[1]杨吴奔,苏立鹏,郑建豹,等.基于 AAC-GAN 的起重机缺陷图像生成方法及其在缺陷分类中的应用[J].机械与电子,2025,(09):67-73.
　YANG Wuben,SU Lipeng,ZHENG Jianbao,et al.Crane Defect Image Generation Method Based on AAC-GAN and Its Application in Defect Classification[J].Machinery & Electronics,2025,(09):67-73.
点击复制

基于 AAC-GAN 的起重机缺陷图像生成方法及其在缺陷分类中的应用
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
期数:
2025年09期
页码:
67-73
栏目:
机电一体化
出版日期:
2025-09-25

文章信息/Info

Title:
Crane Defect Image Generation Method Based on AAC-GAN and Its Application in Defect Classification
文章编号:
1001-2257 ( 2025 ) 09-0067-07
作者:
杨吴奔 1 ; 苏立鹏 1 ; 郑建豹 1 ; 娄益凡 1 ; 郑 磊 1 ; 易灿灿 2
1. 温州市特种设备检测科学研究院,浙江 温州 325000 ;2. 武汉科技大学机械工程学院,湖北 武汉 430081
Author(s):
YANG Wuben1 ;  SU Lipeng1 ;  ZHENG Jianbao1 ;  LOU Yifan1 ;  ZHENG Lei1 ;  YI Cancan2
( 1.Wenzhou Institute of Special Equipment Inspection and Research , Wenzhou 325000 , China ;2.School of Mechanical Engineering , Wuhan University of Science and Technology , Wuhan 430081 , China )
关键词:
起重机; 图像生成; 生成对抗网络; 注意力机制; 感知损失
Keywords:
crane ;  image generation ;  GAN ;  attention module ;  perceptual loss
分类号:
TP183 ;TH213
文献标志码:
A
摘要:
针对起重机缺陷图像数量不足导致其模型分类准确率低、安全风险辨识效果差的问题,提出了一种基于注意力和辅助分类器的生成对抗网络( AAC-GAN )用于对起重机缺陷图像进行多样性生成和分类。首先,将生成器设计为一个基于编码器 解码器的网络,该生成器能够生成与缺陷类型相关的编码作为潜在信息从而生成复杂的起重机结构损伤图像。其次,除了使用传统的对抗损失外,还提出感知损失和注意力模块来生成高质量的缺陷图像。最后,使用空间图像特征相似性距离( SIFID )和结构相似性指数( SSIM ) 2 项指标来评估该网络的性能,用来衡量生成图像的多样性和质量。实验结果表明,所提方法的多样性分数和结构相似性分数分别达到了 56.2% 和 59.6% ,相较于现有先进方法表现更好,能够有效地生成高质量的起重机缺陷图像,具有较高的应用价值。
Abstract:
To address the challenges of insufficient crane defect image data leading to low model classification accuracy and poor safety risk identification , this study proposes an Attention and Auxiliary Classifiers Generative Adversarial Network ( AAC-GAN ) for diversified generation and classification of crane defect images.First , the generator is designed as an encoder decoder based network capable of producing defect type related encodings as latent information to generate complex crane structural damage images. Second , in addition to employing traditional adversarial loss , perceptual loss and attention modules are introduced to synthesize high quality defect images.Finally , the network ’ s performance is evaluated using two metrics : Spatial Feature Identity Distance ( SIFID ) and Structural Similarity Index Measure ( SSIM ), which measure the diversity and quality of generated images , respectively.Experimental results demonstrate that the proposed method achieves diversity and structural similarity scores of 56.2% and 59.6% , respectively , outperforming existing state of the art methods.The approach effectively generates high quality crane defect images with significant practical application value.This method addresses the critical need for enhanced defect image datasets in industrial safety inspections while advancing GAN based image synthesis techniques through its novel integration of attention mechanisms and auxiliary classifiers.The quantitative improvements in both diversity and fidelity metrics validate its superiority over conventional approaches for this specialized domain.

参考文献/References:

[ 1 ] 胡晓旭,张鹏,王欣凯,等 . 起重机钢丝绳缺陷检测装置结构优化与仿真分析[ J ] . 起重运输机械, 2025 ( 4 ): 28-35.

