Bangladesh’s rapid increase in automobile traffic has underscored the need for advanced solutions in transportation management, law enforcement, and public security, particularly through automated Number Plate Recognition (NPR) systems. This work presents a comprehensive Bangla number plate recognition system customized to the unique challenges of the Bangla script, which is primarily used in Bangladesh’s vehicle registration plates. The proposed system addresses the complexities of the Bangla script using deep learning and image processing techniques. The methodology encompasses data preprocessing, augmentation, and a custom Convolutional Neural Network (CNN) architecture, trained on a dataset of 2,000 images across 17 classes, augmented to 19200 images. The system achieves a test accuracy of 96%, outperforming established models like InceptionV3 (91.28%), MobileNetV2 (93%), VGG16 (94%), and ResNet50 (71%). Extensive evaluation through hyperparameter tuning confirms the model’s robustness and generalization ability. This work contributes to developing localized NPR systems for non-Latin scripts, offering a scalable solution for traffic management, automated tolling, and security applications in Bangladesh, with potential applicability to other regions with similar linguistic challenges.