Intelligent Extraction of Surface Cracks on LNG Outer Tanks Based on Close-Range Image Point Clouds and Infrared Imagery.

Highlights: Most of the studies on oil tanks have focused on the analysis of deformation and settlement, and more research needs to be done on crack extraction from external LNG tanks. Oil tanks are more sensitive to temperature due to the lower temperature inside the tank. Using infrared images as a dataset for crack recognition can identify cracks that the naked eye cannot see, and a convolutional neural network that introduces a channel attention mechanism is used for crack recognition with a recognition accuracy of 85.9%. The automatic extraction of three-dimensional (3D) crack point clouds using depth images is novel and accurate, with an accuracy of about 97.6%. The precise detection and ongoing surveillance of surface fractures on exterior LNG storage tanks are crucial in guaranteeing the secure transit and storage of natural gas. Undetected fractures have the potential to result in the release of liquefied natural gas (LNG), hence presenting a significant risk to both public health and the environment. This paper presents a novel approach for crack identification, which involves the integration of thermal infrared images and point clouds derived from close-range images captured by unmanned aerial vehicles (UAVs). The aim of this approach is to overcome the limitations of conventional manual detection methods, namely in terms of efficiency and safety concerns. The primary approach for acquiring two-dimensional photographs of the tank surface is the utilization of infrared technology to generate an infrared dataset capturing the presence of fractures on the storage tanks' exterior. The utilization of the attention mechanism convolutional neural network is employed during the process of model training. The UAV close-range photos were utilized in close-range photogrammetry to generate an accurate point cloud model. This was achieved by incorporating control point coordinates and matching feature points. The infrared photos that were discovered were subsequently matched with this particular model. The 3D model that was officially was employed as a point of reference on the unfolded two-dimensional (2D) plane. To construct the depth image, a least-squares approach of least-column fitting was utilized. In order to validate the accuracy of the automated extraction process, a manual crack extraction was conducted on the original close-range image point cloud of the tank exterior. The results indicated that the extracted cracks exhibited an accuracy level of around 97.6%. The experimental findings demonstrate that the process of crack extraction exhibits a high level of accuracy, hence presenting numerous possible applications in the realms of maintenance management and intelligent monitoring. The utilization of this technology is appropriate for the purpose of monitoring the thermal conditions and structural soundness of LNG storage tanks. [ABSTRACT FROM AUTHOR]