<p>Injection molding is a manufacturing process characterized by complex and variable operating conditions, where product quality is often influenced by multiple factors. In practical production, there are significant differences in the quality distribution of products under different operating conditions, leading to the common issue of poor generalization in traditional quality prediction models, making it difficult to accurately predict product quality under new operating conditions. Therefore, this paper proposes a multi-operating conditions injection molding product quality prediction method based on multi-feature fusion and fine-tuned transfer learning. Initially, a multi-feature fusion model based on SAMSBG-CNN is proposed as the base model for fine-tuned transfer learning. The model modularizes the modeling of temporal features, local spatial features, and state features to extract common characteristics of injection molded products under mixed operating conditions. The model’s predictive accuracy is improved by achieving collaborative modeling of multi-source information through multi-feature fusion. In addition, to address the issue of insufficient model generalization in new operating conditions and small sample scenarios, we introduce a fine-tuning transfer learning strategy. By leveraging transfer learning, we enable the cross-operating-conditions transfer and reuse of common features in injection molding products. Furthermore, the model is fine-tuned with a small amount of new operating conditions data, enhancing its generalization and adaptability in novel operational environments. The results demonstrate that the proposed method achieves superior performance under both mixed operating conditions and small sample new operating conditions. Additionally, it validates that the fine-tuning strategy for local spatial features significantly impacts the model’s generalization capability.</p>