Traffic Condition Classification Model Based on Traffic-Net.

Area of Science:

• Computer Science
• Transportation Engineering
• Artificial Intelligence

Background:

• Accurate traffic status classification is crucial for urban smart transportation systems.
• Existing methods struggle with real-time accuracy due to factors like weather, lighting, and labeling costs.
• There's a need for efficient methods applicable to small-sample road traffic datasets.

Purpose of the Study:

• To develop an effective method for transferring knowledge from large-scale image datasets to small-sample road traffic datasets.
• To improve the real-time classification and detection accuracy of traffic status.
• To optimize road traffic condition classification using a novel approach.

Main Methods:

• Utilized transfer learning by applying pretrained models from large-scale image datasets to small-sample road traffic data.
• Employed techniques such as sharing common visual features, model weight parameter migration, and fine-tuning.
• Developed and applied a classification model named Traffic-Net.

Main Results:

• Achieved a prediction accuracy exceeding 96% for road traffic condition classification.
• Significantly reduced model training time compared to traditional methods.
• Demonstrated the method's effectiveness and suitability for practical applications.

Conclusions:

• The proposed transfer learning approach effectively addresses the limitations of existing traffic classification methods.
• The Traffic-Net model offers high accuracy and efficiency for real-time traffic status detection.
• This research contributes to the advancement of smart transportation systems by improving traffic management and traveler information.