SAR ATR with full‐angle data augmentation and feature polymerisation

Zhai, Yongjie; Ma, Hui; Liu, Jian; Deng, Wenbo; Shang, Lijuan; Sun, Bing; Jiang, Ziyi; Guan, Huixin; Zhi, Yihang; Wu, Xi; Zhou, Jihua

doi:10.1049/joe.2019.0219

Cited by 3 publications

(10 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Except for works [41] and [37], we did not find reproducible data augmentation strategies that explicitly synthesize samples at new poses. As pointed out earlier, our approach can be used in conjunction with most of the other strategies outlined in section I-A.…”

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 82%

“…Adding Our poses and sub-pixel shifts (B+S+P) Full-data Features (F) Method Error(%) using 100% data Error(%) using ≤ 20% data SVM (2016) [32] 13.27 47.75 (at 20%) SRC (2016) [32] 10.24 36.35 (at 20%) A-ConvNet (2016) [27] 0.87 35.90 (at 20%) Ensemble DCHUN (2017) [28] 0.91 25.94 (at 20%) CNN-TL-bypass (2017) [26] 0.91 2.85 (at 18%) ResNet (2018) [29] 0.33 5.70 (at 20%) DFFN (2019) [34] 0.17 Baseline data with proposed sub-pixel augmentations as B+S, Baseline data with proposed sub-pixel and pose augmentations as B+S+P, baseline data with simple rotations added (from [41]) as B+R and baseline data with linearly interpolated poses (from [37]) as B+L.…”

Section: Sub-samplingmentioning

confidence: 99%

“…We add real-time pixel level translations for all experiments. For the sake of completion, the simple rotations are produced in [41] using the rotation matrix followed by appropriate cropping and the We observe that even the rotation is small the ground-truth image has a different scattering behavior, which does not get captured by rotation. The comparison is tabulated in table II and can be seen in Fig.…”

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 99%

“…Cha et al [40] use images from a SAR data simulator and refine them using a learned function from real images. Simple rotations of radar images were considered as a data-augmentation method [41]. In [15], Zhong et al suggested key ideas to incorporate prior knowledge in training the model.…”

Section: A Related Workmentioning

confidence: 99%

“…As pointed out earlier, our approach can be used in conjunction with most of the other strategies outlined in section I-A. So, we do a detailed comparison of our pose synthesis approach and sub-pixel level translations with pose synthesis methods in [41] and [37]. We add real-time pixel level translations for all experiments.…”

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 99%

See 4 more Smart Citations

Sparse Signal Models for Data Augmentation in Deep Learning ATR

Agarwal¹,

Sugavanam²,

Ertin³

2020

Preprint

View full text Add to dashboard Cite

Automatic Target Recognition (ATR) algorithms classify a given Synthetic Aperture Radar (SAR) image into one of the known target classes using a set of training images available for each class. Recently, learning methods have shown to achieve state-of-the-art classification accuracy if abundant training data is available, sampled uniformly over the classes, and their poses. In this paper, we consider the task of ATR with a limited set of training images. We propose a data augmentation approach to incorporate domain knowledge and improve the generalization power of a data-intensive learning algorithm, such as a Convolutional neural network (CNN). The proposed data augmentation method employs a limited persistence sparse modeling approach, capitalizing on commonly observed characteristics of wide-angle synthetic aperture radar (SAR) imagery. Specifically, we exploit the sparsity of the scattering centers in the spatial domain and the smoothly-varying structure of the scattering coefficients in the azimuthal domain to solve the ill-posed problem of overparametrized model fitting. Using this estimated model, we synthesize new images at poses and sub-pixel translations not available in the given data to augment CNN's training data. The experimental results show that for the training data starved region, the proposed method provides a significant gain in the resulting ATR algorithm's generalization performance.

show abstract

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 82%

Section: Sub-samplingmentioning

confidence: 99%

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 99%

Section: A Related Workmentioning

confidence: 99%

Section: B Comparison With Existing Sar-atr Modelsmentioning

confidence: 99%

See 3 more Smart Citations

Sparse Signal Models for Data Augmentation in Deep Learning ATR

Agarwal¹,

Sugavanam²,

Ertin³

2020

Preprint

View full text Add to dashboard Cite

show abstract

CNN Performance Analysis for SAR Object Classification

Wangiyana

2022

2022 23rd International Radar Symposium (IRS)

View full text Add to dashboard Cite

Sparse Signal Models for Data Augmentation in Deep Learning ATR

Agarwal,

Sugavanam,

Ertin

2023

Remote Sensing

View full text Add to dashboard Cite

Automatic target recognition (ATR) algorithms are used to classify a given synthetic aperture radar (SAR) image into one of the known target classes by using the information gleaned from a set of training images that are available for each class. Recently, deep learning methods have been shown to achieve state-of-the-art classification accuracy if abundant training data are available, especially if they are sampled uniformly over the classes and in their poses. In this paper, we consider the ATR problem when a limited set of training images are available. We propose a data-augmentation approach to incorporate SAR domain knowledge and improve the generalization power of a data-intensive learning algorithm, such as a convolutional neural network (CNN). The proposed data-augmentation method employs a physics-inspired limited-persistence sparse modeling approach, which capitalizes on the commonly observed characteristics of wide-angle synthetic aperture radar (SAR) imagery. Specifically, we fit over-parametrized models of scattering to limited training data, and use the estimated models to synthesize new images at poses and sub-pixel translations that are not available in the given data in order to augment the limited training data. We exploit the sparsity of the scattering centers in the spatial domain and the smoothly varying structure of the scattering coefficients in the azimuthal domain to solve the ill-posed problem of the over-parametrized model fitting. The experimental results show that, for the training on the data-starved regions, the proposed method provides significant gains in the resulting ATR algorithm’s generalization performance.

show abstract

SAR ATR with full‐angle data augmentation and feature polymerisation

Cited by 3 publications

References 10 publications

Sparse Signal Models for Data Augmentation in Deep Learning ATR

Sparse Signal Models for Data Augmentation in Deep Learning ATR

CNN Performance Analysis for SAR Object Classification

Sparse Signal Models for Data Augmentation in Deep Learning ATR

Contact Info

Product

Resources

About