一种轻量化空谱双分支Transformer网络用于白细胞偏振高光谱图像分类

杨雨诗; 晏佳轩; 谢奕; 邱礼佳; 黄丹飞

doi:10.11999/JEIT260124

一种轻量化空谱双分支Transformer网络用于白细胞偏振高光谱图像分类

doi: 10.11999/JEIT260124 cstr: 32379.14.JEIT260124

杨雨诗^{1, 2},
晏佳轩^{1, 2},
谢奕^{1, 2},
邱礼佳³,
黄丹飞^{1, 2, ,}

1.
长春理工大学光电工程学院长春 130022
2.
长春理工大学中山研究院中山 528400
3.
南方医科大学珠江医院潮南医院汕头 515144

基金项目: 国家自然科学基金(62105245)

详细信息

作者简介:
杨雨诗：女，硕士生，研究方向为计算机视觉与图像处理技术

晏佳轩：男，硕士生，研究方向为光电检测与图像分析

谢奕：男，博士生，研究方向为光电检测与图像分析

邱礼佳：男，检验师，主要从事临床血液学检验工作

黄丹飞：女，教授，研究方向为计算机视觉与图像处理技术

通讯作者:
黄丹飞　hdanfei@163.com

中图分类号: O433.4; TP183; R331.1+42
计量
- 文章访问数: 206
- HTML全文浏览量: 89
- PDF下载量: 13
- 被引次数: 0
出版历程
- 收稿日期: 2026-01-30
- 修回日期: 2026-06-29
- 录用日期: 2026-06-29
- 网络出版日期: 2026-07-08

A Lightweight Spatial-Spectral Dual-Branch Transformer Network for Classifying Polarized White Blood Cell Hyperspectral Images

YANG Yushi^{1, 2},
YAN Jiaxuan^{1, 2},
XIE Yi^{1, 2},
QIU Lijia³,
HUANG Danfei^{1, 2
, ,}

1.
College of Optoelectronic Engineering, Changchun University of Science and Technology, Changchun, 130022, China
2.
Zhongshan Institute, Changchun University of Science and Technology, Zhongshan, 528400, China
3.
Chaonan Hospital of Zhujiang Hospital, Southern Medical University, Shantou, 515144, China

Funds: National Natural Science Foundation of China (62105245)

