Advanced vehicle monitoring in smart port utilizing deep denoising real-time object detectors integrated multi-resolution attention-augmented CRNN
Anh Son Ta, Luan Thanh Le, Linh Bui-Duy
Abstract
The development of smart ports can solve supply chain disruptions caused by uncertainties and alleviate congestion at seaports. Intelligent technology helps detect license plates (LPs) of vehicles and container identification numbers (IDs). A system using a novel hybrid Multi-Resolution Attention Augmented Convolutional Recurrent Neural Network (MR-AA-CRNN) integrated with the deep denoising (DD) for a real-time object detection technique (YOLOv12) is employed to detect and recognize LPs and IDs. The input surveillance video is processed using a DD model to remove noise and artifacts. YOLOv12 is employed to detect and localize vehicle LPs and container IDs in the images. The MR approach is deployed to enhance the low-quality image to a high-quality one after extracting the LP region. The processed images are then passed through an AA-CRNN, which enhances the model’s focus on important image regions, extracts features for character sequence prediction, and decodes the sequence to produce the final output. The proposed hybrid model achieves recognition accuracy of up to 99.71 % for LPs ([email protected] = 99.50 %) and 99.57 % for container IDs ([email protected] = 99.14 %) under ideal conditions. The originality of this study is in creating a smart system for seaport gates. The system significantly reduces the time required to monitor vehicle entry and exit at the port compared to traditional methods.