In this paper, an investigation will be systematically made on the architecture design and core algorithm optimization of an intelligent inspection integration platform in order to meet the safety and smart O&M demands of gas power plants. YOLOv5 detection algorithm is upgraded by means of integrating the attention mechanism of CBAM, multi-scale feature extraction of BiFPN and Focal EIoU loss function for enhancing oil leakage detection performance. A combination of Comsol simulation of temperature field and infrared image processing with SVM classification method is used for steam leakage identification. Experiment results of both oil leakage and steam leakage under simulation conditions are reported to test the feasibility of the intelligent inspection integration platform proposed in this paper. It is found that with the application of the enhanced algorithm, the mean average precision of real oil leakage sample reaches 80.14%, which is 24.13 percentage points higher than that of baseline YOLOv5 detection algorithm. In addition, the steam leakage alarm rate is up to 99.13%.
