基于深度学习的水下钢结构锈蚀识别与评价

Corrosion Recognition and Evaluation of Underwater Steel Structures Based on Deep Learning

锈蚀是影响输水工程中水下钢结构能否稳定运行的重要因素.由于水下环境复杂,常规方法难以在通水情况下实现对水下钢结构锈蚀情况进行有效的检测与评价.本文基于深度学习,提出了一种利用水下机器人采集的视频图像,自动检测水下钢结构锈蚀并进行锈蚀等级评价的方法.首先,采用构建的卷积神经网络(CNN)模型对预处理后的水下钢结构锈蚀图像进行锈蚀识别;然后,利用颜色直方图对锈蚀区域进行语义分割,通过计算的锈蚀率来评价锈蚀等级.实例结果表明:水下钢结构锈蚀识别的准确率可达95.5%,实现了锈蚀程度评价.本文方法可为通水情况下输水工程水下钢结构的锈蚀快速识别及健康诊断提供新的途径.

Corrosion is an important factor affecting the safe operation of underwater steel structures in water conveyance projects.The complex underwater environment renders it difficult for conventional methods to effectively detect and evaluate the corrosion of underwater steel structures under the condition of water flow.Based on deep learning,this research proposes a method for automatic detection involving corrosion recognition and corrosion grade evaluation of underwater steel structures.First,the constructed convolutional neural network(CNN)model is used to identify corrosion in preprocessed images of underwater steel structures.Next,the corrosion area is semantically segmented using color histograms,after which the corrosion grade is evaluated by the calculated corrosion rate.The experimental results show that the accuracy of corrosion recognition of underwater steel structure can reach 95.5%,and the evaluation of the corrosion degree is realized.This study demonstrates a novel method for the rapid identification of corrosion and the health diagnosis of underwater steel structures used in water conveyance projects under the condition of water flow.