Application of infrared thermal images for sinkhole detection with time-series and time-difference index data through a convolution neural network

ABSTRACT

Sinkholes in the subgrade disrupt the transportation capacity and usability of roads. The objective of this study is to use infrared thermal images to find sinkholes in the subgrade, and a CNN is applied. Cavities of small and large sizes are artificially simulated to conduct field testing, and cavities are installed in the top, middle and bottom positions. The temperature variations are measured via an infrared camera of 240 × 180 pixels to account for heating and cooling conditions. A convolution neural network (CNN) is applied to classify the measured image according to the sinkhole characteristics, including size, depth and surface materials. The highest accuracy is 90.9% for the locally distributed sinkhole with a surface material of sand. To reflect the characteristics of temperature change, the temperature difference index (TDI) is calculated and the numerical data are converted into an image through the recurrence plot (RP) algorithm. The results of the CNN are raised to 17% through TDI. This study shows that the suggested technique is an alternative method for detecting sinkholes in roads via infrared thermal images.

KEYWORDS:

• Convolution neural network (CNN)
• field test
• infrared thermal image
• recurrence plot (RP)
• sinkhole

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Education (NRF-2020R1A2C2012113).