<![CDATA[Efficient systems tailored to countries with bilingual license plates remain scarce, and existing methods often struggle to adapt the unique challenges posed by these scenarios. Countries like Bangladesh and India face challenges with the existing techniques. Accurate recognition of bilingual license plates is crucial for traffic management, but conventional methods with limited resources often fail to work effectively with low-quality footage. Thus, this research provides a framework for the resource-efficient solution for bilingual number plates, ESRGAN for resolution enhancement and explainable CNNs with transfer learning for language classification and character recognition. We divide the task into two sub-tasks and train four specialized CNN models for Bangla and English characters and numerals to maximize the system’s performance. The proposed two-tier architecture optimizes computational efficiency by dynamically loading models based on recognition requirements. Overall, using the gradient-based explainability techniques in the proposed framework increases the credibility of decision-making. The proposed models achieved accuracies of 96%, 97.48%, 97.15%, and 97.62% for English license plate recognition, Bangla license plate recognition, Bangla number recognition, and English number recognition, respectively. While maintaining competitive accuracy, the proposed approach outperforms state-of-the-art works in resource efficiency and computational time, which are crucial for edge deployments. Furthermore, the license plate detection and character extraction algorithm proposed in the paper demonstrates excellent accuracy for both black and white-colored plates. This approach ensures consistent performance and efficient resource utilization while showing the ability to work with a variety of traffic conditions, making it suitable for automated vehicle management systems.]]>
