Abstract: How to detect and control the quality of resistance spot welding joint has become an urgent issue. Ultrasonic detection and measurement is widely used in the quality assessment of resistance spot welding joints because it can directly show the size of weld nugget and the position and magnitude of defects. This paper introduces small wave conversion and gray uniformization techniques to reduce noise and enhance the image to extract the image characteristics of welding nuclei more accurately, while utilizing artificial intelligence technology to establish a predictive model of welding site melt nuclei morphology and mass prediction. The results showed that after the noise reduction enhancement treatment, the measurement of the melting nucleus diameter was more than 90% accurate and consistent with the actual melting nuclei diameter measured at the gold phase. The method significantly reduces the cost of performing disruptive tests and improves the reliability of evaluations.