FTAN: Feature Transform and Alignment Network for cross-domain specific emitter identification

Deep learning models have obtained state-of-the-art results for medical image analysis. However, CNNs require a massive amount of labelled data to achieve a high performance. Moreover, many supervised learning approaches assume that the training/source dataset and test/target dataset follow the same probability distribution. Nevertheless, this assumption is hardly satisfied in real-world data and when the models are tested on an unseen domain there is a significant performance degradation. In this work, we present an unsupervised Cross-Modality Adversarial Domain Adaptation (C-MADA) framework for medical image segmentation. C-MADA implements an image-level and feature-level adaptation method in a two-step sequential manner. First, images from the source domain are translated to the target domain through an unpaired image-to-image adversarial translation with cycle-consistency loss. Then, a U-Net network is trained with the mapped source domain images and target domain images in an adversarial manner to learn domain-invariant feature representations and produce segmentations for the target domain. Furthermore, to improve the network’s segmentation performance, information about the shape, texture, and contour of the predicted segmentation is included during the adversarial training. C-MADA is tested on the task of brain MRI segmentation from the crossMoDa Grand Challenge and is ranked within the top 15 submissions of the challenge.

Specific emitter identification technology aims to identify the signal emitted by a specific emitter from multiple signals. Existing techniques for specific emitter identification are faced with the problem that the identification performance decreases with the passage of time. In this paper, we propose a method for specific emitter identification at different time. The method treats the signals of radiation sources at different time as signals of separate domains, and resolves the influence of time on specific emitter identification by eliminating the factor of different domains. The experimental results show that the method has better performance on both data corresponding to the time involved in training and data corresponding to the time not involved in training compared with both domain-free migration and ordinary domain migration methods. At the same time, the proposed tagging flexible loss function is more efficient in the face of new time tag free data.

Specific emitter identification (SEI), as an important problem in situational awareness, identifies emitters via unique characteristics. However, current SEI methods mostly suffer from appropriately setting the trade-off between comprehensiveness and efficiency when extracting fingerprint features. To address the issue, this paper provides a novel SEI framework with a separate representation module. Within the novel framework, manifolds are proposed to be signal representations and multi-level manifold features are extracted as fingerprint features. We first build the SEI model from the nonlinear dynamic perspective, where the SEI process identifies the nonlinear systems via a measurement sequence. Then, we demonstrate that manifolds can represent emitters equivalently and prove the one-to-one correspondence between manifolds and emitter individuals. Hence, manifolds can highlight unique nonlinear dynamic characteristics and simultaneously describe comprehensive system working processes. The coordinate delayed technique and manifold learning methods are employed to reconstruct the phase space and manifold, respectively. For accomplishing the identification task, multi-level manifold features, comprising intrinsic dimension, topological features, conformal features, and Riemannian metric features, are extracted from the reconstructed manifolds and input to an ensemble learning scheme, named Adaboost. Extensive simulation and real-world experiments agree with our analytical conclusions and confirm the proposed method’s efficiency. The results also demonstrate that the proposed method achieves a high recognition accuracy, outstanding adaptability, and strong robustness.

Internet of Things (IoT) deployments continue at an accelerated rate and are projected to reach 75 billion by 2025. Most IoT devices employ weak or no encryption at all due to constrained onboard resources, prohibitive manufacturing costs, and challenges due to implementing and managing encryption at scale. Specific Emitter Identification (SEI) is an approach intended to address the security risks associated with unencrypted or weakly encrypted IoT deployments. SEI passively exploits inherent, unintentional waveform features that are imparted during normal device operations, thus it does not require modification of the device being identified. Recently, Deep Learning (DL)-based SEI has garnered a lot of interest due to its ability to learn discriminatory features directly from an emitter's digitally sampled waveforms, thus eliminating feature engineering commonly associated with SEI. DL-based SEI uses the entirety of the digitally sampled waveform, which includes the unintentional as well as intentional waveform structure. The intentional waveform structure does not convey emitter specific features, thus it is information not used by the DL-based SEI process. In fact, this work shows that the intentional waveform structure acts as a confuser that inhibits the DL network's ability to learn the features needed to discern one emitter from another and its negative impacts worsen as the number of emitters increases. Our work shows that frequency-domain removal of the intentional waveform structure coupled with a Long Short-Term Memory (LSTM) results in superior SEI performance.

