Varying lighting conditions pose significant challenges to image classification models, particularly Convolutional Neural Networks (CNNs). This research aims to introduce a GAN-based preprocessing method to generate illumination-invariant images, expecting to enhance the robustness of CNNs in classifying images captured under diverse lighting conditions. By exploring architectures such as Conditional GAN, GauGAN, and techniques like multi-scale feature extraction and illumination attention, this study seeks to minimize lighting-related discrepancies in image datasets. The proposed method is expected to preprocess images, normalizing lighting variations and improving classification consistency. Experiments will be conducted on diverse datasets with varying lighting conditions to evaluate the effectiveness of the GAN-enhanced preprocessing. The anticipated outcome is a significant improvement in classification accuracy, demonstrating the potential of illumination-invariant preprocessing to address challenges posed by lighting inconsistencies in image recognition systems.