Page 237 - 《应用声学》2025年第2期

第 44 卷 第 2 期 Vol. 44, No. 2
2025 年 3 月 Journal of Applied Acoustics March, 2025

⋄ 研究论文 ⋄

改进Faster R-CNN的大型铸造不锈钢机匣超声

相控阵检测图像的缺陷智能识别 ∗

赵玉琦 1 李 婧 2 董德秀 2 黄鑫章 1 陈振华 1† 卢 超 1

(1 南昌航空大学 无损检测技术教育部重点实验室 南昌 330063)
(2 中国航发沈阳黎明航空发动机有限责任公司 沈阳 110043)

摘要：大型铸造不锈钢机匣的超声相控阵检测技术具有检测能力强、检测效率高的优势。然而，相控阵图像中
显示的缺陷类型仍需检测人员判读，存在主观性强、易误判、效率低、可靠性不足等问题。据此，提出基于深度
学习的机匣超声相控阵检测图像缺陷类型的自动识别方法。首先，采集机匣典型铸造缺陷的超声相控阵图像，
对缺陷图像扩充并制备数据集；其次，在 Faster R-CNN 深度学习网络的特征提取网络、多层特征信息融合网
络、感兴趣区域模块等方面进行优化改进；最后，对比分析改进前后深度学习网络模型的缺陷识别与分类准确
率。结果表明：相比于原始 Faster R-CNN 深度学习网络，在采用深度残差网络、特征金字塔网络、区域一致性
池化等优化措施后，平均准确率均值提高至 95.3%，模型对缺陷图像的识别精度得到了有效的提高；改进的
Faster R-CNN 目标识别算法克服了超声相控阵缺陷图像人工识别与分类的问题，具有较好的工程应用价值。
关键词：超声相控阵检测；改进 Faster R-CNN；缺陷智能识别
中图法分类号: TG115.28 文献标识码: A 文章编号: 1000-310X(2025)02-0497-08
DOI: 10.11684/j.issn.1000-310X.2025.02.025

Improved Faster R-CNN for the intelligent recognition of defects in ultrasonic
phased array inspection images of large cast stainless steel casing

1
1
ZHAO Yuqi , LI Jing , DONG Dexiu , HUANG Xinzhang , CHEN Zhenhua and LU Chao 1
1
2
2
(1 Key Laboratory of Nondestructive Testing of Ministry of Education, Nanchang Hangkong University,
Nanchang 330063, China)
(2 AECC Shenyang Liming Aero-Engine Co., Ltd., Shenyang 110043, China)
Abstract: The ultrasonic phased array testing technology for large stainless steel machine boxes has the
advantages of strong detection capability and high detection efficiency. However, the defect types displayed
in the phased array images still need to be judged by the testing personnel, leading to problems such as
strong subjectivity, easy misjudgment, low efficiency, and lack of reliability. In response to this, an automatic
identification method for detecting image defect types of machine box ultrasonic phased array based on deep
learning is proposed. Firstly, ultrasonic phased array images of typical casting defects of the machine box
are collected, and the defect images are augmented to prepare the dataset. Secondly, optimization and im-
provement are carried out in the feature extraction network, multi-layer feature fusion network, and region
of interest module of the Faster regions with convolutional neural network (R-CNN) deep learning network.
Finally, the defect recognition and classification accuracy of the deep learning network model before and after
the optimization are compared and analyzed. The results show that, compared with the conventional Faster
2023-10-23 收稿; 2023-12-27 定稿
国家科技重大专项 (J2019-VII-0002-0142), 国家市场监督管理总局科技计划项目 (2021Mk065)
∗
作者简介: 赵玉琦 (1997– ), 男, 辽宁锦州人, 硕士研究生, 研究方向: 声学检测, 智能检测技术。
† 通信作者 E-mail: zhenhuachen@yeah.net