An Early Warning Method for Abnormal Fluttering Behavior in Transmission Lines Based on Deep Binocular Vision

Under adverse weather conditions, ice accumulation and strong winds cause low-frequency and high-amplitude swaying of transmission lines, increasing the risk of wire fatigue, hardware damage, and tower collapse, posing a threat to the safety of the power grid. The existing monitoring methods suffer from inaccurate 3D spatial data acquisition, leading to errors in edge detection, difficulty in suppressing background interference, insufficient feature extraction, and reduced accuracy in detecting abnormal oscillations. Therefore, this article proposes a warning method based on deep binocular vision, which captures three-dimensional data with a stereo camera and converts it into two-dimensional images. Mathematical morphological operators are used to determine the edge of the transmission line, solving the problem of inaccurate edge detection. The morphological edge results are input into the ResNet LSTM network to suppress background noise and extract features. Finally, compare the detection results with the actual values to achieve multi-level alarms. The experiment shows that this method is efficient and accurate, with a detection efficiency of 0.958-0.968. Improved the reliability of early warning.