021:1506.02106.
[10]	ZHANG L, ZHANG Y N, YAN H Q, et al. Salient object detection in hyperspectral imagery using multi-scale spectral-spatial gradient[J]. Neurocomputing, 2018, 291:215-225. doi: 10.1016/j.neucom.2018.02.070 URL
[11]	ITTI L, KOCH C, NIEBUR E. A model of saliency-based visual attention for rapid scene analysis[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 1998, 20(11): 1254-1259.
[12]	MOAN S L, MANSOURI A, HARDEBERG J Y, et al. Saliency in spectral images[J]. Scandinavian Conference on Image Analysis, 2011: 114-123.
[13]	LIANG J, ZHOU J, BAI X, et al. Salient object detection in hyperspectral imagery[C]//Proceedings of International Conference on Image Processing. Melbourne, VIC, Australia:IEEE, 2013.
[14]	IMAMOGLU N, DING G Q, FANG Y M, et al. Salient object detection on hyperspectral images using features learned from unsupervised segmentation task[C]//Proceedings of ICASSP 2019- 2019 IEEE International Conference on Acoustics, Speech and Signal Processing. Brighton, UK:IEEE,2019.
[15]	HUANG C, XU T F, ZHANG Y H, et al. Salient object detection on hyperspectral images in wireless network using CNN and saliency optimization[J]. Ad Hoc Networks, 2021, 112(8):102369. doi: 10.1016/j.adhoc.2020.102369 URL
[16]	PAN C, LIU J F, YAN W Q, et al. Salient object detection based on visual perceptual saturation and two-stream hybrid networks[J]. IEEE Transactions on Image Processing: a Publication of the IEEE Signal Processing Society, 2021, 30:4773-4787. doi: 10.1109/TIP.2021.3074796 URL
[17]	ZHOU Y, HUO S W, XIANG W, et al. Semi-supervised salient object detection using a linear feedback control system model[J]. IEEE Transactions on Cybernetics, 2019, 49(4):1173-1185. doi: 10.1109/TCYB.2018.2793278 pmid: 29993850
[18]	ZHANG Q J, CONG R M, LI C Y, et al. Dense attention fluid network for salient object detection in optical remote sensing images[J]. IEEE Transactions on Image Processing, 2021, 30:1305-1317. doi: 10.1109/TIP.2020.3042084 pmid: 33306467
[19]	LIN Z Y, LI P T, HUANG W J, et al. Fixed feature with attention based on recurrent criss-cross salient object detection network[C]//Proceedings of the 2020 8th International Conference on Digital Home.Dalian, China:IEEE, 2020.
[20]	QIN X B, ZHANG Z C, HUANG C Y, et al. U2-net: going deeper with nested U-structure for salient object detection[J]. Pattern Recognition, 2020, 106:107404. doi: 10.1016/j.patcog.2020.107404 URL
[21]	WANG Y D, ZHANG J, KAN M N, et al. Self-supervised equivariant attention mechanism for weakly supervised semantic segmentation[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, US:IEEE, 2020.
[22]	JAMPANI V, SUN D Q, LIU M Y, et al. Superpixel Sampling Networks[C]//Proceedings of European Conference on Computer Vision.Beilin,Germany:Springer, 2018.
[23]	QIN X B, ZHANG Z C, HUANG C Y, et al. BASNet: boundary-aware salient object detection[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA, US:IEEE, 2019.
[24]	YAN H Q, ZHANG Y N, WEI W, et al. Salient object detection in hyperspectral imagery using spectral gradient contrast[C]//Proceedings of IGARSS 2016- 2016 IEEE International Geoscience and Remote Sensing Symposium. Beijing, China: IEEE, 2016.
[25]	TONG N, LU H C, RUAN X, et al. Salient object detection via bootstrap learning[C]//Proceedings of IEEE Conference on Computer Vision & Pattern Recognition. Boston, MA, US:IEEE, 2015:1884-1892.
[26]	TONG N, LU H C, ZHANG L H, et al. Saliency detection with multi-scale superpixels[J]. IEEE Signal Processing Letters, 2014, 21(9):1035-1039. doi: 10.1109/LSP.2014.2323407 URL
[27]	LI H Y, LU H C, LIN Z, et al. Inner and inter label propagation: salient object detection in the wild[J]. IEEE Transactions on Image Processing, 2015, 24(10):3176-3186. doi: 10.1109/TIP.2015.2440174 pmid: 26054065
[28]	YANG C, ZHANG L H, LU H C, et al. Saliency detection via graph-based manifold ranking[C]//Proceedings of Computer Vision & Pattern Recognition.Portland, OR, US: IEEE, 2013.
[29]	ZHU W J, LIANG S, WEI Y C, et al. Saliency optimization from robust background detection[C]//Proceedings of IEEE Conference on Computer Vision & Pattern Recognition.Columbus, OH, US:IEEE, 2014:2814-2821.
[30]	ZHANG J M, SCLAROFF S, LIN Z, et al. Minimum barrier salient object detection at 80 FPS[C] //Proceedings of IEEE International Conference on Computer Vision. Washington,D.C., US: IEEE, 2015.
[31]	TU W C, HE S F, YANG Q X, et al. Real-time Salient object detection with a minimum spanning Tree[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas, NV, US: IEEE, 2016.
[32]	ZHOU Y, HUO S W, XIANG W, et al. Semi-supervised salient object detection using a linear feedback control system model[J]. IEEE Transactions on Cybernetics, 2019, 49(4):1173-1185. doi: 10.1109/TCYB.2018.2793278 pmid: 29993850

