Precise control and accurate endpoint determination in the thawing of frozen beef are vital for maintaining its quality and safety in the food industry. Traditional thawing methods, which are time-controlled, often lead to inconsistencies like under or over-thawing, adversely affecting texture, color, nutritional value, and increasing the risk of microbial contamination. This study introduces a novel, non-destructive approach using ultrasonic signals and ultrasonic velocity for the quantification of beef thawing. It involves analyzing thermal images of beef cross-sections to assess thawing levels and gathering ultrasonic data at various thawing stages. Machine learning algorithms, including KNN, ANN, Extra-Trees, and LightGBM, were employed to develop prediction models that accurately identify the thawing endpoint. The models exhibit high accuracy, with R^2 values exceeding 0.9, some reaching as high as 0.986. This method represents a significant advancement in non-destructive quality control, enhancing safety and quality management in the beef processing sector.