An Ensemble Broad Learning Scheme for Semisupervised Vehicle Type Classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Transportation Engineering

Background:

• Vehicle type classification is crucial for intelligent transportation systems (ITSs) but often relies on supervised learning, limiting its use with abundant unlabeled data.
• Existing methods face challenges with time-consuming training and the burden of unlabeled samples in real-world ITS applications.

Purpose of the Study:

• To propose a novel semisupervised vehicle type classification scheme for ITS.
• To enhance the applicability and efficiency of vehicle classification by overcoming limitations of supervised learning.

Main Methods:

• A semisupervised learning approach is employed to train multiple base broad learning system (BLS) classifiers, reducing training time and addressing unlabeled data issues.
• A dynamic ensemble structure combines these classifiers, leveraging diverse characteristics to determine vehicle type probability and improve generalization.

Main Results:

• The proposed semisupervised ensemble broad learning method demonstrates superior performance compared to single BLS classifiers.
• Experiments on the BIT-Vehicle and MIO-TCD datasets validate the effectiveness and efficiency of the developed classification scheme.

Conclusions:

• The semisupervised ensemble broad learning approach offers an effective and efficient solution for vehicle type classification in ITS.
• This method significantly improves generalization performance and broadens the applicability of vehicle classification in intelligent transportation systems.