Traditional cross-correlation algorithms are prone to time-of-flight (TOF) calculation errors under conditions of strong noise interference and complex temperature gradients, resulting in a decline in the accuracy of ultrasonic temperature measurement. To this end, this paper proposes an ultrasonic temperature measurement method that combines YOLOv11 target detection with energy-type weighted cross-correlation algorithm. The YOLOv11 model is utilized to conduct target detection and key area positioning on the ultrasonic signal waveform diagram, automatically identifying characteristic waveforms such as node waves and end face waves, and achieving adaptive extraction of the effective signal interval. Further introduce the energy-based weighted cross-correlation algorithm. Based on the signal energy distribution, the cross-correlation results are weighted and processed to enhance the main wave response and suppress noise interference. Experiments show that the YOLOv11 model has high detection accuracy (Precision = 0.987, Recall = 0.958, mAP@50 = 0.988); The proposed method maintains the stability of time delay estimation under strong noise and high temperature (> 1200℃), with the average time delay error reduced by approximately 35% to 50% compared to traditional algorithms. This verifies its high robustness and temperature measurement accuracy in complex environments, and it has a promising engineering application prospect.