摘要

摘要: 白细胞类型与形态是血常规分析的关键，然而传统镜检方法依赖主观经验，现有深度学习方法在空谱特征建模和计算效率方面仍存在不足。为此，该文将偏振高光谱成像技术应用于白细胞分析，提出一种轻量化空谱双分支Transformer网络(LSDBT)用于白细胞偏振高光谱图像分类。该网络通过双分支特征提取模块联合学习局部空间和空间−光谱特征，并结合轻量化设计降低计算复杂度。结果表明，LSDBT在实现高精度分类的同时兼顾模型轻量化，验证了偏振高光谱成像在白细胞分类中的应用价值，为临床辅助诊断提供高效可靠的技术方案。
- 偏振高光谱成像 /
- 白细胞分类 /
- Swin Transformer
Abstract: Objective White Blood Cell (WBC) classification and morphology are essential in routine blood analysis and provide important information for disease diagnosis and health assessment. Current WBC classification mainly relies on hematology analyzers and manual microscopic examination. Hematology analyzers cannot acquire cellular images, limiting classification accuracy when abnormal cellular characteristics are present, whereas manual microscopic examination depends on operator experience and is susceptible to human error. Although automated WBC classification based on deep learning and computer vision has attracted considerable attention, methods using stained images are sensitive to staining conditions and image quality. In addition, conventional hyperspectral imaging has limited ability to distinguish WBC subtypes with highly similar morphological and spectral characteristics. To address these limitations, this study combines Polarized Hyperspectral Imaging (PHSI) with deep learning and proposes a lightweight classification framework for polarized hyperspectral WBC images, providing an efficient and reliable approach for clinical decision support. Methods A polarized hyperspectral microscopic imaging system is established to acquire images at multiple polarization angles. Based on Stokes Vector Theory, polarization parameters are calculated to generate multidimensional data cubes containing both light intensity and polarization-state information. Degree of Linear Polarization (DOLP) images are then computed to construct a PHSI dataset of WBCs. To exploit the multidimensional characteristics of PHSI, a Lightweight Spatial-Spectral Dual-Branch Transformer Network (LSDBT) is proposed. After preprocessing, the input data are fed into a dual-branch feature extraction module that separately extracts local spatial features and joint spatial-spectral features. An adaptive scaling factor is introduced to fuse the two feature streams and balance their contributions, enabling effective utilization of the multidimensional information contained in PHSI. A lightweight Swin Transformer backbone performs effective global feature modeling while reducing computational complexity. Global average pooling and a fully connected layer are used for classification. Model performance is evaluated using Overall Accuracy (OA), Precision, Recall, Specificity, and F1-Score. Ablation studies, comparative experiments, and feature visualization are conducted to validate the proposed method. Results and Discussions The DOLP spectra and PHSI visualizations of monocytes, lymphocytes, and neutrophils (Figures. 5 and 6) demonstrate different polarization characteristics that reflect the selective absorption and scattering of light by their internal structures. Compared with conventional intensity images, PHSI improves image contrast and provides additional polarization information that enhances discrimination among WBC types. The proposed LSDBT achieves an OA of 99.29% on the test set, with consistently high classification performance across all cell categories (Table 1). Analysis of the adaptive scaling factor shows that classification performance first improves and then decreases slightly as the scaling factor increases, with the optimal value of 3 providing the best balance between spatial and spatial-spectral features (Figure. 8). Ablation experiments (Tables 2 and 3) demonstrate that the dual-branch feature extraction module substantially improves classification performance, whereas the lightweight design greatly reduces computational complexity and model parameters with only a marginal reduction in accuracy. Compared with conventional hyperspectral imaging, the PHSI dataset achieves higher classification accuracy with all evaluated classifiers, indicating that polarization information provides complementary physical features that improve discrimination among WBC types (Table 4). Comparisons with representative methods show that LSDBT achieves the best overall classification performance across multiple evaluation metrics (Table 5). Furthermore, t-SNE visualization (Figure. 9) shows compact intra-class distributions and clear separation among different cell types, confirming the strong discriminative capability of the learned features. Although LSDBT does not have the lowest computational cost among the compared methods, it achieves the best balance between classification performance, model size, and computational efficiency (Table 6). Conclusions This study proposes a lightweight dual-branch Transformer network for polarized hyperspectral WBC classification. To the best of our knowledge, this is the first study to combine PHSI with deep learning for WBC classification. Comparative experiments with conventional hyperspectral imaging validate the superiority of PHSI for WBC classification. The proposed LSDBT integrates spatial and spatial-spectral information through a dual-branch feature extraction module and performs efficient global feature modeling using a lightweight Swin Transformer backbone. The network maintains high classification performance while substantially reducing computational complexity and model parameters. These results demonstrate that LSDBT provides an accurate and computationally efficient solution for automated WBC classification and supports the application of PHSI in cellular microscopic analysis and clinical auxiliary diagnosis.
- Polarized Hyperspectral Imaging /
- White Blood Cell Classification /
- Swin Transformer

HTML全文

图 1 白细胞偏振高光谱显微成像系统

下载: 全尺寸图片幻灯片

图 2 LSDBT网络框架

下载: 全尺寸图片幻灯片

图 3 双分支特征提取模块

下载: 全尺寸图片幻灯片

图 4 SwinT窗口划分和计算自注意力的移位窗口方法示意图

下载: 全尺寸图片幻灯片

图 5 白细胞线偏振度谱

下载: 全尺寸图片幻灯片

图 6 白细胞偏振高光谱图像可视化结果

下载: 全尺寸图片幻灯片

图 7 混淆矩阵

下载: 全尺寸图片幻灯片

图 8 不同缩放因子的分类性能比较

下载: 全尺寸图片幻灯片

图 9 深层特征的t−SNE可视化

下载: 全尺寸图片幻灯片

表 1 LSDBT在3种类型白细胞上的高光谱图像分类结果(%)