[ 2 ] 武亚琪,杨晓英,杨逢海,等 . 基于数据挖掘的复杂机械产品质量关联分析方法[ J ] . 组合机床与自动化加工技术,2025 ( 3 ): 78-82.
[ 3 ] 李洋 . 起重机轨道表面缺陷机器视觉检测技术与设备研究[ D ] . 武汉:武汉工程大学,2024.
[ 4 ] 王华,张燕超,吴波,等 . 基于二阶段网络的起重机小样本图像锈蚀检测[ J ] . 起重运输机械, 2022 ( 19 ): 47-55.
[ 5 ] ISOLA P , ZHU J Y , ZHOU T H , et al.Image-to-image translation with conditional adversarial networks [ C ]// 2017 IEEE Conference on Computer Vision and Pattern Recognition ( CVPR ) .New York : IEEE , 2017 : 1125-1134.
[ 6 ] ZHU J Y , PARKT , ISOLA P , et al.Unpaired image-to-image translation using cycle-consistent adversarial networks [ C ]// 2017 IEEE International Conference on Computer Vision ( ICCV ) .New York : IEEE , 2017 : 2223-2232.
[ 7 ] KARRAS T , LAINE S , AITTALA M , et al.Analyzing and improving the image quality of styleGAN [ C ]// 2020 IEEE Computer Society Conference on Computer Vision and Pattern Recognition ( CVPR ), 2020 : 8107-8116.
[ 8 ] 蔡梓文,赵云,陆煜锌,等 . 基于变分自编码器的多源数据融合窃电检测方法[ J ] . 电力系统保护与控制, 2025 , 53 ( 4 ): 176-187.
[ 9 ] 张明华,刘佳艺,石少华,等 . 基于全局特征提取和提示学习的水下图像增强[ J ] . 华中科技大学学报(自然科学版),2025 , 53 ( 3 ): 31-40.
[ 10 ] 戚鑫,魏维,陈柯宇,等 . 基于复杂注意力与神经符号感知网络的文本图像恢复[ J ] . 计算机应用与软件, 2025 , 4 ( 14 ): 1-7.
[ 11 ] 王亮,周朦,邱硕 . 浅谈基于卷积神经网络的 AR 起重机识别系统[ J ] . 中国设备工程, 2024 ( 8 ): 141-144.
[ 12 ] 刘晶晶,刘业峰 . 一种新型激活函数的机床能耗预测神经网络研究[ J ] . 控制工程, 2025 , 32 ( 3 ): 492-499.
[ 13 ] 王俊霞,张岩波,余红梅,等 . 基于高斯混合模型双向聚类重采样和随机森林构建 DLBCL 早期复发预测模型[ J ] . 中国卫生统计, 2025 , 42 ( 1 ): 7-11 , 17.
[ 14 ] 张宇,李宝山,高迪,等 . 基于改进的 InceptionV3 模型在肉牛体侧识别中的应用研究[ J ] . 内蒙古科技大学学报,2024 , 43 ( 4 ): 365-370.
[ 15 ] 宫永立,王玉超,刘志文,等 . 基于四分位数和 Sigmoid 改进模型的风电数据清洗方法[ J ] . 电力科学与工程,2025 , 41 ( 3 ): 55-62.
[ 16 ] BAKUROV I , BUZZELLI M , SCHETTINI R , et al. Structural similarity index ( SSIM ) revisited : a data driven approach [ J ] .Expert systems with applications , 2022 , 189 : 116087.
[ 17 ] YAN L , ZHENG W B , GOU C , et al.IPGAN : identity preservation generative adversarial network for unsupervised photo-to-caricature translation [ J ] .Knowledge-based systems , 2022 , 241 : 108223.
[ 18 ] 胡梦婷,罗晨 . 基于 MCNN-LSTM 和交叉熵损失函数的轴承故障诊断[ J ] . 制造技术与机床, 2024 ( 9 ): 16-22.

相似文献/References:

[1]徐小东,韩颖.某型号起重机仿真分析与试验研究[J].机械与电子,2016,(06):33.
　XU Xiaodong,HAN Ying.Simulation Analysis and Experimental Research on Cranes[J].Machinery & Electronics,2016,(09):33.
[2]李文俊,胡 泓,李峥嵘.一种用于引线键合视觉定位的图像质量评估方法[J].机械与电子,2021,(03):60.
　LI Wenjun,HU Hong,LI Zhengrong.An Image Quality Assessment Method for Visual Positioning of Wire Bonder[J].Machinery & Electronics,2021,(09):60.
[3]巩 盛.双梁抓斗起重机负载运动自动化防摇控制方法[J].机械与电子,2022,(12):63.
　GONG Sheng.Automatic Anti-sway Control Method for Load Movement of Double-beam Grab Crane[J].Machinery & Electronics,2022,(09):63.

备注/Memo

备注/Memo:
收稿日期: 2025-05-06
基金项目:国家自然科学基金资助项目( U1709210 );温州市基础性科研项目(2023G0277 )
作者简介:杨吴奔 ( 1982- ),男,浙江温州人,高级工程师,研究方向为特种设备健康监测;娄益凡 ( 1985- ),男,浙江温州人,硕士,工程师,研究方向为特种设备无损检测,通信作者, E-mail : 15549872671@163.com 。
更新日期/Last Update: 2025-09-29