Hyperspectral gunshot residue detection based on depth-separable convolution and channel attention mechanism

WANG Jiaju; LI Yunpeng; BAO Jiaxin; WU Lianquan; DAI Xuejing; ZHANG Hongguo

您当前的位置：

首页 >

文章列表页 >

Hyperspectral gunshot residue detection based on depth-separable convolution and channel attention mechanism

Optical Image Processing | 更新时间：2026-07-23

- Hyperspectral gunshot residue detection based on depth-separable convolution and channel attention mechanism
- OPTICAL TECHNIQUE Vol. 52, Issue 4, Pages: 500-507(2026)
- 作者机构：
  
  中国刑事警察学院，辽宁沈阳 110854
- 作者简介：
- 基金信息：
- DOI：
  CLC： O433.4
- Received：04 March 2026，
  
  Revised：2026-05-16，
  
  Accepted：20 May 2026，
  
  Online First：23 July 2026，
  
  Published：20 July 2026
- 稿件说明：
移动端阅览
王嘉驹,李云鹏,包佳鑫等.基于深度可分离卷积与通道注意力机制的枪击残留物高光谱图像检测[J].光学技术,2026,52(04):500-507.

WANG Jiaju,LI Yunpeng,BAO Jiaxin,et al.Hyperspectral gunshot residue detection based on depth-separable convolution and channel attention mechanism[J].OPTICAL TECHNIQUE,2026,52(04):500-507.
王嘉驹,李云鹏,包佳鑫等.基于深度可分离卷积与通道注意力机制的枪击残留物高光谱图像检测[J].光学技术,2026,52(04):500-507. DOI：

WANG Jiaju,LI Yunpeng,BAO Jiaxin,et al.Hyperspectral gunshot residue detection based on depth-separable convolution and channel attention mechanism[J].OPTICAL TECHNIQUE,2026,52(04):500-507. DOI：

摘要

在安检与安防现场，复杂纺织物背景下枪击残留物（GSR）的高光谱检测受光谱冗余、噪声干扰与样本不均衡等因素影响，难以实现快速、无损、精准识别。为此，构建了六类不同材质纺织物表面的GSR高光谱数据集，并提出一种融合二维‑三维深度可分离卷积与通道注意力机制的轻量化检测网络（Hybrid DSCDSENet）。方法采用三层级处理框架：首先运用三维深度可分离卷积提取跨波段关联特征，再通过二维深度可分离卷积细化空间结构，最后引入通道注意力机制自适应增强有效谱段、抑制噪声与冗余信息，并结合小波软阈值对输入数据进行预处理。实验结果表明，所提模型在测试集上总体识别精度达到98.55%，相比3D-Res-CNN与RSSAN等典型方法分别提升16.9%和4.91%，在参数量与计算复杂度显著降低的同时，具备更好的鲁棒性与场景泛化能力，为复杂勘查环境中GSR的现场快速识别提供了可行的技术途径。

Abstract

At the security inspection and security protection sites

the high-spectrum detection of gunshot residues （GSR） in complex textile backgrounds is affected by factors such as spectral redundancy

noise interference and sample imbalance

making it difficult to achieve rapid

non-destructive and accurate identification. Therefore

a high-spectrum data set of GSR is constructed on the surfaces of six different types of textile materials

and proposes a lightweight detection network （Hybrid DSCDSENet） that integrates two-dimensional-three-dimensional depth separable convolution and channel attention mechanism. This method adopts a three-level processing framework: first

three-dimensional depth separable convolution is used to extract cross-band correlation features; then

two-dimensional depth separable convolution is used to refine the spatial structure; finally

the channel attention mechanism is introduced to adaptively enhance the effective spectral bands

suppress noise and redundant information

and combine wavelet soft thresholding to preprocess the input data. Experimental results show that the proposed model achieves a total recognition accuracy of 98.55% on the test set

which is 16.9% and 4.91% higher than typical methods such as 3D-Res-CNN and RSSAN

respectively. While significantly reducing the parameter quantity and computational complexity

it has better robustness and scene generalization ability

providing a feasible technical approach for the rapid on-site identification of GSR in complex investigation environments.

关键词

Keywords

references

Ledergerber T D , Barbosa L , Dalzell K , et al . Characterization and classification of organic gunshot residue from neat smokeless powder, cartridge cases, and the hands of known shooters [J ] . Forensic Chemistry , 2025 , 45 : 100683 .

Teng Y , Zhang K , Lv X , et al . Gunshots detection, identification, and classification: Applications to forensic science [J ] . Science & Justice , 2024 , 64 ( 6 ): 625 — 636 .

Medici V , Paone N , Pandarese G , et al . Active Thermography for Gunshot Residue (GSR) pattern estimation on textiles [C ] // 2023 IEEE International Workshop on Technologies for Defense and Security (TechDefense) . Bari, Italy : IEEE , 2023 : 104 — 109 .

Bai L , Chen Z , Zhou L , et al . HSI classification: a comprehensive review, dataset, and future trends [J ] . Journal of the Indian Society of Remote Sensing , 2025 , 53 ( 12 ): 3983 — 4004 .

