Abstract: In order to solve the problems of weak adaptability and low recognition rate of current target detection algorithm for oil injection robots, a new oil injection hole detection algorithm YOLOv5-S was proposed by improving YOLOv5 algorithm. ShuffleNetv2 was used as the feature extraction of the image, the resolution of the input image was adjusted, and the depth convolution kernel size in some basic units was expanded to ensure that the algorithm has both a lightweight network structure and a high-precision detection level. The collected oil injection hole images under different working conditions were classified and labeled, and YOLOv5-S, YOLOv5, YOLOv3 tiny algorithms were used to train them to verify the effectiveness of the proposed algorithm. The results show that the detection accuracy of YOLOv5-S on the oil injection hole dataset remains at 99.4%. Compared with the original algorithm, its model capacity is reduced by 77%, and the detection speed is increased by 11.7 F/s. The detection algorithm proposed in this paper has good recognition accuracy and detection speed under the condition of limited computing power and storage resources of industrial computer.