Abstract: The accurate and timely determination of crop growth stages is fundamental for precision agriculture. To enable rapid, non-destructive, and precise identification of winter wheat growth stages, this study proposed and validated a methodological framework centered on the fusion of multi-source parameters, their optimization, and probabilistic model determination. A comprehensive methodological framework of “multi-source parameter fusion—parameter optimization and dimensionality reduction—probabilistic model determination” was established. Specifically, a Bayesian classification model was constructed by integrating canopy RGB color skew distribution parameters with cumulative light–heat meteorological factors. Field experiments were conducted over four consecutive years at established observation sites in three distinct ecological zones in Shandong Province, China, namely Heze, Tai'an, and Huantai. High-resolution canopy Red-Green-Blue (RGB) images and concurrent meteorological data were systematically collected throughout the growth cycles. Twenty canopy color skew-distribution parameters and four cumulative light–heat indices were extracted. Using manually observed phenological stages as prior knowledge, correlation analysis was performed to identify key variables. Bayesian classification models were then constructed based on canopy color parameters, meteorological factors, and their fusion. Model performance was compared under different input conditions, and cross-year and cross-ecological-zone samples were used to verify generalization ability and robustness. Results showed that both canopy color skew distribution parameters and cumulative light–heat indices were significantly associated with the growth stages of winter wheat, providing a theoretical foundation for intelligent identification. The Bayesian model using only meteorological parameters achieved an overall accuracy of 87.77%, outperforming the model based solely on canopy color information. However, its performance and generalizability across different years and ecological regions required further enhancement. While the initial model integrating all canopy and meteorological parameters showed potential, it also exhibited a risk of overfitting. Subsequent parameter optimization through correlation analysis identified a refined and most effective parameter set for model input: red channel skewness (R_Skewness), green channel kurtosis (G_Kurtosis), accumulated temperature (AT), and accumulated photosynthetically active radiation (AR_E). The Bayesian model constructed with this optimized parameter set demonstrated superior performance, achieving a classification accuracy of 91.70%. More importantly, this optimized model exhibited significantly enhanced robustness and generalizability, delivering high and stable identification accuracies ranging from 84.00% to 100% for independent samples across different years and ecological zones, substantially outperforming the single-source parameter models. This research successfully demonstrates that the fusion of canopy digital image features derived from RGB imagery with key meteorological variables, processed through digital image processing and machine learning techniques, effectively improves the accuracy and robustness of winter wheat growth stage determination. The proposed "multi-source parameter fusion-optimization-probabilistic determination" framework and the specifically developed Bayesian model offer valuable technical support for precision crop management, agricultural meteorological services, and field-level disaster early warning, thereby contributing to the advancement of agricultural intelligence.