密集异形截面建材智能计数：专用模型及大模型对比研究

刘思媛; 陈宇韬; 黄茜; 李洋; 陈隽

doi:10.3724/j.gyjzG24071003

密集异形截面建材智能计数：专用模型及大模型对比研究

doi: 10.3724/j.gyjzG24071003

刘思媛¹,
陈宇韬¹,
黄茜²,
李洋³,
陈隽^2, ,

1. 新疆大学建筑工程学院, 乌鲁木齐 830047;
2. 同济大学土木工程学院, 上海 200092;
3. 中国联合工程有限公司, 杭州 310052

基金项目:

中国机械工业集团有限公司青年科技基金项目（QNJJ-PY-2025-22）。

详细信息

作者简介:
刘思媛，硕士研究生，主要从事智能建造方面的研究。

通讯作者:
陈隽，博士，教授，主要从事工程结构大数据防灾方面的研究，cejchen@tongji.edu.cn。

计量
- 文章访问数: 23
- HTML全文浏览量: 7
- PDF下载量: 0
- 被引次数: 0
出版历程
- 收稿日期: 2024-07-10
- 网络出版日期: 2026-07-06

Intelligent Counting of Densely-Packed Building Materials with Special-Shaped Sections： a Comparative Study of Specialized Models and Large Models

1. College of Architecture and Civil Engineering, Xinjiang University, Urumqi 830047, China;
2. College of Civil Engineering, Tongji University, Shanghai 200092, China;
3. China United Engineering Co. Ltd., Hangzhou 310052, China

摘要

摘要: 通过建立专用模型和基于大模型二次开发两种方式，对比研究了异形截面建材智能计数问题。首先，实地拍摄并标注了大量的角钢与轮扣图片，结合图片增强构建了基础数据集。进而，通过对经典YOLOv8框架引入挤压与激励网络（SE）注意力机制、WIoU损失函数、动态蛇形卷积以及轻量化网络架构改进等措施，实现了对密集异形截面建材的高精度计数，在现场环境下模型对角钢与轮扣的检测平均精度分别能达到91.8%与99.4%，具有很好的实际应用效果。随后，采用相同的数据集在大模型平台EasyDL上进行了异形截面检测模型的二次开发。对比结果表明，目前基于大模型平台二次开发模型的检测精度、训练效率与展示效果清晰度均低于专用模型，但由于其开发技术门槛低、建模便捷且通用性强，在未来通用类计数任务的开发中仍是一种非常有前景的方式。
- 目标检测 /
- 智能建造 /
- 建材计数 /
- EasyDL
Abstract: In this paper， the intelligent counting problem of special-shaped section construction materials was comparatively studied through two methods： establishing a dedicated model and conducting secondary development based on large models. First， a large number of images of angle steel and wheel locks were taken on-site and labeled， and a basic dataset was constructed combined with image enhancement. Furthermore， by introducing measures such as the SE attention mechanism， WIoU loss function， dynamic snake convolution， and lightweight network architecture improvement to the classic YOLOv8 framework， high-precision counting of densely arranged special-shaped section construction materials was achieved. The average detection accuracy of the model in this paper for angle steel and wheel locks in the field environment reached 91.8% and 99.4%， respectively， showing good practical application effects. Subsequently， the same dataset was used for the secondary development of the special-shaped section detection model on the large model platform EasyDL. The comparison results showed that currently， the detection accuracy， training efficiency， and clarity of display effect of the secondary development model based on the large model platform were lower than those of the dedicated model. However， due to its low development technical threshold， convenient modeling， and strong versatility， it remained a very promising approach for the development of future general counting tasks.
- object detection /
- intelligent construction /
- building material counting /
- EasyDL

HTML全文

参考文献(20)

[1]	刘育江.智能建造与建筑工业化协同发展影响因素探究[J].广东建材，2024，40（6）:159-161.
[2]	靳惠怡.十部门印发《绿色低碳转型产业指导目录（2024年版）》多项涉及建材领域[J].中国建材，2024（3）:29-30.
[3]	ZHANG D，XIE Z，WANG C.Bar section image enhancement and positioning method in on-line steel bar counting and automatic separating system[C]//2008 Congress on Image and Signal Processing.Sanya:2008:319-323.
[4]	秦学文.基于图像处理的现场钢管计数研究[D].武汉:华中科技大学，2014.
[5]	ZHAO J，XIA X，WANG H，et al.Design of real-time steel bars recognition system based on machine vision[C]//2016 8th International Conference on Intelligent Human-Machine Systems and Cybernetics（IHMSC）.Hangzhou:2016:505-509.
[6]	LIU X H，OUYANG J N，Research on steel bar detection and count ing method based on contours[C]//2018 International Conference on Electronics Technology（ICET）.Chengdu:2018:294-297.
[7]	袁中锦，刘宇，时娟.基于空间相关性的成捆钢筋图像计数算法研究[J].电脑与信息技术，2019，27（4）:10-13.
[8]	LI Y，LU Y J，CHEN J.A deep learning approach for real-time rebar counting on the construction site based on YOLOv3 detector[J].Automation in Construction，2021，124:103602.
[9]	LI Y，CHEN J.Computer vision-based counting model for dense steel pipe on construction sites[J].Journal of Construction Engineering and Management，2022，148（1）:04021178.
[10]	李志鹏.基于目标检测算法的钢筋检测方法研究[D].杭州:杭州电子科技大学，2023.
[11]	高洪飞，杨仁松，张运波，等.基于增强Mask RCNN算法的隧道掌子面节理智能识别方法与应用[J].工业建筑，2023，53（增刊2）:583-587.
[12]	JIANG Y，TAN Z，WANG J，et al.GiraffeDet:a heavy-neck paradigm for object detection[EB/OL].[ 2022-05-10]（2024-07-10）.https://arxiv.org/abs/2202.04256.
[13]	STRUDEL R，GARCIA R，LAPTEV I，et al.Segmenter:Transformer for semantic segmentation[C]//Proceedings of IEEE/CVF International Conference on Computer Vision.Washington D C，2022:7242-7252.
[14]	HU J，SHEN L，ALBANIE S，et al.Squeeze-and-excitation networks[J].IEEE Transactions on Pattern Analysis and Machine Intelligence，2020，42（8）:2011-2023.
[15]	TONG Z J，CHEN Y H，XU Z W，et al.Wise-IoU:bounding box regression loss with dynamic focusing mechanism[EB/OL].[ 2023-04-08]（2024-07-10）.https://arxiv.org/abs/2301.10051.
[16]	QI Y，HE Y，QI X，et al.Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation[C]//2023 IEEE/CVF International Conference on Computer Vision（ICCV）.Paris:2023:6047-6056.
[17]	HOWARD A，SANDLER M，CHEN B，et al.Searching for MobileNetV3[C]//2019 IEEE/CVF International Conference on Computer Vision（ICCV）.Seoul:2019:1314-1324，
[18]	ZHENG Z，WANG P，REN D，et al.Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J].IEEE Transactions on Cybernetics，2021，52（8）:8574-8586.
[19]	范立军.基于轻量级卷积神经网络MobileNetV2的混凝土结构裂缝识别[J].工业建筑，2023，53（7）:231-236.
[20]	谢永康.深度学习开发平台与应用[J].人工智能，2020（3）:43-53.