	Precision	Recall	Specificity	F1−Score
Monocytes	99.34	99.07	99.67	99.21
Lymphocytes	99.60	99.47	99.80	99.54
Neutrophils	98.95	99.34	99.47	99.14

下载: 导出CSV

表 2 消融实验结果

Swin Transformer	轻量化	空间−光谱特征提取模块	双分支特征提取模块	OA (%)
√	×	×	×	97.70
√	√	×	×	97.66 (–0.04)
√	×	√	×	98.77 (+1.07)
√	×	×	√	99.34 (+1.64)
√	√	√	×	98.72 (+1.02)
√	√	×	√	99.29 (+1.59)

下载: 导出CSV

表 3 模型轻量化前后的计算复杂度

模型	参数量(M)	GFLOPs	内存大小(MB)	OA (%)
双分支特征提取+SwinT	28.32	9.10	108.24	99.34
LSDBT	1.42	3.31	5.49	99.29

下载: 导出CSV

表 4 高光谱数据集与偏振高光谱数据集分类结果

数据集	模型	OA (%)
	KNN	93.1
偏振高光谱数据集	RF	95.6
	SVM	94.9
	KNN	91.2
高光谱数据集	RF	93.8
	SVM	93.3

下载: 导出CSV

表 5 不同模型的分类结果(%)

模型	类别	Precision	Recall	Specificity	F1−Score	OA
本文LSDBT	Monocytes	99.3	99.1	99.7	99.2	99.2 (±0.1)
	Lymphocytes	99.6	99.5	99.8	99.5
	Neutrophils	99.0	99.3	99.5	99.1
MambaHSI^[8]	Monocytes	99.2	94.9	99.6	97.0	95.4 (±1.4)
	Lymphocytes	92.8	99.3	96.2	95.9
	Neutrophils	98.5	96.0	99.3	97.2
HSIFormer^[13]	Monocytes	98.9	97.4	99.5	98.2	98.4 (±0.1)
	Lymphocytes	97.1	99.6	98.5	98.4
	Neutrophils	99.3	98.2	99.6	98.7
SGTFormer^[22]	Monocytes	99.3	97.8	99.6	98.5	97.3 (±0.3)
	Lymphocytes	97.8	96.7	98.9	97.2
	Neutrophils	96.1	98.5	98.0	97.3
MACLST^[23]	Monocytes	98.5	97.8	99.3	98.2	97.7 (±0.4)
	Lymphocytes	97.5	98.9	98.7	98.2
	Neutrophils	98.2	97.4	99.1	97.8

下载: 导出CSV

表 6 不同模型计算成本分析

模型	参数量 (M)	GFLOPs	内存大小 (M)	OA (%)
本文LSDBT	1.42	3.31	5.49	99.2 (±0.1)
MambaHSI	0.12	0.70	0.46	95.4 (±1.4)
HSIFormer	0.53	1.32	2.14	98.4 (±0.1)
SGTFormer	219.45	13.85	837.13	97.3 (±0.3)
MACLST	12.70	4.10	48.40	97.7 (±0.4)

下载: 导出CSV

参考文献(23)

