Human cognition excels in lifelong learning, seamlessly acquiring, retaining, and transferring knowledge. In contrast, deep neural networks suffer from catastrophic forgetting, where training on new tasks rapidly erases previously acquired knowledge. In the human brain, memory reactivation is crucial for preserving memories over time. Similarly, generative replay in artificial neural networks shows potential for addressing forgetting by reactivating learned representations. However, existing generative replay techniques struggle with complex tasks and high-dimensional data, as they shift the burden from the task-solving network to the generative network, which is also prone to catastrophic forgetting. In this paper, we propose C3GAN, a brain-inspired model that combines Contrastive Clustering and Conditional Generative Adversarial Networks to emulate the memory consolidation processes of the brain. C3GAN uses contrastive class structuring to consolidate recent memories, mimicking hippocampal functions, and incorporates a conditional generative adversarial network to facilitate long-term knowledge storage, akin to the prefrontal cortex. Additionally, an amygdala-inspired module enhances selective replay of indistinguishable classes by prioritizing memory that is emotionally salient, similar to the amygdala's role in strengthening the retention of significant experiences. C3GAN achieves state-of-the-art performance on class-incremental learning benchmarks without raw data, providing a novel solution for lifelong memory retention in artificial systems.

The broadcast nature of device authentication systems in the Internet of Things (IoT) and maritime transportation makes real-time signals susceptible to forgery attacks, leading to critical risks, such as information leakage and navigation failures. specific emitter identification (SEI) plays a crucial role in protecting communication systems from spoofing and tampering. However, traditional SEI networks struggle with limited generalizability, high-computational costs, and susceptibility to extraction and inversion attacks. To overcome these challenges, this article introduces sliding window linear attention and the convolution mechanism (SWLC)-Conformer, a novel hybrid SEI architecture that seamlessly integrates SWLC. The proposed architecture dynamically captures both local signal characteristics and global temporal dependencies through a hybrid design incorporating window-based attention blocks and depth-wise convolution layers. Furthermore, to strengthen system security against model inversion threats, we incorporate a perturbation mechanism that employs reverse sigmoid activation in the final feature representation. Experimental results demonstrate that SWLC-Conformer achieves high recognition accuracy, reaching 91.10% on the AIS100 dataset and 97.89% on the WiSig dataset, while maintaining FLOPs at 22.17M and 11.11M, respectively. In terms of security, SWLC-Conformer reduces the accuracy of stolen models by over 40% when the stolen dataset ratio is 0.9. This result highlights the effectiveness of the introduced defense mechanism in degrading the usability of stolen models through controlled feature obfuscation.

Infrared small target detection (IRSTD) plays a critical role in both civilian and military applications, yet it still faces inherent challenges stemming from faint targets, complex noise interference, and difficulties in preserving shape integrity. Despite significant progress in detecting general small targets, existing methods often struggle to balance detection accuracy and false alarms due to limited sensitivity to low-intensity signals, inaccurate perception of confusing noise, and inadequate edge refinement. To break this dilemma, we propose ISGLNet, which centers on a U-shaped architecture specifically tailored to preserve salient target responses, along with a guided learning strategy that progressively enhances target–noise distinction while refining boundary details. Specifically, we introduce the Context-aware Local-Global Module (CLGM) as the cornerstone of the model, which incorporates multi-branch large receptive fields and multi-dimensional adaptive attention mechanisms, effectively capturing rich contexts while preserving critical target information. This ensures reliable feature modeling throughout the extraction and fusion process. Furthermore, the Multi-frequency Perception Module (MFPM) and the Edge Refinement Module (ERM) replace conventional skip connections to refine semantic patterns through guidance. Among these, the MFPM operates in the deeper layers, primarily identifying discriminative clues by evaluating and dynamically selecting multi-frequency information to amplify the distinction between targets and complex noise. The ERM further works in the shallower layers with a progressive strategy to refine uncertain target boundaries, enabling precise segmentation of fine-grained target shapes. Extensive experiments on multiple public datasets demonstrate that ISGLNet achieves superior performance in both detection and segmentation accuracy. The code is available at https://github.com/fuqingzhang/ISGLNet.

The L-band passive microwave data from the Soil Moisture Active Passive (SMAP) observatory are investigated for remote sensing of ocean surface winds during severe storms. The surface winds of Joaquin derived from the real-time analysis of the Center for Advanced Data Assimilation and Predictability Techniques at Penn State support the linear extrapolation of the Aquarius and SMAP geophysical model functions (GMFs) to hurricane force winds. We apply the SMAP and Aquarius GMFs to the retrieval of ocean surface wind vectors from the SMAP radiometer data to take advantage of SMAP's two-look geometry. The SMAP radiometer winds are compared with the winds from other satellites and numerical weather models for validation. The root-mean-square difference (RMSD) with WindSat or Special Sensor Microwave Imager/Sounder is 1.7 m/s below 20-m/s wind speeds. The RMSD with the European Center for Medium-Range Weather Forecasts direction is 18° for wind speeds between 12 and 30 m/s. We find that the correlation is sufficiently high between the maximum wind speeds retrieved by SMAP with a 60-km resolution and the best track peak winds estimated by the National Hurricane Center and the Joint Typhoon Warning Center to allow them to be estimated by SMAP with a correlation coefficient of 0.8 and an underestimation by 8%-18% on average, which is likely due to the effects of spatial averaging. There is also a good agreement with the airborne Stepped-Frequency Radiometer wind speeds with an RMSD of 4.6 m/s for wind speeds in the range of 20-40 m/s.