基于DeepLabv3+的船体结构腐蚀检测方法
On the Corrosion Detection Method of Hull Structures Based on DeepLabv3+
向林浩 1方昊昱 1周健 2张瑜 2李位星2
作者信息
- 1. 中国船级社,北京 100007
- 2. 北京理工大学 自动化学院,北京 100081
- 折叠
摘要
利用图像识别方法对无人机、机器人所采集的实时图像开展船体结构腐蚀检测,可有效提高检验检测效率和数字化、智能化水平,具有极大的应用价值和潜力,将改变传统的船体结构检验检测方式.提出一种基于DeepLabv3 +的船体结构腐蚀检测模型,通过收集图像样本并进行三种腐蚀类别的分割标注,基于DeepLabv3 +语义分割模型进行网络的训练,预测图片中腐蚀的像素点类别和区域,模型在测试集的精准率达到52.92%,证明了使用DeepLabv3 +检测船体腐蚀缺陷的可行性.
Abstract
Using image recognition method to carry out hull structure corrosion detection on real-time video collected by UAV and robot can effectively improve the efficiency and the level of digitization and intelligence of inspection and survey,which has great application value and potential,and will change the traditional hull structure inspection and survey method.A detection model for hull structure corrosion based on DeepLabv3 + was proposed.By collecting the image samples and labeling of three corrosion categories,the DeepLabv3 + model was trained to predict pixel types and areas of corrosion in the image,and the pre-cision of the model could reach 52.92%in the test set,which proved the feasibility of using DeepLabv3 + to detect hull corro-sion defects.
关键词
船体结构/腐蚀检测/深度学习/DeepLabv3+Key words
hull structure/corrosion detection/deep learning/DeepLabv3 +引用本文复制引用
基金项目
国家自然科学基金(61973036)
交通部交通运输行业重点科技项目(2022)(2022-MS2-088)
出版年
2024
NETL
NSTL
万方数据