[1]	林宇晖. 基于人工智能的血常规分析技术在临床应用中的研究进展[J]. 中国冶金工业医学杂志, 2025, 42(4): 384–385. doi: 10.13586/j.cnki.yjyx1984.2025.04.066. LIN Yuhui. Research progress on the clinical application of artificial intelligence−based complete blood count analysis technology[J]. Chinese Medical Journal of Metallurgical Industry, 2025, 42(4): 384–385. doi: 10.13586/j.cnki.yjyx1984.2025.04.066.
[2]	阮受杰. 比较全自动血液分析仪与血涂片镜检在血常规检验中的准确性差异[J]. 中国医疗器械信息, 2025, 31(12): 88–90. doi: 10.15971/j.cnki.cmdi.2025.12.018. RUAN Shoujie. Comparison of accuracy difference between automatic blood analyzer and blood smear microscopic examination in blood routine test[J]. China Medical Device Information, 2025, 31(12): 88–90. doi: 10.15971/j.cnki.cmdi.2025.12.018.
[3]	温晶. 全自动血细胞分析仪联合血涂片细胞形态学在血常规检验中的应用效果[J]. 中国医疗器械信息, 2025, 31(12): 91–93. doi: 10.15971/j.cnki.cmdi.2025.12.017. WEN Jing. Application effect of automatic hematology analyzer combined with blood smear cell morphology in routine blood tests[J]. China Medical Device Information, 2025, 31(12): 91–93. doi: 10.15971/j.cnki.cmdi.2025.12.017.
[4]	ASGHAR R, KUMAR S, SHAUKAT A, et al. Classification of white blood cells (leucocytes) from blood smear imagery using machine and deep learning models: A global scoping review[J]. PLoS One, 2024, 19(6): e0292026. doi: 10.1371/journal.pone.0292026.
[5]	WIBAWA M S. A comparison study between deep learning and conventional machine learning on white blood cells classification[C]. 2018 International Conference on Orange Technologies (ICOT), Nusa Dua, Indonesia, 2018. doi: 10.1109/ICOT.2018.8705892.
[6]	蔡轶珩, 谭美伶, 潘建军, 等. 基于多尺度非对称密集网络的高光谱图像分类[J]. 电子与信息学报, 2024, 46(4): 1448–1457. doi: 10.11999/JEIT230651. CAI Yiheng, TAN Meiling, PAN Jianjun, et al. Hyperspectral image classification based on multi−scale asymmetric dense network[J]. Journal of Electronics & Information Technology, 2024, 46(4): 1448–1457. doi: 10.11999/JEIT230651.
[7]	XIE Yi, HUANG Danfei, YAN Jiaxuan, et al. Integrated laser speckle contrast and multispectral imaging for in vivo multimodal assessment of small intestinal viability[J]. Journal of Photochemistry and Photobiology B: Biology, 2025, 272: 113264. doi: 10.1016/j.jphotobiol.2025.113264.
[8]	LI Yapeng, LUO Yong, ZHANG Lefei, et al. MambaHSI: Spatial-spectral mamba for hyperspectral image classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5524216. doi: 10.1109/tgrs.2024.3430985.
[9]	廖帝灵, 赖涛, 黄海风, 等. LightMamba: 一种轻量级Mamba用于高光谱图形和激光雷达数据联合分类网络[J]. 电子与信息学报, 2025, 47(12): 4937–4947. doi: 10.11999/JEIT250981. LIAO Diling, LAI Tao, HUANG Haifeng, et al. LightMamba: A lightweight mamba network for the joint classification of HSI and LiDAR data[J]. Journal of Electronics & Information Technology, 2025, 47(12): 4937–4947. doi: 10.11999/JEIT250981.
[10]	DUAN Puhong, JIN Shiyu, LU Xiaotian, et al. Domain-adaptive mamba for cross-scene hyperspectral image classification[J]. IEEE Transactions on Image Processing, 2026, 35: 1206–1217. doi: 10.1109/TIP.2026.3657209.
[11]	MAHDAVIPOUR Z, ALROMEMA N, KHADER A, et al. TGDHTL: Hyperspectral image classification via transformer-graph convolutional network-diffusion with hybrid domain adaptation[J]. Remote Sensing, 2026, 18(2): 189. doi: 10.3390/rs18020189.
