Xu Zhou, Rui Zhang, Guogen Li, Gang Wei, and Baishuang Liu
TensorFlow, power safety monitoring, OpenPose, abnormal network traffic.
The rapid information technology development has led to electricity demand increasing, which has brought enormous opportunities and challenges to power production enterprises. However, traditional power safety monitoring has certain limitations, which prevent unified supervision, management, and testing analysis of power safety production, resulting in frequent occurrence of various power safety accidents. Therefore, this study proposes an inspection robot power safety monitoring system based on TensorFlow trained attitude recognition model for monitoring power safety issues. In the thousands of datasets, the recognition accuracy of objects can reach over 89%, while the recognition accuracy of human posture is only about 25.8%. When the dataset is increased to ten thousand, the recognition rate for human posture increases to around 48.2%. And after OpenPose preprocessing, this model’s recognising accuracy is improved by about 9% compared to a small number of datasets. The datasets are more; this model’s recognising accuracy is higher after OpenPose preprocessing. This model that has been preprocessed by OpenPose can be monitored in real time and will automatically alert when encountering abnormal situations. It improves the model’s accuracy and makes power safety monitoring timelier.
Important Links:
Go Back
