Xin Tan

Xin Tan

Auto-generated profile, based on publicly available information

Xin Tan

Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun, China

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
14 August 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
ZMZhang Mingle
Zhang Mingle
HXHongyu Xie
Hongyu Xie
MYMingyu Yang
Mingyu Yang
QJQingbin Jiao
Qingbin Jiao
LXLiang Xu
Liang Xu
XTXin Tan
Xin Tan
Convolutional neural networks (CNNs) and transformers have demonstrated remarkable capabilities in capturing spatial–spectral contextual dependencies, making them widely applicable for hyperspectral unmixing tasks. However, constrained by limited receptive fields and high computational complexity, these methods often struggle to effectively capture both global information and fine-grained local features, thereby compromising the accuracy of unmixing results. Recently, the Mamba architecture, based on state-space models, has shown great potential in low-dimensional scenarios by combining the advantages of long-range sequence modeling with linear computational efficiency. Inspired by this, we propose a novel dual-stream spatial–spectral feature extraction network named MBUNet, which comprehensively integrates information from four dimensions of hyperspectral images: spatial, spectral, local, and global, thereby achieving more effective unmixing. Specifically, we design a spatial feature extraction module and a spectral feature extraction module, employing CNNs and Mamba as local and global feature extractors, respectively. To fully capture long-range dependencies across the entire image, we introduce a bidirectional Mamba module that enables precise pixel-level global information acquisition and fusion. Through a weight-sharing strategy, the two branches of the dual-stream network can efficiently exchange information while avoiding potential interference caused by traditional feature fusion modules. This design allows both branches to focus on their respective unmixing tasks, with end-to-end collaborative training and mutual correction ultimately yielding more accurate and interpretable results. Extensive experiments on both synthetic and real datasets demonstrate that MBUNet significantly outperforms state-of-the-art unmixing methods in terms of endmember extraction and abundance estimation. If this work is accepted for publication, we will make our code publicly available.

12 January 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
XHXie Hongyu
Xie Hongyu
ZMZhang Mingle
Zhang Mingle
MYMingyu Yang
Mingyu Yang
QJQingbin Jiao
Qingbin Jiao
LXLiang Xu
Liang Xu
XTXin Tan
Xin Tan
Hyperspectral images(HSIs) defocus deblurring and defocus map estimation(DME) are crucial and challenging, especially when dealing with large-scale data. Existing convolutional neural network-based methods exhibit inherent limitations in modeling long-range dependencies, while Transformer-based approaches suffer from high computational complexity when capturing global contextual correlations. Although existing Mamba-based HSI methods (such as SSUMamba and 3DSS-Mamba) have shown promise, they primarily focus on additive noise removal or classification tasks, and their scanning strategies often disrupt the spatial-spectral intrinsic continuity required for modeling spatially varying blurring. To address these challenges, Joint Defocus Map Estimation and deblurring Reconstruction (JDMEDR) framework with the spatial-spectral Transformer Mamba(SSFMamba) modules network is proposed for enhancing the reconstruction of defocused HSIs effectively. The SSFMamba, constructed by hybridizing Mamba context interaction modules with a Spatial-Spectral Layer-by-Layer Scanning (SSLS) strategy, is designed to balance long-range perception and reduction of computation complexity as state-space model. In addition, a DME network is proposed based on reconstructed HSIs adjusted through the defocus generation mechanism by ResNet, enabling the learning of a direct mapping from clean input images to blurred HSIs, thereby further evaluating the effectiveness of the defocused HSI reconstruction algorithm. Extensive experiments conducted on both synthetic and real HSIs datasets demonstrate that the proposed SSFMamba achieves state-of-the-art reconstruction performance and defocus estimation accuracy compared with other state-of-the-art methods.

6 May 2026 in IEEE Access
Verified
Journal Article
Open access
JLJiguo Li
Jiguo Li
XZXin Zhang
Xin Zhang
QJQingbin Jiao
Qingbin Jiao
HHHuansong Huang
Huansong Huang
MGMinghui Gu
Minghui Gu
XTXin Tan
Xin Tan
To enable high-sensitivity and high-stability detection with Fano resonant metasurfaces, the illumination light field must satisfy stringent specifications in three critical aspects: angular sensitivity, wavefront fidelity, and spatial spot matching with the metasurface sensing region. Conventional metasurface illumination systems suffer from several interrelated limitations—including geometric mismatch between the illumination spot and the active sensing area, poor irradiance uniformity, insufficient control over incident angle distribution, wavefront aberrations, and low optical integration density. To address these challenges, this work introduces a purpose-built, high-uniformity illumination optical system tailored specifically for efficient excitation of Fano resonant metasurfaces. Grounded in the principle of optical étendue conservation, the system performs secondary light redistribution of a halogen source using an ellipsoidal reflector. A square-shaped light integrator rod ensures highly uniform spatial intensity distribution, while a custom-designed relay lens group simultaneously accomplishes beam compression, wavefront correction, and strict confinement of the marginal ray exit angles. Experimental validation confirms that the system achieves >95.65% illumination uniformity across the target area, maintains marginal ray angles within ±11°, yields a root-mean-square (RMS) wavefront error of <λ/6, and produces a precisely sized 1.2 mm² square illumination spot aligned to the chip footprint. All output parameters fully comply with the physical excitation requirements of Fano resonant metasurfaces. By combining miniaturization, cost-effectiveness, and high optical performance, this illumination strategy provides a practical and scalable foundation for next-generation metasurface-based biochemical detection platforms.

13 April 2026 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Verified
Journal Article
Open access
ZMZhang Mingle
Zhang Mingle
XHXie Hongyu
Xie Hongyu
QJQingbin Jiao
Qingbin Jiao
LXLiang Xu
Liang Xu
MYMingyu Yang
Mingyu Yang
XTXin Tan
Xin Tan
In hyperspectral imaging systems, stripe noise is a common interference phenomenon, severely affecting the quality and usability of data. Although existing denoising methods have achieved certain progress, most algorithms are still built upon specific model assumptions. However, stripe noise in real-world scenarios often exhibits more complex statistical characteristics and unknown distribution patterns. Such model mismatch problems significantly limit the performance of existing methods in practical applications. Leveraging its powerful feature extraction and data generation capabilities, the variational autoencoder (VAE) can effectively capture the underlying distribution structure of data and generate samples highly consistent with real scenes. To address the issue that short-wave infrared hyper spectral images (HSIs) are highly susceptible to stripe noise induced specifically by dark current interference, this paper proposes a novel VAE-based noise modeling and removal method. Specifically, this study utilizes a small number of real noisy images to learn the true distribution characteristics of stripe noise, thereby generating diverse noise samples. By combining the generated noise with existing high-quality standard datasets, a large-scale paired noisy-clean image dataset is constructed, overcoming the limitations of traditional single noise generation models. On this basis, this paper further proposes an improved stripe noise removal model based on a UNet backbone network. Experimental results demonstrate that the proposed method achieves superior performance on both simulated data and real world scenarios, offering a new solution for stripe noise removal in HSIs.