Abstract: At present, the traditional manual installation and visual inspection of rubber sealing ring for automobile pipe fittings have low efficiency and high false detection rate. Aimed at this problem, an improved Faster RCNN method for detecting installation of rubber sealing rings for automotive pipes is put forward. This improved method replaced the backbone feature extraction network with ResNet50. The CSPNet structure was added to the backbone network, and the ordinary convolution in the original algorithm was replaced with deep separable convolution, achieving lighter network structure and reducing the amount of model parameters and calculation costs. The channel shuffling unit and the use of Mish activation function was introduced to further improve the accuracy of the network. Experimental results show that based on 5 500 image dataset, the accuracy rate of improved Faster RCNN network model is 91.45%, which meets real production needs.