Application of an ensemble CatBoost model over complex dataset for vehicle classification.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Traditional vehicle classification methods face challenges like inefficiency and errors.
• Machine learning offers a promising solution for accurate vehicle image analysis.

Purpose of the Study:

• To develop an effective machine learning model for vehicle classification (VC).
• To improve accuracy and efficiency in categorizing vehicles from large datasets, even in challenging conditions.

Main Methods:

• Utilized contrast enhancement for image pre-processing.
• Employed Mask-R-CNN for feature segmentation and VGG16 for feature extraction.
• Implemented an autoencoder for feature selection and the CatBoost (CB) algorithm for final classification.

Main Results:

• The CatBoost algorithm demonstrated superior performance in vehicle classification.
• Achieved a high accuracy rate of 98.89% on the UFPR-ALPR dataset.
• The model proved effective in diverse environments including adverse weather and partial blockage.

Conclusions:

• The proposed machine learning approach significantly enhances vehicle classification accuracy.
• The integrated method effectively addresses limitations of traditional techniques for large-scale vehicle image analysis.