基于卷积神经网络的隧道掌子面图像质量评价方法研究

Research on image quality assessment method of tunnel face based on convolutional neural network

鉴于隧道掌子面图像的多样性和复杂性,提出一种基于深度卷积神经网络(Convolutional Neural Network,CNN)的图像质量评价方法,以筛选出满足工程需求的掌子面图像。基于多条隧道创建掌子面图像数据集,采用Keras深度学习框架,应用多种主流的CNN进行对比试验,并结合传统的图像评价指标,分别从清晰度、分类和相似度3个方面对掌子面图像质量进行评价。其中基于DenseNet169的多分类模型可达到88.7%的准确率。研究结果表明:跟传统的图像处理技术相比,深度学习方法在隧道掌子面图像识别上具有精度高、效率高的显著优势。该方法可为实现隧道掌子面的自动素描提供技术支持,具有良好的工程应用前景。

In view of the diversity and complexity of tunnel face images,an image quality assessment method based on deep convolution neural network was proposed to select tunnel face images to meet engineering needs.A tunnel face image dataset based on several tunnels was created.The Keras deep learning framework was adopted.Various mainstream convolutional neural networks(CNN)were applied to carry out comparative experiments.Combining with traditional image evaluation indexes,the tunnel face image quality was evaluated from three aspects:clarity,classification and similarity.The multi-classification model based on DenseNet169 achieved 88.7%accuracy.The results show that,compared with the traditional image processing technologies,the deep learning method has the remarkable advantages of high accuracy and high efficiency in tunnel face image recognition.This method can provide technical support for realizing automatic sketch of tunnel face,and it has a good prospect in engineering application.