Optimization of deep learning-based faster R-CNN network for vehicle detection

Document Type

Article

Publication Title

Scientific Reports

Abstract

Optimizing hyperparameters in object detection models is critical for enhancing performance, particularly in domain-specific tasks such as vehicle detection. This research systematically investigates the optimization of key hyperparameters for the Faster R-CNN model to maximize its efficiency in detecting vehicles. We evaluated the impact of various base CNN architectures (VGG-16, ResNet-50, Inceptionv3), solvers (sgdm, rmsprop, adam), learning rates (10− 5, 10− 4, 10− 3), and detection thresholds (0.1, 0.2, 0.3) on model performance. Our findings reveal that the optimal performance, achieving an average precision-recall (PR avg) value of 82%, was obtained using ResNet-50 with a learning rate of 10− 5 and a detection threshold of 0.1, employing the rmsprop solver across all learning rates and detection thresholds studied. The results demonstrate a clear trend wherein decreasing the learning rate from 10− 3 to 10− 5 steadily enhances network efficiency. Additionally, the choice of solver and detection threshold significantly influences the model’s performance. These insights emphasize the importance of meticulous hyperparameter tuning to improve the accuracy and reliability of object detection models. The proposed optimization methodology can be applied to various object detection tasks beyond vehicle detection, offering a framework for systematically enhancing model performance in diverse applications such as surveillance, autonomous driving, and traffic management systems.

DOI

10.1038/s41598-025-22828-z

Publication Date

12-1-2025

This document is currently not available here.

Share

COinS