Traditional specific emitter identification (SEI) methods need to artificially design extraction methods of the signal fingerprint features, which require a large amount of calculation and low classification accuracy. A supervised SEI framework based on deep residual shrinkage Network (DRSN) is proposed in this paper, which is used to adaptively learn the deep features of transmitted signals and classify them. In our framework, soft thresholding is introduced to reduce the noise of the original I/Q signal which contains a lot of noise, so as to reduce the influence of noise on the inaccuracy of RF fingerprint feature extraction. The I/Q signal of the receiver is sliced and spliced (pre-processing step) before inputting to the DRSN model to reduce the computational complexity. The experimental results show that the proposed scheme can correctly distinguish 16 credible transmitters with 98% accuracy respectively. Compared with the existing SEI methods, our scheme has the advantages of excellent identification accuracy and strong anti-interference ability.

Unsupervised domain adaptation (UDA) has emerged as an effective paradigm for reducing the huge manual annotation cost for Person Re-Identification (Re-ID). Many of the recent UDA methods for Re-ID are clustering-based and select all the pseudo-label samples in each iteration for the model training. However, there are many wrong labeled samples that will mislead the model optimization under this circumstance. To solve this problem, we propose a Progressive Unsupervised Domain Adaptation (PUDA) framework for image-based Person Re-ID to reduce the negative effect of wrong pseudo-label samples on the model training process. Specifically, we first pretrain a CNN model on a labeled source dataset, then finetune the model on unlabeled target dataset with the following three steps iteratively: 1) estimating pseudo-labels for all the images in the target dataset with the model trained in the last iteration; 2) extending the training set by adding pseudo-label samples with higher label confidence; 3) updating the CNN model with the expanded training set in a supervised manner. During the iteration process, the number of pseudo-label samples added increased progressively. In particular, a Moderate Initial Selections (MIS) strategy for pseudo-label sampling is also proposed to reduce the negative impacts of random noise features in the early iterations and mislabeled samples in the late iterations on the model. The proposed framework with MIS strategy is validated on the Duke-to-Market, Market-to-Duke unsupervised domain adaptation tasks and achieves improvements of 4.2 points (absolute, i.e., 80.0% vs. 75.8%) and 1.7 points (absolute, i.e., 70.7% vs. 69.0%) in mAP correspondingly.

Optimal transport (OT) can be applied to cross-domain adaptation in speaker verification (SV) by converting speakers' probability distributions from source to target domains. However, in scenarios involving over-massive categories (speakers) or difficult samples in discrimination, OT often has difficulty computing effective transports. To address this challenge, we propose an OT-based unsupervised domain adaptation (UDA) framework for SV, OT with a diversified memory bank, called DMB-OT, which ensures the accuracy of transfers by two strategies: (1) It regularizes the solution space of OT, which attempts to plan transformations between audio samples from the same speaker with high confidence; (2) it integrates a dynamic curriculum learning algorithm, preventing OT from calculating transport couplings based on hard-discriminative samples in the early stage of UDA. Experiments under different target domains showed that our unsupervised DMB-OT could significantly improve the performance of OT-based UDA and could even match the performance of the supervised PLDA-based adaptation.

