Visible spectrum images capture limited information from just three discrete bands, often resulting in suboptimal performance in underwater depth estimation (UDE) due to significant information loss from water absorption. In contrast, HSIs, which include hundreds of continuous bands, provide abundant spectral information that offers greater resilience against the adverse effects of water absorption. In this paper, we conduct a comprehensive study to investigate how spectral information can enhance remote sensing UDE through two key aspects: the benchmark dataset and the general framework. For the benchmark dataset, we construct a real-world hyperspectral UDE (HUDE) dataset ATR-HUDE, comprising approximately 500 synchronized hyperspectral and LiDAR data pairs collected from diverse coastal scenes and flight altitudes. Regarding the general framework, we integrate recent advances in state space models and physical imaging models to design a novel HUDE framework named HUDEMamba that estimates underwater depth using both model-driven and data-driven approaches. Experimental results on the constructed benchmark dataset validate the potential of HUDE and the effectiveness of HUDEMamba.

Cross-Modality Re-Identification (VT-ReID) aims to achieve around-the-clock target matching, benefiting from the strengths of both RGB and infrared (IR) modalities. However, the field is hindered by limited datasets, particularly for vehicle VT-ReID, and by challenges such as modality bias training (MBT), stemming from biased pre-training on ImageNet. To tackle the above issues, this paper introduces an dataset benchmark, named UCM-VeID V2, for vehicle VT-ReID, and proposes a new self-supervised pre-training method, Cross-Modality Patch-Mixed Self-Supervised Learning (PMSL). UCM-VeID V2 dataset features a significant increase in data volume, along with enhancements in multiple aspects. PMSL addresses MBT by learning modality-invariant features through Patch-Mixed Image Reconstruction (PMIR) and Modality Discrimination Adversarial Learning (MDAL), and enhances discriminability with Modality-Augmented Contrasting Cluster (MACC). Comprehensive experiments are carried out to validate the proposed method.

Unmanned aerial vehicles (UAV)-based object detection with visible (RGB) and infrared (IR) images facilitates robust around-the-clock detection, driven by advancements in deep learning techniques and the availability of high-quality dataset. However, the existing dataset struggles to fully capture real-world complexity for limited imaging conditions. To this end, we introduce a high-diversity dataset ATR-UMOD covering varying scenarios, spanning altitudes from 80m to 300m, angles from 0° to 75°, and all-day, all-year time variations in rich weather and illumination conditions. Moreover, each RGB-IR image pair is annotated with 6 condition attributes, offering valuable high-level contextual information. To meet the challenge raised by such diverse conditions, we propose a novel prompt-guided condition-aware dynamic fusion (PCDF) to adaptively reassign multimodal contributions by leveraging annotated condition cues. By encoding imaging conditions as text prompts, PCDF effectively models the relationship between conditions and multimodal contributions through a task-specific soft-gating transformation. A prompt-guided condition-decoupling module further ensures the availability in practice without condition annotations. Experiments on ATR-UMOD dataset reveal the effectiveness of PCDF.

Hyperspectral imaging (HSI) provides fine-grained spectral information essential for material identification and target detection, particularly in complex environments such as underwater scenarios. However, hyperspectral underwater target detection (HUTD) remains challenging due to severe spectral distortions and variability introduced by wavelength-dependent absorption and the dynamic nature of aquatic environments. Existing separation-based and characterization-based methods are often constrained by weak signal responses or a heavy reliance on accurate environmental parameter estimation, which is difficult to achieve in practice. To overcome these limitations, we propose PCL-HUTD, a novel physics-informed curriculum learning framework for robust underwater target characterization without requiring explicit environmental modeling. PCL-HUTD integrates a physics-guided target construction module with a hard-sample aware contrastive learning strategy, enhanced by unsupervised clustering and a perturbation-consistency based sample selection mechanism. Furthermore, a closed-loop curriculum learning paradigm is introduced to progressively refine target representations throughout training. Extensive experiments on three real-world HUTD datasets demonstrate that PCL-HUTD achieves state-of-the-art performance in both detection accuracy and robustness, particularly under challenging conditions with strong background interference. These results validate the effectiveness of our parameter-free, physics-informed approach for underwater hyperspectral target detection.

Visible-infrared (RGB-IR) image fusion has shown great potentials in object detection based on unmanned aerial ve-hicles (UAVs). However, the weakly misalignment problem between multimodal image pairs limits its performance in object detection. Most existing methods often ignore the modality gap and emphasize a strict alignment, resulting in an upper bound of alignment quality and an increase of implementation costs. To address these challenges, we propose a novel method named Offset-guided Adaptive Feature Alignment (OAFA), which could adaptively adjust the relative positions between multimodal features. Considering the impact of modality gap on the cross-modality spa-tial matching, a Cross-modality Spatial Offset Modeling (CSOM) module is designed to establish a common sub-space to estimate the precise feature-level offsets. Then, an Offset-guided Deformable Alignment and Fusion (ODAF) module is utilized to implicitly capture optimal fusion po-sitions for detection task rather than conducting a strict alignment. Comprehensive experiments demonstrate that our method not only achieves state-of-the-art performance in the UAVs-based object detection task but also shows strong robustness to the weakly misalignment problem.

