Chaoyang Niu

Synthetic Aperture Radar (SAR) and optical imagery exhibit strong complementarity, which can meet the needs of real-time Earth observation under extreme conditions representing a cutting-edge research direction in the field of Multimodal Change Detection (MCD). However, Multimodal Remote Sensing Images (MRSI) encounter a significant domain gap due to differences in imaging mechanisms. At the same time, due to the scarcity of high-quality annotations, existing deep learning models typically suffer from weak generalization and limited detection accuracy. This paper proposes a lightweight MCD framework named the Heuristic Teaching and Hypergraph Structural Constraints Multimodal Change Detection Network (HT-HSCNet), which leverages the powerful prior knowledge of Multimodal Vision Foundation Models (MVFMs) to train a lightweight change detection model without introducing additional inference overhead. Specifically, a Deep Semantic Knowledge Injection (DSKI) module is introduced, employing a heuristic teaching strategy to dynamically inject rich semantic features from a frozen MVFM into a trainable lightweight CNN backbone, thereby enhancing its representational capacity without compromising inference efficiency. Furthermore, to explicitly model and utilize the inherent and consistent structural spatial layout in unchanged regions of multimodal images, a Hypergraph-based Structure Knowledge Enhance Module (HSKEM) is proposed to capture complex spatial topological relationships among features. Experimental results on 3 optical-to-SAR change benchmarks, California, XiongAn and OSCD(multi-sensor) demonstrate that the proposed method achieves significant improvements across multiple key evaluation metrics. Meanwhile, HT-HSCNet achieves minimal parameters and computational overhead, striking a balance between high performance and efficiency, thereby offering a novel perspective for addressing intelligent MCD challenges. The code will be released at https://github.com/JackW0820/HT-HSCNet https://github.com/JackW0820/HT-HSCNet.

When lightweight backbones pretrained on natural scene datasets are applied to object detection in aerial remote sensing images (ARSIs), the detection performance varies significantly. This variation stems from factors including the data domain gap, dataset scale, training configuration, and model architecture. Since remote sensing foundation models (RSFMs) are pretrained on large-scale remote sensing datasets and exhibit strong feature extraction capabilities, we propose an RSFM-based auxiliary framework to enable existing lightweight backbones to achieve enhanced performance on ARSI datasets of varying scales. Specifically, the RSFM is leveraged to efficiently extract features with robust and rich representational capabilities from input ARSIs. A foundation feature fusion module is designed to fuse the features extracted by the RSFM with those from the lightweight backbone, addressing the inadequacy in representational capacity of various lightweight backbones when extracting ARSI features. Furthermore, a feature aggregation and expansion module is introduced to enhance the representational power of the fused features. Experimental results on four ARSI datasets of different scales demonstrate that the performance of various lightweight backbones is improved when integrated with the proposed RSFM-based auxiliary framework. In most cases, this performance is superior to that of larger-scale networks. Specifically, on the DOTA 1.5 and DIOR datasets, the performance of these lightweight backbones (integrated with the framework) is significantly enhanced compared to that of state-of-the-art (SOTA) models. Collectively, these results validate the effectiveness of the proposed RSFM-based auxiliary framework, confirming its ability to effectively improve the performance of existing backbones and thereby facilitate the popularization and application of existing technical advancements.

With the rapid development of UAV technology, the illegal use of UAVs poses a serious threat to public safety, and the feature extraction of UAV echo signals is crucial for target identification, however the micro-Doppler feature extraction based on convolutional neural network has the problem of insufficient feature extraction, which has a greater impact on the accurate identification of UAV targets . In this paper, a new UAV recognition network model is proposed , named Wavelet Collaborative and Effective Channel Spatial Attention Network (WTCESNet),which takes RepViT as the backbone network, and designs a collaborative and effective channel spatial attention module (CES),which is capable of fine-tuning the micro-Doppler features of the echo signals through the synergistic effect of effective channel attention and spatial attention. At the same time, wavelet convolution is introduced to expand the sensory field of the model feature extraction, which effectively improves the network’s recognition ability. Finally, a hybrid loss function is designed, which consists of a focus loss and a label smoothing loss function . Avoid the error due to sample imbalance. The experimental results show that the proposed method achieves accuracy rates of 95.98 % (15 dB) , 93.76% (10 dB) , and 86.11% (-10 dB) under different signal-to-noise ratios compared with Vit, Swin Transformer, and other methods.

The deep unfolding network (DUN) for image compressive sensing (ICS) integrates a traditional optimization algorithm with a neural network, providing clear interpretability and demonstrating exceptional performance. Nevertheless, the inherent paradigm of the DUN lies in the independent proximal mapping between iterations and the limited information flux, potentially constraining the mapping capability of the deep unfolding method. This paper introduces a Feature-Domain FISTA-Inspired Deep Unfolding Network (FDFI-DUN) for ICS. FDFI-DUN comprises a Feature-Domain Nesterov Momentum Module (FNMM), a Feature-Domain Gradient Descent Module (FGDM), and a Two-level Multiscale Proximal Mapping Module (TMPMM). Specifically, the Nesterov momentum term and gradient descent term in the FISTA are tailored to the feature domain, enhancing the information flux of the entire DUN and augmenting the feature information within and between iterations while maintaining clear interpretability. Furthermore, the TMPMM, encompassing intra-stage and inter-stage components, is designed to further augment the information flux and effectively utilize multiscale feature information for reconstructing image details. Extensive experimental results demonstrate that the proposed FDFI-DUN surpasses state-of-the-art methods in both quality and vision. Our codes are available at: https://github.com/giant-pandada/FDFI-DUN.

