Abstract: In response to the complex internal environment of the tunnel and the difficulty in monitoring surrounding rock settlement during construction, an improved YOLOv5 algorithm for automatic identification and measurement of tunnel surrounding rock settlement is proposed. The observation targets are installed at key locations of tunnel construction, and industrial cameras are used to shoot the targets in real-time, and then the target detection model is used to automatically identify the targets. The settlement of surrounding rock is indirectly measured by calculating the settlement of the targets. Based on the YOLOv5s network model, an ECA-Net channel attention mechanism is introduced to enable the network to focus on the target to be detected in channel aspect, in order to avoid the channel dimension reduction and enhance its perception ability for the target. The use of an improved α-IoU target regression loss function has improved the learning ability of the model for image features and target detection accuracy on the whole. The experimental results show that the improved target detection model achieves recognition accuracy and recall rate of 98.5% and 98.4% respectively under the industrial camera shooting target dataset, with high detec? tion accuracy. By analyzing the error, it can be seen that the absolute error is within 6 mm when shooting at a distance of 0~10 meters. When shooting at a distance of 15~25 meters from the target to be tested, the absolute error is within 15 mm, which is relatively small.