In the image processing domain applications, which includes image retrieval, autonomous navigation, and content recommendation, scene categorization is crucial. Deep learning procedures have demonstrated the tremendous skill in overcoming the obstacles associated with visual scene categorization. This work gives a thorough investigation of deep learning-based visual scene categorization. Convolutional neural networks (CNNs) are used in the suggested method to extract features and recognize hierarchical patterns. To leverage the discriminative characteristics found in diverse scene categories, we investigate a range of architectures, including cutting-edge pre-trained models. The usefulness of data augmentation approaches like geometric transformations, color, and intensity transformations to improve the model's generalization capacity is also examined in this work, particularly in situations when the availability of labeled training data is constrained.