[1]孙丽丽,冯丽佳.基于深度神经网络的试卷批改标记检测与识别研究[J].机械与电子,2026,44(09):34-43.
　SUN Lili,FENG Lijia.Detection and Recognition of Examination Paper Marking Traces Based on Deep Neural Networks[J].Machinery & Electronics,2026,44(09):34-43.
点击复制

基于深度神经网络的试卷批改标记检测与识别研究()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年09期
页码:
34-43
栏目:
智能检测
出版日期:
2026-09-25

文章信息/Info

Title:
Detection and Recognition of Examination Paper Marking Traces Based on Deep Neural Networks
文章编号:
1001-2257(2026)09-0034-10
作者:
孙丽丽1; 冯丽佳2
1.河北省教育考试院,河北 石家庄 050091;
2.河北工程技术学院,河北 石家庄 050091
Author(s):
SUN Lili1; FENG Lijia2
(1.Hebei Education Examinations Authority,Shijiazhuang 050091,China;
2.Hebei University of Engineering Science,Shijiazhuang 050091,China)
关键词:
试卷批改标记; 频域注意力; 轻量化网络; 损失函数优化
Keywords:
examination paper marking traces; frequency-domain attention; lightweight network; loss function optimization
分类号:
TP391.4
文献标志码:
A
摘要:
针对试卷批改标记形态多样、尺度差异大和易受印刷文字等复杂背景干扰等问题,提出一种融合频域增强与Ghost轻量化的尺度形状解耦试卷批改标记检测算法FGSD-YOLO。首先,提出频率增强混合注意力模块(FHAT),通过对特征图振幅谱施加通道级自适应加权,增强与批改标记相关的特征响应,抑制印刷文字等背景干扰。其次,引入GhostConv轻量化卷积,通过半数通道标准卷积与轻量逐通道线性变换相结合,有效压缩模型参数量与计算开销。最后,提出高斯尺度形状解耦IoU 损失函数(GSSD-IoU),以高斯衰减形式重构中心距离惩罚并将形状惩罚解耦为宽度与高度2个独立约束项,提升小目标与异形标记的回归精度。实验结果表明:FGSD-YOLO 在Examination Paper Mark Trace数据集上的mAP@0.5达到97.1%,较基准模型提升4.3百分点;在Examination Paper Multiple Choice数据集上的mAP@0.5达到98.4%,较基准模型提升3.3百分点;同时参数量仅为2.8×106,GFLOPs为7.4,实现了检测精度与模型大小的平衡,并具有良好的泛化能力、稳定性和复杂场景适应能力。
Abstract:
To address the challenges posed by diverse morphology,large scale variations,and complex background interference (e.g.,printed text) in examination paper marking traces,FGSD-YOLO,a scale-shape decoupled detection algorithm that integrates frequency domain enhancement and Ghost-based lightweight design,is proposed in this study.First,a Frequency-enhanced Hybrid Attention module (FHAT) is introduced,which applies channel-wise adaptive weighting to the amplitude spectrum of feature maps,thereby enhancing feature responses relevant to marking traces while suppressing background interference such as printed text.Second,GhostConv lightweight convolution is adopted,combining standard convolution on half of the channels with lightweight pointwise linear transformations,effectively reducing model parameters and computational overhead.Finally,a Gaussian Scale-Shape Decoupled IoU loss (GSSD-IoU) is proposed,which reconstructs the center distance penalty via Gaussian decay and decouples the shape penalty into two independent constraints for width and height,thereby improving regression accuracy for small targets and irregularly shaped marks.Experimental results demonstrate that FGSD-YOLO achieves an mAP@0.5 of 97.1% on the Examination Paper Mark Trace dataset,outperforming the baseline model by 4.3 percentage points,and an mAP@0.5 of 98.4% on the Examination Paper Multiple Choice dataset,exceeding the baseline by 3.3 percentage points.Moreover,the model has only 2.8×106 parameters and the computational cost is 7.4 GFLOPs,achieving a favorable trade-off between detection accuracy and model compactness,while exhibiting strong generalization capability,robustness,and adaptability to complex scenarios.

参考文献/References:

[1] 刘伊薇.基于深度学习的试卷手写信息检测与识别[D].北京:北京邮电大学,2022.

