Intelligent Recognition of Leakage Infrared Thermal Imaging of Embankment Dams Based on ResNet Networks

With the development of modern water conservancy, the leakage potential of embankment dams has received more and more attention. However, the current mainstream detection method is still manual inspection, which is often accompanied by high cost, time-consuming, damaging, and dangerous defects. How to effectively and accurately detect the leakage potential is of great significance for the prevention and control of floods and the overall safety monitoring of dykes. In response to the above urgent needs, this paper builds an intelligent identification system of infrared thermal imaging of levee leakage based on the ResNet network. The system takes the actual earth and stone embankment dam as the structural prototype and builds an indoor simulation model of the embankment dam. The infrared thermography spectral data of the embankment dam are collected using an infrared thermography camera. A ResNet network is used to classify and train the data-enhanced infrared image data to achieve the detection of leakage potentials of the embankment dam. The maximum accuracy of the ResNet network under two sample divisions can reach 99.9%, and the anomalous fluctuation in the test is small. Further, based on the accuracy optimal classification model, the leakage prediction is carried out on the real-time captured images of the acquired infrared thermal imager, and the prediction results are displayed in real-time. The results show that the system can achieve correct, efficient, and real-time detection of dam leakage. The proposed infrared thermal imaging intelligent identification system for dam leakage improves the identification efficiency of dam leakage and ensures classification accuracy, which provides a reference for scientific and safe dike defense.