To solve the problem that the neural network-based specific emitter identification (SEI) method is limited by the training scene and has poor generalization ability in the complex scene, an SEI method based on deep adversarial domain adaptation (DADA) is proposed. DADA integrates deep neural networks and adversarial methods into the domain adaptation problem of transfer learning. It trains the feature extractor by maximizing the domain discrimination loss and transfers the high-quality deep radio frequency fingerprint (RFF) features extracted under weak interference conditions to the unlabeled emitter of the same class affected by the channel and noise. At the same time, the loss of the sample classifier is minimized, and the deep RFF features of unlabeled emitters can be accurately classified and identified. By repeating the above-mentioned adversarial process, the network realizes unsupervised transfer learning based on deep RFF features. The analysis of the identification results of 20 CC2530 devices under the Rayleigh channel and different signal-to-noise ratio (SNR) conditions proves that DADA can effectively improve the network's identification performance for unlabeled emitters under different conditions and has good performance of adaptability and generalization.

The specific emitter identification is widely used in electronic countermeasures, spectrum control, wireless network security and other civil and military fields. In response to the problems that the traditional specific emitter identification algorithm relies on a priori knowledge and has poor generalizability, and the existing specific emitter identification algorithm based on deep learning has poor feature selection and the adopted feature extraction network is not targeted, etc., the specific emitter identification algorithm based on multi-sequence feature learning is proposed. Firstly, multiple sequence features of the emitted signal of the communication radiation source are extracted, and these features are combined into multiple sequence features. Secondly, the multiple sequence fusion convolutional network is constructed to fuse and deeply extract the multiple sequence features and complete the classification of individual communication radiation sources through the classifier of neural network. The selected sequence features of this algorithm contain more and more essential RFF information, while the targeted design of the multi-sequence feature fusion learning network can effectively extract the essential RFF information. The results show that the algorithm can significantly improve the performance of SEI compared with the benchmark algorithm, with a recognition rate gain of about 17%.

In recent years, the growing complexity of the electromagnetic environment has rendered traditional methods of specific emitter identification (SEI) inadequate in meeting the increasing demands for both high accuracy and generalization capability. Moreover, the high cost and confidentiality associated with emitter data have made the acquisition of large-scale labeled datasets exceedingly difficult, further intensifying the challenges in SEI tasks. To address these issues, this paper proposes a deep metric learning-based method for SEI. The approach first leverages a large-scale emitter signal dataset to train a high-accuracy, generalizable pre-trained model. Subsequently, the pre-trained model is fine-tuned in few-shot scenarios to achieve accurate emitter identification under data-scarce conditions. Experimental results indicate that the fine-tuning method, built upon the pre-trained model, demonstrates excellent SEI performance even with limited training data. Specifically, when only 10% of the training data is used, the model achieves an impressive accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0 \%}$</tex>.

In the specific emitter identification, the current traditional classification and identification methods have difficulty in extracting fingerprint features, and the accuracy of fingerprint identification is low. In order to better extract the fingerprint features, a specific emitter identification algorithm is proposed based on the Convolutional Neural Network Dense Block module, called the Deep Dense Feature Network (DDFN). The network can effectively realize automatic extraction and multiplexing of fingerprint features, strengthen the transfer of features, and reduce the problem of vanishing gradients. The experimental results show that DDFN can achieve a classification accuracy of 93.17%, and can better realize the identification of specific radiation sources in a complex electro-magnetic environment.

In an e-commerce business, the ability to parse postal addresses into sub-component entities (such as building, locality) is essential to take automated actions at scale for successful delivery of shipments. The entities can be leveraged to build applications for logistics related operations, e.g. geocoding, assessing address completeness. Training an accurate address parser requires a significant number of manually labeled examples which is very expensive to create, especially when trying to build model(s) for multiple countries with unique address structure. To tackle this problem, in this paper, we present a novel Unsupervised Domain Adaptation (UDA) framework to transfer knowledge acquired by training a parser on labeled data from one country (source domain) to another (target domain) with unlabeled data. We specifically propose a multi-task student-teacher model comprising of three components: 1) specialized teachers trained on source data to create a pseudo labeled dataset, 2) consistency regularization, that uses a new data augmentation technique for sequence tagging data, and 3) boundary detection, leveraging signals in addresses like commas and text box boundaries. Multiple experiments on diverse address datasets (In this paper, we do not reveal the name of the e-commerce countries on which we evaluate our models due to business confidentiality. We also mask finer address details with (XX) to preserve customer’s privacy.) demonstrate that our approach outperforms state-of-the-art UDA baselines for Named Entity Recognition (NER) task in terms of F1-score by 2–9%.

