In recent years, synthetic aperture radar (SAR) ship detection methods based on convolutional neural networks (CNNs) have attracted wide attention in remote sensing fields. However, these methods require a large number of labeled SAR images to train the network, where labeling for SAR images is more expensive and time-consuming than that for optical images. To address the problem of lacking labeled SAR images, in this article, we proposed an unsupervised domain adaptation (UDA) framework based on progressive transfer for SAR ship detection by transferring knowledge from the optical domain to the SAR domain. Due to the prominent difference between the optical and SAR images, our approach progressively transfers knowledge from three levels: pixel level, feature level, and prediction level. At the pixel level, considering the difference in imaging mechanism, we propose a special data augmentation method for ship targets and build the generator with skip connection based on generative adversarial networks (GANs) to generate transition domain, which can reduce the appearance discrepancy between the optical and SAR images. At the feature level, the detector is trained to learn the domain-invariant features with adversarial alignment. At the prediction level, we further use the predicted pseudo-labels obtained by the feature-aligned detector to learn more discriminative features of the SAR images directly and propose the robust self-training (RST) method to reduce the influence of noisy pseudo-labels on detector training. Specially, RST is formulated as a loss minimization problem for object detection. The experimental results based on the domain adaptation from optical dataset to SAR dataset demonstrate that our approach achieves superior SAR ship detection performance with unlabeled SAR images.
In the remote sensing image processing field, the synthetic aperture radar (SAR) target-detection methods based on convolutional neural networks (CNNs) have gained remarkable performance relying on large-scale labeled data. However, it is hard to obtain many labeled SAR images. Semi-supervised learning is an effective way to address the issue of limited labels on SAR images because it uses unlabeled data. In this paper, we propose an improved faster regions with CNN features (R-CNN) method, with a decoding module and a domain-adaptation module called FDDA, for semi-supervised SAR target detection. In FDDA, the decoding module is adopted to reconstruct all the labeled and unlabeled samples. In this way, a large number of unlabeled SAR images can be utilized to help structure the latent space and learn the representative features of the SAR images, devoting attention to performance promotion. Moreover, the domain-adaptation module is further introduced to utilize the unlabeled SAR images to promote the discriminability of features with the assistance of the abundantly labeled optical remote sensing (ORS) images. Specifically, the transferable features between the ORS images and SAR images are learned to reduce the domain discrepancy via the mean embedding matching, and the knowledge of ORS images is transferred to the SAR images for target detection. Ultimately, the joint optimization of the detection loss, reconstruction, and domain adaptation constraints leads to the promising performance of the FDDA. The experimental results on the measured SAR image datasets and the ORS images dataset indicate that our method achieves superior SAR target detection performance with limited labeled SAR images.
14 June 2021 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Recent years have witnessed great progress in synthetic aperture radar (SAR) target detection methods based on deep learning. However, these methods generally assume the training data and test data obey the same distribution, which does not always hold when the radar parameters, imaging algorithm, viewpoints, scenes, etc., change in practice. When such a distribution mismatch occurs, it will cause a significant performance drop. Domain adaptation methods provide an effective way to address this problem by transferring knowledge from the source domain (training data) to the target domain (test data). In this article, we proposed an unsupervised faster R-CNN SAR target detection framework based on domain adaptation, which can improve SAR target detection performance in the unlabeled target domain by borrowing the knowledge of the labeled source domain. Our approach is composed of the following three stages: pixel-domain adaptation (PDA), multilevel feature domain adaptation (MFDA), and iterative pseudolabeling (IPL). By generating transition domain using generative adversarial networks, the PDA stage can reduce the appearance differences of SAR images. At the MFDA stage, the detector can not only learn the domain-invariant global features and instance-level regional features via multilevel adversarial learning in the common feature space but also reweight the low-level global features according to their relative importance to the target domain. At the IPL stage, we design an iterative pseudo labeling strategy that can select pseudo-labels on instance level and image level to encourage the detector to learn more discriminative features of the target domain directly. We evaluate our method using miniSAR and FARADSAR datasets. The experimental results demonstrate the effectiveness of the proposed unsupervised domain adaptation target detection approach.
