Lighting condition variations present a very significant challenge in maintaining the accuracy and reliability of computer vision models in real-world applications. This paper introduces a novel transfer learning approach to adapt models for diverse lighting conditions, particularly focusing on training under poor lighting conditions and testing against ideal lighting scenarios. The proposed framework initially trains deep learning models, specifically in Convolutional Neural Networks (CNNs), the images are captured in poor lighting conditions, enabling the model to learn robust invariant features. These features are subsequently transferred and evaluated in well light conditions to ensure consistent performance across diverse environments. Experimental results demonstrate that this approach substantially improves model accuracy and stability, achieving a noticeable improvement over traditional methods when dealing with variable lighting conditions. The research provides a foundation for applications that require reliable vision systems such as autonomous driving, surveillance, and medical imaging where lighting consistency cannot always be ensured.