Electroencephalogram (EEG) signals convey abundant physiological and psychological information. Decoding EEG is fundamental for brain-computer interaction and medical rehabilitation. Nevertheless, the nonstationarity and inter-individual variability of EEG signals impedes current models from achieving robust crosssubject generalization without expensive subject-specific recalibration, thereby restricting practical deployment. Unsupervised domain adaptation (UDA) aims to improve generalization by minimizing distribution discrepancies between source and target domains. Recent studies regard different subjects as distinct domains and leverage UDA to facilitate cross-subject EEG classification by learning domain-invariant features through discrepancy minimization or adversarial training. However, these conventional methods are all developed in Euclidean space, which are insufficient to capture the non-linear structures inherent in EEG data. To address this limitation, we propose a Deep Adversarial Neural Network on the Symmetric Positive Definite (SPD) matrix manifold, referred to as SPD-DANN, which facilitates the extraction of subject-invariant features through adversarial learning. Additionally, we design an SPD domain feature alignment loss and an SPD class prototype pair loss to simultaneously promote feature alignment across subjects and enhance feature space discriminability. Extensive experiments on four brain-computer interface (BCI) datasets demonstrate that our method surpasses several state-of-the-art UDA techniques. Furthermore, the proposed loss functions are readily adaptable to broader unsupervised or semi-supervised domain adaptation frameworks.

Specific emitter identification (SEI) and automatic modulation classification (AMC) aregenerally two separate tasks in the field of radio monitoring. Both tasks have similarities in terms oftheir application scenarios, signal modeling, feature engineering, and classifier design. It is feasibleand promising to integrate these two tasks, with the benefit of reducing the overall computationalcomplexity and improving the classification accuracy of each task. In this paper, we propose adual-task neural network named AMSCN that simultaneously classifies the modulation and thetransmitter of the received signal. In the AMSCN, we first use a combination of DenseNet andTransformer as the backbone network to extract the distinguishable features; then, we design amask-based dual-head classifier (MDHC) to reinforce the joint learning of the two tasks. To trainthe AMSCN, a multitask cross-entropy loss is proposed, which is the sum of the cross-entropyloss of the AMC and the cross-entropy loss of the SEI. Experimental results show that our methodachieves performance gains for the SEI task with the aid of additional information from the AMCtask. Compared with the traditional single-task model, our classification accuracy of the AMC isgenerally consistent with the state-of-the-art performance, while the classification accuracy of the SEIis improved from 52.2% to 54.7%, which demonstrates the effectiveness of the AMSCN.

Unsupervised Domain adaptation (UDA) attempts to recognize the unlabeled target samples by building a learning model from a differently-distributed labeled source domain. Conventional UDA concentrates on extracting domain-invariant features through deep adversarial networks. However, most of them seek to match the different domain feature distributions, without considering the task-specific decision boundaries across various classes. In this paper, we propose a novel Adversarial Dual Distinct Classifiers Network (AD <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> CN) to align the source and target domain data distribution simultaneously with matching task-specific category boundaries. To be specific, a domain-invariant feature generator is exploited to embed the source and target data into a latent common space with the guidance of discriminative cross-domain alignment. Moreover, we naturally design two different structure classifiers to identify the unlabeled target samples over the supervision of the labeled source domain data. Such dual distinct classifiers with various architectures can capture diverse knowledge of the target data structure from different perspectives. Extensive experimental results on several cross-domain visual benchmarks prove the model's effectiveness by comparing it with other state-of-the-art UDA.