Infrared video small object detection is pivotal in numerous security and surveillance applications. However, existing deep learning-based methods, which typically rely on a two-step paradigm of frame-by-frame detection followed by temporal refinement, struggle to effectively utilize temporal information. This is particularly challenging when detecting small objects against complex backgrounds. To address these issues, we introduce the One-Step Transformer (OSFormer), a novel method that pioneeringly integrates a small-object-friendly transformer with a one-step detection paradigm. Unlike traditional methods, OSFormer processes the video sequence only through a single inference, encoding the sequence into cube format data and tracking object motion trajectories. Additionally, we propose the Varied-Size Patch Attention (VPA) module, which generates patches of varying sizes to capture adaptive attention features, bridging the gap between transformer architectures and small object detection. To further enhance detection accuracy, OSFormer incorporates a Doppler Adaptive Filter, which integrates traditional filtering techniques into an end-to-end neural network to suppress background noise and accentuate small objects. OSFormer outperforms YOLOv8-s on both the AntiUAV dataset (+3.1% mAP⁵⁰, -35.1% Params) and the InfraredUAV dataset (+4.0% mAP⁵⁰⁻⁹⁵, -51.0% FLOPs), demonstrating superior efficiency and effectiveness in small object detection.

In recent years, multispectral object detection technology has gained widespread attention due to its significant performance in detecting objects with similar colors or textures in complex environments. The mainstream methods adopt the fusion of visible light (RGB) and thermal images to make up for the shortcomings of a single modality, improving the detection accuracy. However, most methods fail to take the inherent differences between modalities into account. RGB and thermal images differ in object attributes, which may lead to inconsistent contribution of each modality to the fusion features. Hence, equally feeding them into the feature extractor will limit the expressiveness of the fusion features. In order to reasonably utilize complementary information cues of each modality, an effective cross-modality feature fusion network is proposed in this letter. It comprises a spectral-spatial enhancing module (SSE) and a feature fusion module via Transformer (FFT).For dual-modality data, in the aerial dataset, our model’s detection accuracy metrics, mAP50 and mAP, are respectively improved by 1.2% and 0.9% compared to the best dual-modality network.Comprehensive experiments on both ground and aerial datasets demonstrate that our approach outperforms existing methods. The achievements are of great significance for enhancing the robustness and accuracy of multispectral object detection.

Hyperspectral remote sensing images provide rich spectral information about land surface features and are widely used in fields such as environmental monitoring, disaster assessment, and land classification. However, effectively leveraging the spectral information in hyperspectral images remains a significant challenge. In this paper, we propose a hyperspectral pansharpening method based on ratio transformation and residual networks, which significantly enhances both spatial details and spectral fidelity. The method generates an initial image through ratio transformation and refines it using a residual attention network. Additionally, specialized loss functions are designed to preserve both spatial and spectral details. Experimental results demonstrate that, when evaluated on the EO-1 and Chikusei datasets, the proposed method outperforms other methods in terms of both visual quality and quantitative metrics, particularly in spatial detail clarity and spectral fidelity. This approach effectively addresses the limitations of existing technologies and shows great potential for high-resolution remote sensing image processing applications.