Missing Dropper Nylon Bush Detection Based on YOLOv4 and AlexNet

Nylon bush is an essential part of dropper in high speed railway contact line equipment, its missing would have high change of causing the carrier cable burnt, which is a high risk to railway security. However, because of small size of nylon bush, complex background and little of negative sample, missing nylon bush detection is difficult. Therefore, this paper uses the method of "detection first and then classification" and presents a proposal to detect missing nylon bushes based on the combination of YOLOv4 and AlexNet. The model detects droppers by YOLOv4 first, then the location coordinate of nylon bush is calculated by the location of dropper. After that, AlexNet model would discriminate whether the nylon bushes are missing in the locations. Besides,some data enhancement methods are applied in this model to solve the problem of sample imbalance. Low detection rate of small target could be avoided in this method,which also meets real-time requirements by using simple classification network to replace the complex operator. The model have relatively high accuracy and low false alarm rate, which can be used in 3C pantograph-OCS intelligent detection system to detect the missing of nylon bushes in real-time.