10 January 2023 in IEEE Transactions on Geoscience and Remote Sensing
In recent years, target detection methods based on deep learning have achieved extensive development in SAR ship detection. However, training such detectors requires target-level annotations of SAR images that are hard to be obtained in practice. To reduce the dependence of network training on expensive target-level annotations, we propose a novel semi-supervised SAR ship detection network via scene characteristic learning. The proposed network focuses on utilizing the scene-level annotations of SAR images to improve the detection performance in the case of limited target-level annotations. Compared with the traditional fully supervised SAR ship detection network, the proposed network constructs a scene characteristic learning branch parallel with the detection branch. In the scene characteristic learning branch, a scene classification loss and a scene aggregation loss are designed to utilize the scene-level annotations. Under the constraint of these two losses, the feature extraction network can fully learn the scene characteristics of SAR images, thus enhancing its feature representation ability for ship targets and clutter. In addition, we propose a hierarchical test process from scene to target. After recognizing the scene types of input SAR images, we design different detection strategies for SAR images recognized as different scenes. The proposed test process can significantly reduce the inland and inshore false alarms, thus leading to higher detection performance. The experiments based on two measured SAR ship detection datasets demonstrate the effectiveness of the proposed method.
It is expensive and time-consuming to obtain a large number of labeled synthetic aperture radar (SAR) images. In the task of small training data size, the results of target detection on SAR images using deep network approaches are usually not ideal. In this study, considering that optical remote sensing images are much easier to be labeled than SAR images, we assume to have a large number of labeled optical remote sensing images and a small number of labeled SAR images with the similar scenes, propose to transfer knowledge from optical remote sensing images to SAR images, and develop a domain adaptive Faster R-CNN for SAR target detection with small training data size. In the proposed method, in order to make full use of the label information and realize more accurate domain adaptation knowledge transfer, an instance level domain adaptation constraint is used rather than feature level domain adaptation constraint. Specifically, generative adversarial network (GAN) constraint is applied as the domain adaptation constraint in the adaptation module after the proposals of Faster R-CNN to achieve instance level domain adaptation and learn the transferable features. The experimental results on the measured SAR image dataset show that the proposed method has higher detection accuracy in the task of SAR target detection with small training data size than the traditional Faster R-CNN.
Usually radar target recognition methods only use a single type of high-resolution radar signal, e.g., high-resolution range profile (HRRP) or synthetic aperture radar (SAR) images. In fact, in the SAR imaging procedure, we can simultaneously obtain both the HRRP data and the corresponding SAR image, as the information contained within them is not exactly the same. Although the information contained in the HRRP data and the SAR image are not exactly the same, both are important for radar target recognition. Therefore, in this paper, we propose a novel end-to-end two stream fusion network to make full use of the different characteristics obtained from modeling HRRP data and SAR images, respectively, for SAR target recognition. The proposed fusion network contains two separated streams in the feature extraction stage, one of which takes advantage of a variational auto-encoder (VAE) network to acquire the latent probabilistic distribution characteristic from the HRRP data, and the other uses a lightweight convolutional neural network, LightNet, to extract the 2D visual structure characteristics based on SAR images. Following the feature extraction stage, a fusion module is utilized to integrate the latent probabilistic distribution characteristic and the structure characteristic for the reflecting target information more comprehensively and sufficiently. The main contribution of the proposed method consists of two parts: (1) different characteristics from the HRRP data and the SAR image can be used effectively for SAR target recognition, and (2) an attention weight vector is used in the fusion module to adaptively integrate the different characteristics from the two sub-networks. The experimental results of our method on the HRRP data and SAR images of the MSTAR and civilian vehicle datasets obtained improvements of at least 0.96 and 2.16%, respectively, on recognition rates, compared with current SAR target recognition methods.