Search

Data augmentation has been widely used to improve generalization in training deep neural networks. Recent works show that using worst-case transformations or adversarial augmentation strategies can significantly improve accuracy and robustness. However, due to the non-differentiable properties of image transformations, searching algorithms such as reinforcement learning or evolution strategy have to be applied, which are not computationally practical for large-scale problems. In this work, we show that by simply applying consistency training with random data augmentation, state-of-the-art results on domain adaptation (DA) and generalization (DG) can be obtained. To further improve the accuracy and robustness with adversarial examples, we propose a differentiable adversarial data augmentation method based on spatial transformer networks (STNs). The combined adversarial and random-transformation-based method outperforms the state-of-the-art on multiple DA and DG benchmark datasets. Furthermore, the proposed method shows desirable robustness to corruption, which is also validated on commonly used datasets.

With the development of the Industrial Internet of Things (IIoT), industrial wireless sensors need to upload the collected private data to the cloud servers, resulting in a large amount of private data being exposed on the Internet. Private data are vulnerable to hacking. Many complex wireless-sensor-authentication protocols have been proposed. In this paper, we proposed an efficient authentication protocol for IIoT-oriented wireless sensor networks. The protocol introduces the PUF chip, and uses the Bloom filter to save and query the challenge–response pairs generated by the PUF chip. It ensures the security of the physical layer of the device and reduces the computing cost and communication cost of the wireless sensor side. The protocol introduces a pre-authentication mechanism to achieve continuous authentication between the gateway and the cloud server. The overall computational cost of the protocol is reduced. Formal security analysis and informal security analysis proved that our proposed protocol has more security features. We implemented various security primitives using the MIRACL cryptographic library and GMP large number library. Our proposed protocol was compared in-depth with related work. Detailed experiments show that our proposed protocol significantly reduces the computational cost and communication cost on the wireless sensor side and the overall computational cost of the protocol.

The novel sensing technology airborne passive bistatic radar (PBR) has the problem of being affecting by multipath components in the reference signal. Due to the movement of the receiving platform, different multipath components contain different Doppler frequencies. When the contaminated reference signal is used for space–time adaptive processing (STAP), the power spectrum of the spatial–temporal clutter is broadened. This can cause a series of problems, such as affecting the performance of clutter estimation and suppression, increasing the blind area of target detection, and causing the phenomenon of target self-cancellation. To solve this problem, the authors of this paper propose a novel algorithm based on sparse Bayesian learning (SBL) for direct clutter estimation and multipath clutter suppression. The specific process is as follows. Firstly, the space–time clutter is expressed in the form of covariance matrix vectors. Secondly, the multipath cost is decorrelated in the covariance matrix vectors. Thirdly, the modeling error is reduced by alternating iteration, resulting in a space–time clutter covariance matrix without multipath components. Simulation results showed that this method can effectively estimate and suppress clutter when the reference signal is contaminated.

Multi-object tracking is a crucial problem for autonomous vehicle. Most state-of-the-art approaches adopt the tracking-by-detection strategy, which is a two-step procedure consisting of the detection module and the tracking module. In this paper, we improve both steps. We improve the detection module by incorporating the temporal information, which is beneficial for detecting small objects. For the tracking module, we propose a novel compressed deep Convolutional Neural Network (CNN) feature based Correlation Filter tracker. By carefully integrating these two modules, the proposed multi-object tracking approach has the ability of re-identification (ReID) once the tracked object gets lost. Extensive experiments were performed on the KITTI and MOT2015 tracking benchmarks. Results indicate that our approach outperforms most state-of-the-art tracking approaches.