Pavement crack severity management based on deep learning

초록

This study presents a deep learning-based analysis and management method for Pavement Maintenance System (PMS). Existing systems collect and analyze a variety of data manually in order to assess the condition of the pavement and develop a maintenance plan. However, the manual crack analysis process takes a lot of time, and many human errors may occur due to the difficulty of objective and correct identification of fine or small cracks. In the proposed method, a CNN (Convolution Neural Network) is trained for the classification using pavement analysis information and surface images accumulated in the PMS. A memory efficient architecture SqueezeNet is used as the architecture of the training model. In addition, the effectiveness of training has been increased by using various Keras modules. The trained model classifies the images into three classes depending on the severity of the cracks in the image: no crack, preventive maintenance needed, and maintenance required. The innovative deep learning-based method proposed in this paper enables fast and objective road condition evaluation by automating the existing manual crack analysis process. © 2021 ICIC International. All rights reserved.

키워드