Owing to the capacity of performing full-time target searches, cross-modality vehicle re-identification based on unmanned aerial vehicles (UAV) is gaining more attention in both video surveillance and public security. However, this promising and innovative research has not been studied sufficiently due to the issue of data inadequacy. Meanwhile, the cross-modality discrepancy and orientation discrepancy challenges further aggravate the difficulty of this task. To this end, we pioneer a cross-modality vehicle Re-ID benchmark named UAV Cross-Modality Vehicle Re-ID (UCM-VeID), containing 753 identities with16015RGB and13913infrared images. Moreover, to meet cross-modality discrepancy and orientation discrepancy challenges, we present a hybrid weights decoupling network (HWDNet) to learn the shared discriminative orientation-invariant features. For the first challenge, we proposed a hybrid weights siamese network with a well-designed weight restrainer and its corresponding objective function to learn both modality-specific and modality shared information. In terms of the second challenge, three effective decoupling structures with two pretext tasks are investigated to flexibly conduct orientation-invariant feature separation task. Comprehensive experiments are carried out to validate the effectiveness of the proposed method.

The research on attack transferability is of great importance as it can guide how to conduct an adversarial attack without knowing any information about target models. However, it remains challenging for adversarial examples to maintain a good attack transferability performance, especially for the black-box attack implemented in the physical world. To enhance black-box transferability of physical attacks on object detectors, we present a novel adversarial learning method to produce adversarial patches by redistributing separable attention maps. Concretely, we first develop smoothed multilayer attention maps by introducing serial composite transformations, which could suppress model-specific noise on the one hand, and cover objects to be concealed at various resolutions on the other hand. Besides, our method resorts to a scalable mask to separate object attention from the background and adjust their distribution with a novel loss function. Extensive experiments show that our approach outperforms state-of-the-art methods in both the digital space and the physical world. Our code is available at https://github.com/zhangyu13a/transPhyAtt .

Many deep learning methods have been proposed to classify moving targets from seismic signals in recent years. However, the existing deep models are all designed based on the “end-cloud” framework, in which real-time data processing is difficult because of communication delays. To address this problem and achieve on-site target classification, we propose a novel edge intelligence-oriented method, named compressed sensing-edge convolutional neural network (CS-ECNN). In this method, the acquired seismic signals are first mapped onto a compressed domain using CS. This operation reduces data dimensions, while being able to retain the vast majority of valuable seismic features. Following that, a convolutional neural network is employed to extract implicit features directly from the compressed seismic measurements and then classify the feature vectors. To evaluate the proposed method, the seismic data recorded in DARPA’s SensIT project are used as a case study. The experimental results demonstrate that the proposed model is edge-matched, and it achieves comparable classification accuracy to the state-of-the-art cloud-based models with only 1/10 computation time.

Due to the strong nonstationary characteristics of seismic signals, energy criteria-based methods are not robust for detecting moving targets, especially in data with low SNRs. To address this problem, we propose a new method for detecting ground moving target based on fractal dimension (FD) theory named FD-based support vector machine (FD-SVM). In this method, seismic signals are first measured by fractals, which can effectively extract seismic nonlinear features. These fractal features are then fed into an SVM to distinguish moving targets from noise. Two data sets are used to evaluate the proposed method. One is a set of seismic signals induced by wheeled and tracked vehicles. The other is a set of seismic signals generated by human footsteps. Experimental results demonstrate that the proposed FD-SVM algorithm achieves promising results on both data sets. Compared with the benchmark methods, the FD-SVM algorithm achieves a better precision rate, recall rate, and F1 score.

Moving target recognition is a critical task for a variety of applications, ranging from environmental monitoring to regional security protection. Recently, many deep learning (DL) methods have been proposed to recognize the seismic features of moving targets. However, the established DL algorithms are mainly challenged by time-consuming feature extraction and lack of robustness. In this article, a novel moving target recognition method [Compression Observation-Seismic DL (CO-SDL)] is proposed to solve the above two problems simultaneously. CO-SDL first uses a measurement matrix to project the seismic signal onto a compressed domain and obtain compressed seismic measurements. This operation removes redundant data while retaining valuable seismic information and suppressing noise energy. Following that, CO-SDL efficiently and stably extracts deep nonlinear features from compressed seismic measurements and then accurately classifies the feature vectors. To evaluate the proposed method, a comprehensive seismic dataset is developed. This dataset covers six types of common moving targets, and the SNR ranges of all signal types are greater than 15 dB. The proposed method and the benchmark methods are tested on this dataset. Experimental results prove that the presented CO-SDL method is ten times faster than the state-of-the-art methods with comparable accuracy. Furthermore, the CO-SDL method shows the strongest robustness.

In modern seismic data acquisition, real-time data collection is a challenging task due to bandwidth limitations in wireless communications. In this letter, we propose a novel compressive data gathering scheme using generative adversarial networks, named GAN-CDG, to improve the efficiency of data gathering. Instead of collecting the originally acquired data, GAN-CDG gathers data projections in wireless geophone networks. Data compression and load-balanced relay transmission are utilized during the projection process. To speed up the formation of projections, the shortest path routing tree (SPRT) is constructed, which achieves the minimum end-to-end time delay. The sparse domain of seismic signals and its reconstruction mapping are learned by sparsity-constrained adversarial networks. The testing results demonstrate that projections with high compression ratios (e.g., 16) are gathered efficiently with the SPRT. Then, original seismic signals can be reconstructed accurately (over 30 dB) from the projections using the adversarial model, which outperforms the state-of-the-art method.