Abstract: Improving the detection accuracy of leakage defects in tunnel structures is crucial for ensuring the safe operation of tunnels. A cooled infrared thermography was employed to conduct rapid, non-contact inspections of actual tunnels, and temperature data samples were systematically collected. Differing from traditional image processing techniques, a quantitative leakage screening and segmentation algorithm based on second-order gradient features of temperature field data was proposed. First, characteristic values of the raw temperature field data including mean and minimum values were extracted. Second, second-order gradient features were obtained using eight-direction Laplacian operator convolution. Third, leakage samples were screened based on threshold values of second-order gradient features. Finally, the boundary temperature of leakage was calculated to segment the leakage regions. A tunnel structure intelligent inspection vehicle integrating infrared thermal imaging, line-scan CCD cameras, and high-precision positioning systems was developed, along with dedicated data acquisition and processing software. Field testing on a shield tunnel section of Shanghai Metro demonstrated that the second-order temperature gradient feature range for leakage regions was 0.27~0.37°C·mm^-2, and the algorithm’s 10 leakage locations matched perfectly with manual on-site inspection results. The tunnel leakage identification method based on second-order temperature gradient features effectively improves detection accuracy and provides technical support for tunnel operation and maintenance inspections.