Deliot P , Corcelle B , Achard V , et al . Preliminary results to define an active multispectral imager for gunshot residues patterns visualization [C ] // 2012 4th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS) . Shanghai, China : IEEE , 2012 : 1 — 4 .

Christina S Atwater , Marie E Durina , John P Durina , et al . Visualization of gunshot residue patterns on dark clothing [J ] . Journal of Forensic Sciences , 2006 ， 51 ( 5 ): 1091 — 1095 .

Giverts P , Sorokina K , Barash M , et al . The use of machine learning for the determination of a type/model of firearms by the characteristics on cartridge cases [J ] . Forensic Science International , 2024 , 358 : 112021 .

Pisantanaroj P , Tanpisuth P , Sinchavanwat P , et al . Automated firearm classification from bullet markings using deep learning [J ] . IEEE Access , 2020 , 8 : 78236 — 78251 .

Głomb P , Romaszewski M , Cholewa M , et al . Application of hyperspectral imaging and machine learning methods for the detection of gunshot residue patterns [J ] . Forensic Science International , 2018 , 290 : 227 — 237 .

李兴 , 孙振文 , 刘耀 . 法庭科学领域烟火药及其爆炸残留物检验研究进展 [J ] . 化学研究与应用 , 2022 , 34 ( 10 ): 2265 — 2273 .

Li Xin , Sun Zhengwen , Liu Yao . Research Progress on pyrotechnic substances and their explosive residues in the Field of Forensic Science [J ] . Chemical Research and Application , 2022 , 34 ( 10 ): 2265 — 2273 .

Shrivastava P , Jain V K , Nagpal S . Gunshot residue detection technologies—a review [J ] . Egyptian Journal of Forensic Sciences , 2021 , 11 : 1 — 21 .

Arun S A , Akila A S . Land-cover classification with hyperspectral remote sensing image using CNN and spectral band selection [J ] . Remote Sensing Applications: Society and Environment , 2023 , 31 : 100986 .

Asker M E . Hyperspectral image classification method based on squeeze-and-excitation networks, depthwise separable convolution and multibranch feature fusion [J ] . Earth Science Informatics , 2023 , 16 ( 2 ): 1427 — 1448 .

魏林 , 陈哲 , 尹玉萍 . 结合空-谱自注意力与多粒度特征提取的高光谱图像分类 [J ] . 激光与光电子学进展 , 2024 , 61 ( 24 ): 220 — 232 .

Wei Ling , Chen Zhe , Yin Yuping . Hyperspectral image classification combining spatial-spectral self-attention and multi-granularity feature extraction [J ] . Advances in Laser & Optoelectronics , 24 , 61 ( 24 ): 220 — 232 .

Sandler M , Howard A G , Zhu M , et al . Mobilenetv2: Inverted residuals and linear bottlenecks [C ] // 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition . Salt Lake City, USA : IEEE , 2018 : 4510 — 4520 .

Rajendran T , Valsalan P , Amutharaj J , et al . Hyperspectral image classification model using squeeze and excitation network with deep learning [J ] . Computational Intelligence and Neuroscience , 2022 , 2022 : 9430779 .

Alfaifi R , Artoli A M . Human action prediction with 3D-CNN [J ] . SN Computer Science , 2020 , 1 ( 5 ): 286 .

Paoletti M E , Haut J M , Fernandez-Beltran R , et al . Deep pyramidal residual networks for spectral-spatial hyperspectral image classification [J ] . IEEE Transactions on Geoscience and Remote Sensing , 2018 , 57 ( 2 ): 740 — 754 .

Li W , Chen H , Liu Q , et al . Attention mechanism and depthwise separable convolution aided 3DCNN for hyperspectral remote sensing image classification [J ] . Remote Sensing , 2022 , 14 ( 9 ): 2215 .

Roy S K , Krishna G , Dubey S R , et al . HybridSN: Exploring 3-D–2-D CNN feature hierarchy for hyperspectral image classification [J ] . IEEE Geoscience and Remote Sensing Letters , 2019 , 17 ( 2 ): 277 — 281 .

Chhapariya K , Ientilucci E J , Buddhiraju K M , et al . A spectral-spatial classification network for hyperspectral images using a residual attention network [C ] // 2023 13th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS) . Athens, Greece : IEEE , 2023 : 1 — 5 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Research on terahertz spectral reconstruction technology of Mail based on MAsLS

3D Confocal Raman Spectroscopic High-Resolution imaging method based on RDCAN deep learning model

Self-supervised gaze estimation method based on feature disentanglement

Study on a method for evaluating geometric tolerances based on monocular rotational vision

A streak image adaptive reconstruction method based on multispectral polarization information fusion

Related Author

LI Tao

WANG Chao

ZHANG Jian

QIAO Dejun

ZHANG Jing

QI Yakun

XU Demin

CUI Han

Related Institution

Hebei Key Laboratory of Optical Fiber Biosensing and Communication Devices, School of Software, Institute of Information Technology, Handan University

Jinan Technician College

MIIT Key Laboratory of Complex-field Intelligent Exploration, School of Optics and Photonics, Beijing Institute of Technology

⁰