High-accuracy model recognition method of mobile device based on weighted feature similarity

Area of Science:

• Computer Science
• Electrical Engineering

Background:

• Accurate mobile device recognition is crucial for intellectual property protection and identifying counterfeit products.
• Existing methods achieve high accuracy for device category and brand but require improvement in model-level recognition.

Purpose of the Study:

• To develop a high-accuracy mobile device model recognition method named Recognizer.
• To enhance the precision of identifying specific mobile device models beyond brand and category.

Main Methods:

• Extraction of 20 features from device network traffic and physical attributes.
• Development of feature similarity metric rules and feature importance evaluation strategies to assign weights.
• Calculation of inter-device similarity using weighted features for brand and model recognition.

Main Results:

• The Recognizer method demonstrated superior performance in device brand and model identification compared to existing approaches.
• Achieved an average model recognition accuracy of 99.08% using 20 features and 92.08% using 13 features.
• Experiments were conducted on 587 models from 17 major mobile device brands.

Conclusions:

• Recognizer effectively identifies mobile device brands and models with high accuracy.
• The weighted feature similarity approach offers a significant improvement over current recognition techniques.
• The method is robust, achieving high accuracy even with a reduced feature set.