An Embedded Deep Learning Method for Glass Insulator Detection

With the continuous advancement of neural networks and deep learning, numerous deep learning models have improved recognition accuracy and efficiency while increasingly demanding higher performance from hardware devices. However, the performance of basic equipment in the industrial field often struggles to keep pace with the rapid development of these models, resulting in the difficulty of implementing high-performing deep learning tools from the research field into industrial applications. This paper proposes an embedded deep learning model based on the YOLOv4 framework, enabling high-accuracy and real-time glass insulator detection in industrial environments on performance-constrained embedded devices. Field experiments demonstrate that the proposed model achieves a detection accuracy of 97%, with an average detection time of 120ms per image, meeting the demands of industrial production.