[2] 张晨雪.基于深度学习的试卷手写体检测与识别算法研究[D].西安:西安工业大学,2025.
[3] Ali E R,Shati N M.Evaluation system for multiple-choice questions using optical mark recognition:a survey[ J].Journal of Al Qadisiyah for Computer Science and Mathematics,2025,17(1):202-213.
[4] 尹玲,李家乐,黄勃.基于改进YOLOv5s的文档图像版面分析算法[J].软件导刊,2025,24(2):146-154.
[5] 李晓辉,周宇,彭良瑞,等.从文本识别到智能文档解析的技术演进:“文档图像微沙龙”系列活动前沿综述[J/OL].中国图象图形学报,2026-06-16.https:∥link.cnki.net/urlid/11.3758.tb.20260615.1635.002.
[6] 杨大志,王维兰,李洪瑞,等.边缘感知特征融合的藏文报刊文档图像版面分析[J].高原科学研究,2025,9(1):129-140.
[7] Liu Wei,Anguelov D,Erhan D,et al.SSD:Single shot multibox detector[C]∥ Computer Vision – ECCV 2016,2016:21-37.
[8] Redmon J,Divvala S,Girshick R,et al.You only look once:unified,real-time object detection[C]∥ 2016 IEEE Conference on Computer Vision and PatternRecognition (CVPR 2016).New York:IEEE,2016:779-788.
[9] Ren Shaoqing,He Kaiming,Girshick R,et al.Faster R-CNN:towards real-time object detection with region proposal networks[J].IEEE Transactions on PatternAnalysis and Machine Intelligence,2016,39 (6):1137-1149.
[10] 王宇飞,王毅,韩晓萌,等.基于图像分割的自然场景文本检测算法研究[J].航天控制,2026,44(1):28-33.
[11] 张涵,徐丽格,胡东浩,等.基于深度学习的印章文本检测与识别[J].华南师范大学学报(自然科学版),2025,57(6):82-90.
[12] 吕林浩,邓春华.基于三维姿态估计的自然场景中文文本识别[J].计算机工程与设计,2025,46(12):3397-3405.
[13] 薛冀杰,高光勇,高云龙.基于注意力融合的文本检测方法[J].重庆邮电大学学报(自然科学版),2026,38(1):177-186.
[14] 邢沛然,仁青东主,尼玛扎西,等.自然场景下端到端藏文检测与识别方法研究[J/OL].计算机工程与应用,2025-10-21.https:∥link.cnki.net/urlid/11.2127.TP.20251020.1802.017.
[15] Mondal S,De P,Malakar S,et al.OMRNet:a lightweight deep learning model for optical mark recognition[ J].Multimedia Tools and Applications,2024,83(5):14011-14045.
[16] Labrassi M,Ouaarab A.Improved deep learning-based optical mark recognition for automated correction[C]∥Intelligent Data Engineering and Automated Learning-IDEAL 2025,2025:324-330.
[17] Hossain M I,Sifat J A,Chowdhury A,et al.Edge-native digitization of handwritten marksheets:a hybrid heuristic-deep learning framework[C]∥2026 IEEE 2nd International Conference on Quantum Photonics,ArtificialIntelligenceand Networking (QPAIN).New York:IEEE,2026:1-6.
[18] Binmakhashen G M,Mahmoud S A.Document layout analysis:a comprehensive survey[J].ACM Computing Surveys,2019,52(6):1-36.
[19] Qin Zequn,Zhang Pengyi,Wu Fei,et al.Fcanet:frequency channel attention networks[C]∥2021 IEEE/CVF International Conference on Computer Vision (ICCV).New York:IEEE,2021:783-792.
[20] Rao Yongming,Zhao Wenliang,Zhu Zheng,et al.Global filter networks for image classification[J].Advances in Neural Information Processing Systems, 2021,34:980-993.
?[21] Chen Linwei,Fu Ying,Gu Lin,et al.Frequency-aware feature fusion for dense image prediction[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2024,46(12):10763-10780.
[22] Chen Jierun,Kao Shiuhong,He Hao,et al.Run,don’t walk:chasing higher FLOPS for faster neural networks[C]∥ 2023 IEEE/CVF Conference on ComputerVision and Pattern Recognition (CVPR).New York:IEEE,2023:12021-12031.
[23] Qin Danfeng,Leichner C,Delakis M,et al.MobileNetV4: Universal models for the mobile ecosystem [C]∥Computer Vision – ECCV 2024,2024:78-96.
[24] Chen Xiangyu,Wang Xintao,Zhou Jiantao,et al.Activating more pixels in image super resolution transformer[C]∥2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).New York:IEEE,2023:22367-22377.
[25] Han Kai,Wang Yunhe,Tian Qi,et al.Ghostnet:more features from cheap operations[C]∥ 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2020).New York:IEEE,2020:1580-1589.
[26] 徐威.客观题批改与主观题批阅痕迹识别的算法研究及系统实现[D].武汉:华中科技大学,2022.
[27] Zhao Yian,Lv Wenyu,Xu Shangliang,et al.Detrs beat yolos on real-time object detection[C]∥ 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2024).New York:IEEE,2024:16965-16974.
[28] Redmon J,Farhadi A.YOLOv3:an incremental improvement[ PP/OL].arXiv (2018-04-08)[2026-07-28].https:∥arxiv.org/abs/1804.02767.
[29] Wang C Y,Yeh I H,Liao H Y M.YOLOv9:learning what you want to learn using programmable gradient information[C]∥ Computer Vision – ECCV 2024, 2024:1-21.
[30] Wang Ao,Chen Hui,Liu Lihao,et al.YOLOv10:real time end-to-end object detection[J].Advances in Neural Information Processing Systems,2024,37:107984-108011.
[31] Khanam R,Hussain M.YOLOv11:an overview of the key architectural enhancements[PP/OL].arXiv (2024-10-23)[2026-07-28].https:∥arxiv.org/abs/2410.17725.

备注/Memo

备注/Memo:
收稿日期:2026-06-23
基金项目:河北省教育科学“十四五”规划2025年度教育决策咨询专项课题(2504039)
作者简介:孙丽丽 (1984-),女,河北石家庄人,硕士,高级工程师,研究方向为神经网络;冯丽佳 (1984-),女,河北石家庄人,硕士,副教授,研究方向为信息技术教育应用,通信作者,E-mail:19933165829@163.com。
更新日期/Last Update: 2026-10-12