基于非监督对比学习的火星地形特征提取方法

杨博; 魏翔; 于贺; 刘超凡

doi:10.13700/j.bh.1001-5965.2022.0525

基于非监督对比学习的火星地形特征提取方法

doi: 10.13700/j.bh.1001-5965.2022.0525

北京航空航天大学宇航学院，北京 100191

详细信息

通讯作者:
E-mail：yangbo@buaa.edu.cn

中图分类号: V448
计量
- 文章访问数: 236
- HTML全文浏览量: 90
- PDF下载量: 29
- 被引次数: 0
出版历程
- 收稿日期: 2022-06-22
- 录用日期: 2022-09-02
- 网络出版日期: 2022-09-22
- 整期出版日期: 2024-06-27

Martian terrain feature extraction method based on unsupervised contrastive learning

School of Astronautics，Beihang University，Beijing 100191，China

More Information

Corresponding author: E-mail：yangbo@buaa.edu.cn

摘要

摘要:
火星表面地形智能识别对火星车自主探测具有重要意义，火星地形图像的特征提取方法目前主要分为传统的浅层视觉特征提取和基于监督学习的深层特征提取2类。找回丢失图像信息、获取大量带标签数据是要解决的关键问题。为此，提出一种基于非监督对比学习的火星地形特征识别方法，通过建立图像字典数据集，用“问询”和“编码”2组神经网络分别将单个图像与“字典”数据集中其他图像进行对比，用相似度泛函作为损失函数对网络进行训练，从而实现对火星地形图像的特征识别。所提方法还具有对训练数据集之外的新类型地形图像识别能力，后续识别分类优越性突出。仿真结果表明：所提方法识别准确率为85.4%，对新类型地形图像的识别准确率为84.5%。
- 对比学习 /
- 非监督 /
- 深度学习 /
- 火星地形 /
- 特征提取
Abstract:
Intelligent surface terrain recognition of Martian is significant for the autonomous exploration of Mars rovers. At present, the methods used for feature extraction of Martian terrain images are mainly divided into two categories: traditional shallow visual feature extraction and deep feature extraction based on supervised learning. However, these methods tend to lose image information and require a large amount of labeled data, which are key problems to be solved. A Martian terrain feature recognition method based on unsupervised contrastive learning was proposed. By establishing the image dictionary dataset, a single image was compared with other images in the dictionary dataset by using two groups of neural networks, namely “query” and “encode”. Then, the similarity function was used as the loss function to train the network, so as to realize the feature recognition of Martian terrain images. The proposed method could also recognize new types of terrain images outside the training dataset and showed superior performance in subsequent recognition and classification tasks. Simulation results show that the recognition accuracy of the proposed method is 85.4%, and the recognition accuracy of new terrain images is 84.5%.
- contrastive learning /
- unsupervised /
- deep learning /
- Martian terrain /
- feature extraction

HTML全文

图 1 火星典型地形图像熵分布

Figure 1. Entropy of typical Martian terrain images distribution

下载: 全尺寸图片幻灯片

图 2 地形图像分割

Figure 2. Terrain image segmentation

下载: 全尺寸图片幻灯片

图 3 增强处理示意图

Figure 3. Schematic diagram of enhancement processing

下载: 全尺寸图片幻灯片

图 4 对比学习算法实施思路

Figure 4. Implementation ideas of contrastive learning algorithm

下载: 全尺寸图片幻灯片

图 5 仿真验证设计

Figure 5. Design of simulation verification

下载: 全尺寸图片幻灯片

图 6 神经网络卷积层特征输出图

Figure 6. Convolution layers characteristic output images of neural network

下载: 全尺寸图片幻灯片

图 7 分类识别混淆矩阵

Figure 7. Confusion matrix of classification and recognition

下载: 全尺寸图片幻灯片

图 8 不同参数下分类识别准确率

Figure 8. Classification and recognition accuracy under different parameters

下载: 全尺寸图片幻灯片

图 9 不同方法下地形图像识别准确率

Figure 9. Recognition accuracy of terrain image under different methods

下载: 全尺寸图片幻灯片

图 10 火星远景及多种类型地形图像

Figure 10. Martian vista and various terrain images

下载: 全尺寸图片幻灯片

图 11 添加新一类地形图像分类混淆矩阵

Figure 11. Classification confusion matrix with a new class of terrain image

下载: 全尺寸图片幻灯片

图 12 输出特征维度对识别准确率的影响

Figure 12. Influence of output feature dimension on recognition accuracy

下载: 全尺寸图片幻灯片

表 1 5类火星地形示例

Table 1. Examples of five types of Martian terrain

类别	典型图	影响
砂涟漪		导致车轮陷入沙坑无法运转
尖锐岩石		导致车轮破损
细砂		粒度小，车轮滑转率小
粗砂		粒度大，车轮滑转率大
基岩		硬度大，可通行，易损坏车轮

