Diabetic retinopathy (DR), a serious complication of diabetes, is one of the primary causes of vision loss in retinal vascular diseases. While deep learning has been widely used for DR grading, their performance declines significantly when applied to data outside the training distribution due to domain shifts. Domain generalization (DG) addresses this challenge, yet most existing DG methods neglect lesion-specific features, limiting diagnostic accuracy.In this paper, we propose a novel approach that enhances existing DG methods by incorporating structural priors, inspired by the observation that DR grading is heavily dependent on vessel and lesion structures. We introduce Low-rank Adaptive Structural Priors (LoASP), a plug-and-play framework designed for seamless integration with existing DG models. LoASP improves generalization by learning adaptive structural representations that are finely tuned to the complexities of DR diagnosis.Extensive experiments on eight diverse datasets validate its effectiveness in both single-source and multisource domain scenarios. Visualizations show the learned priors align with vessel/lesion structures, enhancing interpretability and diagnostic relevance.

Synthetic Aperture Radar (SAR) is a type of active microwave remote sensing imaging radar that can generate two-dimensional high-resolution images. Its ability to operate in all weather conditions and at all times has led to its widespread use. As a multi-parameter and multi-channel extension of SAR, polarimetric SAR (PolSAR) provides a wealth of scattering information for various applications, including topographic mapping, ocean exploration, polar observation, and target identification. Compared with single-polarization SAR, multi-polarization SAR enhances the information potential of the data by expanding its one-dimensional information, however, this potential cannot be fully realized without a clean SAR echo signal. The electromagnetic environment is becoming increasingly congested with radio frequency interference (RFI) signals, presenting a significant challenge for the subsequent tasks of PolSAR. Although there have been many related studies based on polarization information to carry out the aforementioned applications, there is a lack of research on the joint suppression of interference by using multi-polarization information, and single-polarization data alone is insufficient in effectively mitigating interference. To address these challenges, this paper presents a framework combining multi-polarization data to improve performance of low-rank based methods. Based on the proposed framework, experiments are conducted on real PolSAR data to assess the feasibility of the proposed framework in interference suppression. The results demonstrate that the framework significantly enhances the suppression performance of various low-rank based methods with clearer scene details being recovered.

With the surge of vehicles and transportation, sensing obstacles and warning drivers to avoid accidents have become a great concern in recent years. In the current roadworthy electromagnetic environment, the number of frequency modulated continuous wave (FMCW) millimeter-wave (MMW) automotive radars has exploded due to their unique advantages in environmental sensing. However, the frequency band of the automotive radars is limited from 77 to 81 GHz, hence the burgeoning of radars on the road is bound to cause mutual interference and jeopardize further target detection and parameter estimation. In this paper, two basic schemes are considered to mitigate mutual interference of automotive radars. First, we consider the sparse characteristics of the mutual interference in the two-dimensional (2-D) time domain and employ a sparse interference extraction (SIE) method to tackle the mutual interference. Next, we further consider the low-rank property of the useful echoes across multiple channels and propose a novel three-dimensional (3-D) tensor decomposition (TD) method to decompose the received signals into mutual interference and useful echoes. Several numerical simulations are fulfilled to test the robustness of the proposed TD method, especially for multiple input and multiple output (MIMO) systems under complex electromagnetic circumstances. Furthermore, more experiments are implemented to demonstrate its feasibility in practical applications in comparison to multiple state-of-the-art methods.

The virtual middle platform, which encapsulates original application data and functions without migration, addresses challenges arising from insufficient trust relationships among capability providers. To resolve the capability selection problem in non-trusted environments, we propose a meta-inspired capability selection algorithm (MHCSA) based on homomorphic cryptography, and propose an improvement scheme to optimize the computation latency. Experimental results demonstrate that MHCSA significantly outperforms other meta-inspired algorithms in terms of homomorphic computation delay, providing an effective solution for non-trusted virtual middle platform scenarios.

Synthetic aperture radar (SAR) is a kind of active imaging radar, which can obtain high-resolution wide-swath SAR images, especially for spaceborne SAR systems. In practical electromagnetic environment, due to the overlap of same frequency bands, spaceborne SAR is extremely vulnerable to interferences from other electromagnetic systems, called radio frequency interference (RFI) to SAR systems. RFI seriously reduces the imaging quality of the SAR system and causes resolution reduction and scene occluded. To mitigate RFI in SAR systems, researchers have proposed many methods, in which semi-parametric methods, such as robust principal component analysis (RPCA)-based methods, played important roles in strong RFI mitigation in recent years. However, it is observed that they may be hard to recover the true scene well under extremely strong RFIs since the strong scatterers are also mixed in the extracted low-rank interferences. Therefore, in this paper, we propose a novel adaptive method, which combines the advantages of both semi-parametric method and frequency domain notched filter (FNF) method, called adaptive notch semi-parametric (ANSP) method, where the FNF method, as a non-parametric method, can retain more true scenes when mitigating interferences. As a result, the proposed method can not only effectively deal with strong RFIs but also protect the strong scatterers better with an adaptive threshold. This method can recover the true scene under extremely strong RFI and be applied to both Level-0 and Level-1 SAR data. Finally, we conduct experiments on several real SAR data and demonstrate the effectiveness of the proposed method.