Abstract: In the current phased-array ultrasonic inspection of high-density polyethylene (HDPE) pipe thermal butt fusion joint, there are some problems such as low efficiency of pattern interpretation and high personnel experience requirements. Based on the improved YoloX algorithm, an intelligent defect (taking holes as an example) identification method for the TFM phased array ultrasonic maps of thermal butt fusion joint is proposed in this paper. The convolutional block attention module (CBAM) is introduced into the path aggregation network (PAnet) of YoloX to improve the model attention to defect information. The CIoU loss function is used to calculate regression loss to improve the positioning accuracy of the model and reduce the missing detection probability. By TFM phased array ultrasonic testing experiment, the original defect maps are gathered, and the data set is created after image enhancement. The transfer learning strategy is adopted for training to speed up the convergence of the model. The results show that the recognition accuracy of the method for defects reaches 98.18% and the average speed of detection reaches 23.92 frames per second, which improves the detection accuracy by 2.57 percentage points compared with that of the original YoloX model, and has a better detection effect on small target defects. The method proposed in this paper can identify the defects in the TFM phased array ultrasonic spectrum, which can provide technical support for the accurate detection of thermal butt fusion joint.