下载: 导出CSV

表 2 火星典型地形图像熵

Table 2. Entropy of typical Martian terrain images

类别	均值	方差
砂涟漪	6.3211	0.4777
尖锐岩石	6.6011	0.2715
细砂	7.1354	0.0088
粗砂	7.1962	0.0661
基岩	6.4962	0.1772

下载: 导出CSV

参考文献(22)

[1]	OJHA L, WILHELM M, MURCHIE S, et al. Spectral evidence for hydrated salts in recurring slope lineae on Mars[J]. Nature Geoscience, 2015, 8(11): 829-832.
[2]	张洪华, 梁俊, 黄翔宇, 等. 嫦娥三号自主避障软着陆控制技术[J]. 中国科学:技术科学, 2014, 44(6): 559-568. doi: 10.1360/092014-51 ZHANG H H, LIANG J, HUANG X Y, et al. Autonomous hazard avoidance control for Chang’E-3 soft landing[J]. Scientia Sinica (Technologica), 2014, 44(6): 559-568 (in Chinese). doi: 10.1360/092014-51
[3]	LEE S J, CHEN T L, YU L, et al. Image classification based on the boost convolutional neural network[J]. IEEE Access, 2018, 6: 12755-12768. doi: 10.1109/ACCESS.2018.2796722
[4]	JIAO L C, ZHANG F, LIU F, et al. A survey of deep learning-based object detection[J]. IEEE Access, 2019, 7: 128837-128868. doi: 10.1109/ACCESS.2019.2939201
[5]	JU J, JUNG H, OH Y, et al. Extending contrastive learning to unsupervised coreset selection[J]. IEEE Access, 2022, 10: 7704-7715. doi: 10.1109/ACCESS.2022.3142758
[6]	SUN Q R, LIU Y Y, CHUA T S, et al. Meta-transfer learning for few-shot learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2019: 403-412.
[7]	SHORTEN C, KHOSHGOFTAAR T M. A survey on image data augmentation for deep learning[EB/OL]. (2022-04-19)[2022-04-23].
[8]	BOWLES C, CHEN L, GUERRERO R, et al. GAN Augmentation: Augmenting training data using generative adversarial networks[EB/OL]. (2018-10-25)[2022-04-23].
[9]	FRID-ADAR M, BEN-COHEN A, AMER R, et al. Improving the segmentation of anatomical structures in chest radiographs using U-net with an ImageNet pre-trained encoder[C]//Proceedings of the International Workshop on Reconstruction and Analysis of Moving Body Organs, International Workshop on Breast Image Analysis, International Workshop on Thoracic Image Analysis. Berlin: Springer, 2018: 159-168.
[10]	CUI B G, CHEN X, LU Y. Semantic segmentation of remote sensing images using transfer learning and deep convolutional neural network with dense connection[J]. IEEE Access, 2020, 8: 116744-116755. doi: 10.1109/ACCESS.2020.3003914
[11]	GENG J, DENG X Y, MA X R, et al. Transfer learning for SAR image classification via deep joint distribution adaptation networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(8): 5377-5392. doi: 10.1109/TGRS.2020.2964679
[12]	LI H F, CUI Z Q, ZHU Z Q, et al. RS-MetaNet: Deep metametric learning for few-shot remote sensing scene classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(8): 6983-6994. doi: 10.1109/TGRS.2020.3027387
[13]	VU T H, JAIN H, BUCHER M, et al. DADA: Depth-aware domain adaptation in semantic segmentation[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway: IEEE Press, 2019: 7363-7372.
[14]	JING L L, TIAN Y L. Self-supervised visual feature learning with deep neural networks: A survey[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(11): 4037-4058. doi: 10.1109/TPAMI.2020.2992393
[15]	MISRA I, VAN DER MAATEN L. Self-supervised learning of pretext-invariant representations[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 6706-6716.
[16]	JAISWAL A, BABU A R, ZADEH M Z, et al. A survey on contrastive self-supervised learning[J]. Technologies, 2020, 9(1): 2. doi: 10.3390/technologies9010002
[17]	HJELM R D, FEDOROV A, LAVOIE-MARCHILDON S, et al. Learning deep representations by mutual information estimation and maximization[EB/OL]. (2018-08-20)[2022-04-23].
[18]	HÉNAFF O J, SRINIVAS A, DE FAUW J, et al. Data-efficient image recognition with contrastive predictive coding[EB/OL]. (2019-03-22)[2022-04-24].
[19]	WU Z R, XIONG Y J, YU S X, et al. Unsupervised feature learning via non-parametric instance discrimination[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2018: 3733-3742.
[20]	CHEN T, KORNBLITH S, NOROUZI M, et al. A simple framework for contrastive learning of visual representations[C]//Proceedings of the 37th International Conference on Machine Learning. New York: ACM, 2020: 1597–1607.
[21]	HE K M, FAN H Q, WU Y X, et al. Momentum contrast for unsupervised visual representation learning[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2020: 9726-9735.
[22]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE Press, 2016: 770-778.