Scene-aware Adaptive Compressive Sensing (ACS) has attracted significant interest due to its promising capability for efficient and high-fidelity acquisition of scene images. ACS typically prescribes adaptive sampling allocation (ASA) based on previous samples in the absence of ground truth. However, when confronting unknown scenes, existing ACS methods often lack accurate judgment and robust feedback mechanisms for ASA, thus limiting the high-fidelity sensing of the scene. In this paper, we introduce a Sampling Innovation-Based ACS (SIB-ACS) method that can effectively identify and allocate sampling to challenging image reconstruction areas, culminating in high-fidelity image reconstruction. An innovation criterion is proposed to judge ASA by predicting the decrease in image reconstruction error attributable to sampling increments, thereby directing more samples towards regions where the reconstruction error diminishes significantly. A sampling innovation-guided multi-stage adaptive sampling (AS) framework is proposed, which iteratively refines the ASA through a multi-stage feedback process. For image reconstruction, we propose a Principal Component Compressed Domain Network (PCCD-Net), which efficiently and faithfully reconstructs images under AS scenarios. Extensive experiments demonstrate that the proposed SIB-ACS method significantly outperforms the state-of-the-art methods in terms of image reconstruction fidelity and visual effects. Codes are available at https://github.com/giant-pandada/SIB-ACS_CVPR2025.

Synthetic aperture radar (SAR) automatic target recognition (ATR) systems based on deep neural network models are vulnerable to adversarial examples. Existing SAR adversarial attack algorithms require access to the network structure, parameters, and training data, which are often inaccessible in real-world scenarios. To address this problem, this study proposes an intermediate layer feature attack algorithm that does not rely on training data for the adversary model. Electromagnetic simulation is used to obtain the simulated SAR local data domain during the training stage. A lightweight generator, TinyResNet, is introduced to quickly construct adversarial examples through a one-step mapping process. In addition, the transferability of these examples across different models is improved by eliminating the intermediate layer features of the model. Finally, a domain-agnostic feature attention module is utilized to reduce discrepancies between different data domains from a model perspective, further improving the transferability of adversarial examples across domains. Experimental results on five SAR datasets of ground vehicles, ships, and scene types demonstrate that the proposed algorithm outperforms 13 mainstream adversarial attack algorithms in terms of cross-model and cross-data domain transferability. In particular, the proposed method improves the cross-domain attack success rate by 43.74%–48.88% on the MSTAR dataset.

Synthetic aperture radar automatic target recognition (SAR-ATR) models based on deep neural networks (DNNs) are vulnerable to attacks of adversarial examples. Universal adversarial attack algorithms can help evaluate and improve the robustness of the SAR-ATR models and have become a research hotspot. However, current universal adversarial attack algorithms have limitations. First, considering the difficulty in obtaining information on the attacking SAR-ATR models, there is an urgent need to design a universal adversarial attack algorithm under a black-box scenario. Second, given the difficulty of acquiring SAR images, the effectiveness of attacks under small-sample conditions requires improvement. To address these limitations, this study proposed a black-box universal adversarial attack algorithm: Transferable Universal Adversarial Network (TUAN). Based on the idea of the generative adversarial network, we implemented the game of generator and attenuator to improve the transferability of universal adversarial perturbation (UAP). We designed loss functions for the generator and attenuator respectively, which can effectively improve the success rate of black-box attacks and the stealthiness of attacks. Additionally, U-Net was used as a network structure of generator and attenuator to fully learn the distribution of examples, thereby enhancing the attack success rate under small-sample conditions. TUAN attained a higher black-box attack success rate and superior stealthiness than up-to-date UAP algorithms in non-targeted and targeted attacks.

The extensive application of unmanned aerial vehicle (UAV) technology has increased academic interest in object detection algorithms for UAV images. Nevertheless, these algorithms present issues such as low accuracy, inadequate stability, and insufficient pre-training model utilization. Therefore, a high-quality object detection method based on a performance-improved object detection baseline and pretraining algorithm is proposed. To fully extract global and local feature information, a hybrid backbone based on the combination of convolutional neural network (CNN) and vision transformer (ViT) is constructed using an excellent object detection method as the baseline network for feature extraction. This backbone is then combined with a more stable and generalizable optimizer to obtain high-quality object detection results. Because the domain gap between natural and UAV aerial photography scenes hinders the application of mainstream pre-training models to downstream UAV image object detection tasks, this study applies the masked image modeling (MIM) method to aerospace remote sensing datasets with a lower volume than mainstream natural scene datasets to produce a pre-training model for the proposed method and further improve UAV image object detection accuracy. Experimental results for two UAV imagery datasets show that the proposed method achieves better object detection performance compared to state-of-the-art (SOTA) methods with fewer pre-training datasets and parameters.