Winter injury frequently causes lethal damage to grape buds and reduces productivity, creating major challenges for grapevine production in regions with cold or erratic climates. Conventional dissection based assessment is destructive, labor intensive, and unsuitable for rapid or scalable monitoring. A non destructive and scalable method for evaluating bud viability is therefore needed to improve winter injury management and vineyard decisions. This thesis developed and evaluated an active thermography system combined with machine learning for non destructive grape bud mortality detection. A pulsed thermography setup was assembled and synchronized to capture thermal videos from four cultivars. A processing pipeline extracted thermal response features such as heating and cooling slopes, peak temperature, and principal components. Mixed effects modeling confirmed significant within cultivar differences between viable and non viable buds, with non viable buds showing faster thermal transients and higher peaks due to changes in water content and tissue integrity. Thermal videos were represented as one dimensional features, two dimensional time series, and three dimensional image sequences, and classified using logistic regression, random forest, support vector machines, long short term memory networks, hybrid long short term memory variants, and a video vision transformer. Models were trained under universal and cultivar specific settings with class-aware cross validation. Universal models exceeded seventy percent accuracy, and cultivar specific models approached eighty percent. Non parametric ranking tests showed small performance differences among the models with corresponding feature representation methods, indicating that simple feature based models can perform comparably to deep learning approaches. The video vision transformer underperformed overall but performed relatively well on Concord. These findings suggest that optimal deployment should be both cultivar specific and application specific. Overall, this work demonstrates that active thermography with machine learning can provide a non destructive and rapid approach for grape bud viability assessment. It also reveals new patterns in thermal dynamics that support future investigation of underlying physiological mechanisms. This study offers the first comprehensive benchmark of thermography based grape bud viability classification across multiple cultivars and model families, outlining both the potential and the remaining challenges for vineyard scale implementation.