Jianye Yuan

Adversarial attacks present a severe threat to the trustworthiness of remote sensing uncrewed systems. Existing detection methods are mostly limited to binary classification, lacking fine-grained traceability and relying heavily on adversarial example (AE) training. To overcome these challenges, we propose a training-free hierarchical adversarial attack traceability framework leveraging the zero-shot transfer ability of contrastive language-image pretraining (CLIP), eliminating the dependency on AE training. For the first time, we establish a multigranularity attack propagation path analysis system. Specifically, the framework leverages CLIP for fine-grained adversarial attack classification without fine-tuning, constructing a hierarchical traceability system (HTS) from coarse-grained to fine-grained semantic levels. Based on Bayesian inference, we design a hierarchical probability fusion method that improves coarse-grained results through maximum a posteriori (MAP) estimation of fine-grained classifiers, collaboratively optimizing hierarchical probability distributions. To capture frequency-domain characteristics of adversarial attacks, we propose a frequency-aware kernel approximation combining high-frequency-enhanced radial basis function (RBF) kernels with ridge regression, improving sensitivity to subtle perturbations in the reproducing kernel Hilbert space (RKHS). This study establishes a comprehensive traceability system for attack propagation chains, providing an interpretable, training-free paradigm for remote sensing security, significantly enhancing fine-grained detection capabilities of uncrewed systems. Extensive experiments on two public remote sensing datasets, RSSSN7 and DWEFS, and one proprietary dataset demonstrate that our method significantly improves adversarial classification accuracy across two victim models, achieving gains of 17.13%/18.89%/13.44% and 17.91%/19.58%/12.17%, respectively, compared with state-of-the-art (SOTA) training-free baselines.

Existing aerial forest fire monitoring data primarily consist of infrared or visible light images. However, there is a lack of in-depth research on the ability of models to perceive fire regions across different spectral images. To address this, we first constructed a dataset of infrared and visible light images captured in the same scene, from the same perspective, and at the same time, with pixel-level segmentation annotations of the flame regions in the images. In response to the issues of poor flame segmentation performance in the current fire images and the large number of learnable parameters in large models, we propose an improved large model algorithm, SAMFA (Segmentation Anything Model, Fire, Adapter). Firstly, while freezing the original parameters of the large model, only the additionally incorporated Adapter module is fine-tuned to better adapt the network to the specificities of the flame segmentation task. Secondly, to enhance the network’s perception of flame edges, a U-shaped mask decoder is designed. Lastly, to reduce the training difficulty, a progressive strategy combining self-supervised and fully supervised learning is employed to optimize the entire model. We compared SAMFA with five state-of-the-art image segmentation algorithms on a labeled public dataset, and the experimental results demonstrate that SAMFA performs the best. Compared to SAM, SAMFA improves the IoU by 11.94% and 6.45% on infrared and visible light images, respectively, while reducing the number of learnable parameters to 11.58 M.

Fire detection is crucial due to the exorbitant annual toll on both human lives and the economy resulting from fire-related incidents. To enhance forest fire detection in complex environments, we propose a new algorithm called FD-Net for various environments. Firstly, to improve detection performance, we introduce a Fire Attention (FA) mechanism that utilizes the position information from feature maps. Secondly, to prevent geometric distortion during image cropping, we propose a Three-Scale Pooling (TSP) module. Lastly, we fine-tune the YOLOv5 network and incorporate a new Fire Fusion (FF) module to enhance the network’s precision in identifying fire targets. Through qualitative and quantitative comparisons, we found that FD-Net outperforms current state-of-the-art algorithms in performance on both fire and fire-and-smoke datasets. This further demonstrates FD-Net’s effectiveness for application in fire detection.

With rising global temperatures, wildfires frequently occur worldwide during the summer season. The timely detection of these fires, based on unmanned aerial vehicle (UAV) images, can significantly reduce the damage they cause. Existing Convolutional Neural Network (CNN)-based fire detection methods usually use multiple convolutional layers to enhance the receptive fields, but this compromises real-time performance. This paper proposes a novel real-time semantic segmentation network called FireFormer, combining the strengths of CNNs and Transformers to detect fires. An agile ResNet18 as the encoding component tailored to fulfill the efficient fire segmentation is adopted here, and a Forest Fire Transformer Block (FFTB) rooted in the Transformer architecture is proposed as the decoding mechanism. Additionally, to accurately detect and segment small fire spots, we have developed a novel Feature Refinement Network (FRN) to enhance fire segmentation accuracy. The experimental results demonstrate that our proposed FireFormer achieves state-of-the-art performance on the publicly available forest fire dataset FLAME—specifically, with an impressive 73.13% IoU and 84.48% F1 Score.