算法	MAE	F_avg	F_max	AUC	NSS
GS算法	0.1787	0.1709	0.2703	0.6191	0.9505
SAD算法	0.1920	0.1855	0.2987	0.5967	0.8377
SED算法	0.1255	0.1273	0.2970	0.5826	0.7203
SED-GS算法	0.1856	0.1545	0.3078	0.6051	0.8836
SED-SAD算法	0.2023	0.1818	0.2998	0.6337	1.2311
SUDF算法	0.1687	0.5451	0.5472	0.8685	1.9627
U2Net^†算法	0.1122	0.6525	0.6599	0.8569	1.5000
本文算法	0.1380	0.6482	0.6508	0.8703	1.3611

算法	MAE	F_avg	F_max	AUC	NSS
GS算法	0.1787	0.1709	0.2703	0.6191	0.9505
SAD算法	0.1920	0.1855	0.2987	0.5967	0.8377
SED算法	0.1255	0.1273	0.2970	0.5826	0.7203
SED-GS算法	0.1856	0.1545	0.3078	0.6051	0.8836
SED-SAD算法	0.2023	0.1818	0.2998	0.6337	1.2311
SUDF算法	0.1687	0.5451	0.5472	0.8685	1.9627
U2Net^†算法	0.1122	0.6525	0.6599	0.8569	1.5000
本文算法	0.1380	0.6482	0.6508	0.8703	1.3611

算法	MAE	F_avg	F_max	AUC	NSS
GS算法	0.2169	0.1815	0.2214	0.8252	2.2154
SAD算法	0.2345	0.1397	0.2662	0.7707	1.1767
SED算法	0.1823	0.1541	0.3420	0.7691	1.3498
SED-GS算法	0.1856	0.1708	0.3634	0.8021	1.5908
SED-SAD算法	0.1833	0.1397	0.2662	0.8108	1.5301
SUDF算法	0.1345	0.4668	0.5654	0.8602	2.1200
U2Net^†算法	0.1065	0.6144	0.6214	0.9281	2.3993
本文算法	0.0868	0.7528	0.7583	0.9793	2.7856

算法	MAE	F_avg	F_max	AUC	NSS
GS算法	0.2169	0.1815	0.2214	0.8252	2.2154
SAD算法	0.2345	0.1397	0.2662	0.7707	1.1767
SED算法	0.1823	0.1541	0.3420	0.7691	1.3498
SED-GS算法	0.1856	0.1708	0.3634	0.8021	1.5908
SED-SAD算法	0.1833	0.1397	0.2662	0.8108	1.5301
SUDF算法	0.1345	0.4668	0.5654	0.8602	2.1200
U2Net^†算法	0.1065	0.6144	0.6214	0.9281	2.3993
本文算法	0.0868	0.7528	0.7583	0.9793	2.7856

算法	MAE	F_avg	F_max	AUC	NSS
BL算法	0.2170	0.6871	0.6914	0.8846	1.3545
MS算法	0.1927	0.6722	0.6941	0.9036	1.4209
LPS算法	0.1858	0.6932	0.7055	0.8806	1.2993
GMR算法	0.1889	0.6953	0.7101	0.8891	1.3939
RBD算法	0.1732	0.6795	0.6904	0.8939	1.3781
MBD算法	0.1718	0.6578	0.6693	0.9164	1.4387
MST算法	0.1579	0.6978	0.7111	0.8872	1.3678
LFCS算法	0.1477	0.7017	0.7231	0.9148	1.4260
本文算法	0.1046	0.8166	0.8214	0.9485	1.6645