An infrared image detection method for electrical equipment based on improved YOLOv9

Traditional electrical appliance thermal fault detection methods have issues such as insufficient intelligence to meet practical application requirements, excessive human subjectivity in fault detection, and low fault detection efficiency. Timely detection of problematic equipment can increase the service life of the machine, thereby reducing maintenance costs. Addressing the aforementioned issues, this paper proposes a YOLOv9-SwiftLite model. This method is based on YOLOv9c, with modifications including the replacement of the Backbone network with MobileNetv4 and the introduction of attention mechanisms such as SENet, Polarized Self-Attention, and Triple Attention in the head to enhance model accuracy. Additionally, an LSKA module is embedded in the final three parts of RepNCSPLEAN4 to ensure that its precision meets the requirements for real-time detection in complex environments. The experimental results demonstrate that YOLOv9-SwiftLite achieves an improvement in mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50</inf> from 0.835 of YOLOv9c to 0.882 of YOLOv9-SwiftLite, with a 2% increase in mAP<inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">50-95</inf>. Under the same conditions, when compared to YOLOv5, YOLOv7, YOLOv8, and YOLOv9, the method proposed in this paper addresses the issues of low detection efficiency and inaccuracy in machine detection, achieving rapid and accurate results in real-time detection.