[12]	赵凤, 耿苗苗, 刘汉强, 张俊杰, 於俊. 卷积神经网络与视觉Transformer联合驱动的跨层多尺度融合网络高光谱图像分类方法[J]. 电子与信息学报, 2024, 46(5): 2237-2248. doi: 10.11999/JEIT231209 doi: 10.11999/JEIT231209. ZHAO Feng, GENG Miaomiao, LIU Hanqiang, ZHANG Junjie, YU Jun. Convolutional Neural Network and Vision Transformer-driven Cross-layer Multi-scale Fusion Network for Hyperspectral Image Classification[J]. Journal of Electronics & Information Technology, 2024, 46(5): 2237-2248. doi: 10.11999/JEIT231209.
[13]	ALKHATIB M Q and JAMALI A. HSIFormer: An efficient vision transformer framework for enhanced hyperspectral image classification using local window attention[C]. The 14th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS), Helsinki, Finland, 2024: 1–5. doi: 10.1109/WHISPERS65427.2024.10876533.
[14]	ZHOU Ximing, MA Ling, HALICEK M, et al. Development of a new polarized hyperspectral imaging microscope[C]. SPIE 11213, Imaging, Therapeutics, and Advanced Technology in Head and Neck Surgery and Otolaryngology, San Francisco, USA, 2020: 1121308. doi: 10.1117/12.2549676.
[15]	SONG Yuan, HUANG Danfei, WANG Tianyi, et al. Polarization imaging method for detection of lung cancer based on symmetrical single LCVR[J]. Journal of Biophotonics, 2025, 18(3): e202400470. doi: 10.1002/jbio.202400470.
[16]	WOESSNER A E, MCGEE J D, JONES J D, et al. Characterizing differences in the collagen fiber organization of skin wounds using quantitative polarized light imaging[J]. Wound Repair and Regeneration, 2019, 27(6): 711–714. doi: 10.1111/wrr.12758.
[17]	ZHOU Ximing, MA Ling, MUBARAK H K, et al. Polarized hyperspectral microscopic imaging system for enhancing the visualization of collagen fibers and head and neck squamous cell carcinoma[J]. Journal of Biomedical Optics, 2024, 29(1): 016005. doi: 10.1117/1.JBO.29.1.016005.
[18]	GHASSEMI P, TRAVIS T E, MOFFATT L T, et al. A polarized multispectral imaging system for quantitative assessment of hypertrophic scars[J]. Biomedical Optics Express, 2014, 5(10): 3337–3354. doi: 10.1364/BOE.5.003337.
[19]	DREMIN V, MARCINKEVICS Z, ZHEREBTSOV E, et al. Skin complications of diabetes mellitus revealed by polarized hyperspectral imaging and machine learning[J]. IEEE Transactions on Medical Imaging, 2021, 40(4): 1207–1216. doi: 10.1109/TMI.2021.3049591.
[20]	ZHOU Ximing, MUBARAK H K, MA Ling, et al. Polarized hyperspectral microscopic imaging for white blood cells on Wright−stained blood smear slides[C]. SPIE 12382, Polarized Light and Optical Angular Momentum for Biomedical Diagnostics, San Francisco, USA, 2023: 1238207. doi: 10.1117/12.2655708.
[21]	LIU Ze, LIN Yutong, CAO Yue, et al. Swin transformer: Hierarchical vision transformer using shifted windows[C]. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, Canada, 2021: 9992–10022. doi: 10.1109/ICCV48922.2021.00986.
[22]	DENG Xiangyu, PAN Lihao, and DANG Zhiyan. SGTformer: Improved shifted window transformer network for white blood cell subtype classification[J]. Journal of Electronic Imaging, 2024, 33(4): 043057. doi: 10.1117/1.JEI.33.4.043057.
[23]	Farooque G, Liu Q, Sargano A B, et al. Swin transformer with multiscale 3D atrous convolution for hyperspectral image classification[J]. Engineering Applications of Artificial Intelligence,2023,126: 107070. doi: 10.1016/j.engappai.2023.107070.