SAR Targets Classification Based on Deep Memory Convolution Neural Networks and Transfer Parameters

Shang, Ronghua; Wang, Jiaming; Jiao, Licheng; Stolkin, Rustam; Hou, Biao; Li, Yangyang

doi:10.1109/jstars.2018.2836909

Cited by 101 publications

(48 citation statements)

References 31 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…AIS is a tracking system for monitoring movement of ships that can provide labeling information. Shang et al [10] amended a CNN with an information recorder. The recorder was used to store spatial features of labeled samples, and the recorded features were used to predict the labels of unlabeled data points based on spatial similarity to increase the number of labeled samples.…”

Section: Related Workmentioning

confidence: 99%

“…As a result of the above challenges, generating labeled detests for the SAR domain data is in general difficult. In particular, given the size of most existing SAR datasets, training a CNN leads to overfit models, as the number of data points is considerably less than the required sample complexity of training a deep network [9,10]. When the model is overfit, naturally, it will not generalize well on test sets.…”

mentioning

confidence: 99%

“…The general idea that we focus on is to transfer knowledge from a secondary domain to reduce the amount labeled data that are necessary to train a model. Building upon prior works in the area of transfer learning, several recent works have used the idea of knowledge transfer to address the challenges of SAR domains [6,8,10,[12][13][14]. The common idea in these works is to transfer knowledge from a secondary related problem, where labeled data are easy and inexpensive to obtain.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Deep Transfer Learning for Few-Shot SAR Image Classification

et al. 2019

View full text Add to dashboard Cite

The reemergence of Deep Neural Networks (DNNs) has lead to high-performance supervised learning algorithms for the Electro-Optical (EO) domain classification and detection problems. This success is because generating huge labeled datasets has become possible using modern crowdsourcing labeling platforms such as Amazon’s Mechanical Turk that recruit ordinary people to label data. Unlike the EO domain, labeling the Synthetic Aperture Radar (SAR) domain data can be much more challenging, and for various reasons, using crowdsourcing platforms is not feasible for labeling the SAR domain data. As a result, training deep networks using supervised learning is more challenging in the SAR domain. In the paper, we present a new framework to train a deep neural network for classifying Synthetic Aperture Radar (SAR) images by eliminating the need for a huge labeled dataset. Our idea is based on transferring knowledge from a related EO domain problem, where labeled data are easy to obtain. We transfer knowledge from the EO domain through learning a shared invariant cross-domain embedding space that is also discriminative for classification. To this end, we train two deep encoders that are coupled through their last year to map data points from the EO and the SAR domains to the shared embedding space such that the distance between the distributions of the two domains is minimized in the latent embedding space. We use the Sliced Wasserstein Distance (SWD) to measure and minimize the distance between these two distributions and use a limited number of SAR label data points to match the distributions class-conditionally. As a result of this training procedure, a classifier trained from the embedding space to the label space using mostly the EO data would generalize well on the SAR domain. We provide a theoretical analysis to demonstrate why our approach is effective and validate our algorithm on the problem of ship classification in the SAR domain by comparing against several other competing learning approaches.

show abstract

Section: Related Workmentioning

confidence: 99%

mentioning

confidence: 99%

mentioning

confidence: 99%

See 1 more Smart Citation

Deep Transfer Learning for Few-Shot SAR Image Classification

et al. 2019

View full text Add to dashboard Cite

show abstract

“…On the one hand, this size is big enough to contain the spatial feature that is needed for classification. On the other hand, with the smaller input size, the computational efficiency is increased and the risk of over-fitting is prevented [23].…”

Section: Cv-caementioning

confidence: 99%

Complex-Valued Convolutional Autoencoder and Spatial Pixel-Squares Refinement for Polarimetric SAR Image Classification

et al. 2019

Self Cite

View full text Add to dashboard Cite

Recently, deep learning models, such as autoencoder, deep belief network and convolutional autoencoder (CAE), have been widely applied on polarimetric synthetic aperture radar (PolSAR) image classification task. These algorithms, however, only consider the amplitude information of the pixels in PolSAR images failing to obtain adequate discriminative features. In this work, a complex-valued convolutional autoencoder network (CV-CAE) is proposed. CV-CAE extends the encoding and decoding of CAE to complex domain so that the phase information can be adopted. Benefiting from the advantages of the CAE, CV-CAE extract features from a tiny number of training datasets. To further boost the performance, we propose a novel post processing method called spatial pixel-squares refinement (SPF) for preliminary classification map. Specifically, the majority voting and difference-value methods are utilized to determine whether the pixel-squares (PixS) needs to be refined or not. Based on the blocky structure of land cover of PolSAR images, SPF refines the PixS simultaneously. Therefore, it is more productive than current methods worked on pixel level. The proposed algorithm is measured on three typical PolSAR datasets, and better or comparable accuracy is obtained compared with other state-of-the-art methods.

show abstract

“…Multiview SAR data was generated and a multi-input CNN architecture was proposed in [20] to get the features of targets from different views. In order to reduce the demand of extensive labeled samples, high-order features extracted by CNN were made into feature dictionary, and the end-to-end training was carried out through feature metric and two-stage optimization [21]. Different from the task of target recognition, the purpose of terrain classification is to predict each pixel of a PolSAR image.…”

mentioning

confidence: 99%

Automatic Design of CNNs via Differentiable Neural Architecture Search for PolSAR Image Classification

Dong

Zou

Zhang

et al. 2020

IEEE Trans. Geosci. Remote Sensing

View full text Add to dashboard Cite

Convolutional neural networks (CNNs) have shown good performance in polarimetric synthetic aperture radar (PolSAR) image classification due to the automation of feature engineering. Excellent hand-crafted architectures of CNNs incorporated the wisdom of human experts, which is an important reason for CNN's success. However, the design of the architectures is a difficult problem, which needs a lot of professional knowledge as well as computational resources. Moreover, the architecture designed by hand might be suboptimal, because it is only one of thousands of unobserved but objective existed paths. Considering that the success of deep learning is largely due to its automation of the feature engineering process, how to design automatic architecture searching methods to replace the handcrafted ones is an interesting topic. In this paper, we explore the application of neural architecture search (NAS) in PolSAR area for the first time. Different from the utilization of existing NAS methods, we propose a differentiable architecture search (DAS) method which is customized for PolSAR classification. The proposed DAS is equipped with a PolSAR tailored search space and an improved one-shot search strategy. By DAS, the weights parameters and architecture parameters (corresponds to the hyperparameters but not the topologies) can be optimized by stochastic gradient descent method during the training. The optimized architecture parameters should be transformed into corresponding CNN architecture and re-train to achieve highprecision PolSAR classification. In addition, complex-valued DAS is developed to take into account the characteristics of PolSAR images so as to further improve the performance. Experiments on three PolSAR benchmark datasets show that the CNNs obtained by searching have better classification performance than the hand-crafted ones. He is currently a Professor and Vice Head with the Department of Information Engineering, Harbin Institute of Technology. His research interests include SAR image processing, polarimetric SAR, and polarimetric SAR interferometry.

show abstract

SAR Targets Classification Based on Deep Memory Convolution Neural Networks and Transfer Parameters

Cited by 101 publications

References 31 publications

Deep Transfer Learning for Few-Shot SAR Image Classification

Deep Transfer Learning for Few-Shot SAR Image Classification

Complex-Valued Convolutional Autoencoder and Spatial Pixel-Squares Refinement for Polarimetric SAR Image Classification

Automatic Design of CNNs via Differentiable Neural Architecture Search for PolSAR Image Classification

Contact Info

Product

Resources

About