The detection of container liquid level is an important link in the process of industrial production, storage and transportation of chemical raw materials. Aiming at the problems that the sensor layout in the existing liquid level detection technology is easily limited by space and the short service life of the sensor in special environments such as high temperature, high pressure, dust and humidity, a method of infrared target imaging liquid level detection based on deep learning is proposed in this paper. Through the optimization training of the infrared image annotation data set of the tank liquid level, the model that can accurately identify the percentage content of liquid in the container is obtained. First, construct a standard data set of tank liquid level and build an image detection framework based on Pytorch's deep learning. Then, enhance the data on the image at the input end, adjust the width and depth of the model, and optimize and train the detection model. Finally, the feature pyramid network and path aggregation network structure are used to fuse the feature information of different size feature maps, the complete intersection over union is used to calculate the regression loss of the bounding box, and the weighted non maximum suppression method is introduced in the post-processing process. The experimental results show that the model has good robustness and recognition effect, the mean average precision is up to 0.804 when intersection over union is 0.5.