Infrared and visible image fusion aims to fuse the thermal information of infrared images and the texture information of visible images into images that are more in compliance with people’s visual perception characteristics. However, in the existing related work, the fused images have incomplete contextual information and poor fusion results. This paper presents a new image fusion algorithm—OMOFuse. At first, both the channel and spatial attention mechanisms are optimized by a DCA (dual-channel attention) mechanism and an ESA (enhanced spatial attention) mechanism. Then, an ODAM (optimized dual-attention mechanism) module is constructed to further improve the integration effect. Moreover, a MO module is used to improve the network’s feature extraction capability for contextual information. Finally, there is the loss function ℒ from the three parts of SSL (structural similarity loss), PL (perceptual loss), and GL (gap loss). Extensive experiments on three major datasets are performed to demonstrate that OMOFuse outperforms the existing image fusion methods in terms of quantitative determination, qualitative detection, and superior generalization capabilities. Further evidence of the effectiveness of our algorithm in this study are provided.

Carbon observation satellites based on passive theory (e.g., OCO-2/3, GOSAT-1/2, and TanSat) have relatively high carbon dioxide column concentration (XCO ₂ ) accuracy when the observation conditions are met. Passive satellites have data bias and coverage deficiencies due to cloud cover, low albedo, low light conditions, and aerosol scattering, resulting in carbon observation satellites based on passive theory that cannot meet the demand for high-precision, all-day, all-weather XCO ₂ monitoring. Active detection satellites are urgently needed to support global carbon sources, sinks, and carbon neutrality. China intends to launch a sensor satellite with active detection of XCO ₂ in the coming years. In this work, based on the satellite’s scaled-down airborne experiments, a spectral energy model was developed to optimize the conventional inversion algorithm and achieve a more accurate XCO ₂ inversion. The 1.572 μm IPDA lidar column length is used indirectly to evaluate the accuracy of the spectral energy model for signal extraction. And the experimental results show that the accuracy of the signal extracted by the 1.572 μm IPDA lidar column length is 0.74 and 6.20 m at sea and on land based on the indirect evaluation of the length of the 1.572 μm IPDA lidar column length. The optimized XCO ₂ was evaluated (standard deviation as an evaluation metric) and its XCO ₂ standard deviation reduced by 31%, 63% and 66% in the ocean, plains and mountains, respectively. Our algorithm can obtain the XCO ₂ with a consistent trend by using XCO ₂ from the OCO-2 satellite as a reference. The calculated XCO ₂ is more accurate in areas dominated by anthropogenic factors (plains), owing to the accuracy of the IPDA detection mechanism. This algorithm improves the accuracy and robustness of XCO ₂ inversion and has important reference significance for the IPDA lidar carried by China’s satellites to be launched in this year.

With the increasing frequency of natural disasters becoming, it is very important to classify and identify disasters. We propose a lightweight disaster classification model, which has lower computation and parameter quantities and a higher accuracy than other classification models. For this purpose, this paper specially proposes the SDS-Network algorithm, which is optimized on ResNet, to deal with the above problems of remote sensing images. First, it implements the spatial attention mechanism to improve the accuracy of the algorithm; then, the depth separable convolution is introduced to reduce the number of model calculations and parameters while ensuring the accuracy of the algorithm; finally, the effect of the model is increased by adjusting some hyperparameters. The experimental results show that, compared with the classic AlexNet, ResNet18, VGG16, VGG19, and Densenet121 classification models, the SDS-Network algorithm in this paper has a higher accuracy, and when compared with the lightweight models mobilenet series, shufflenet series, squeezenet series, and mnasnet series, it has lower model complexity and a higher accuracy rate. According to a comprehensive performance comparison of the charts made in this article, it is found that the SDS-Network algorithm is still better than the regnet series algorithm. Furthermore, after verification with a public data set, the SDS-Network algorithm in this paper is found to have a good generalization ability. Thus, we can conclude that the SDS-Network classification model of the algorithm in this paper has a good classification effect, and it is suitable for disaster classification tasks. Finally, it is verified on public data sets that the proposed SDS-Network has good generalization ability and portability.