A Classification Method for Electronic Components Based on Siamese Network.

Area of Science:

• Electronics manufacturing
• Computer vision
• Machine learning

Background:

• Electronic component classification is crucial for managing and recycling electronic waste.
• Current deep learning methods require extensive datasets, which are costly and time-consuming to collect due to the wide variety of components.
• A need exists for efficient classification methods that perform well with limited data.

Purpose of the Study:

• To propose a few-shot learning method for electronic component classification using a Siamese network.
• To improve the feature extraction capabilities for small sample recognition.
• To enhance model generalization and classification accuracy in electronic waste management.

Main Methods:

• Utilized an improved Visual Geometry Group 16 (VGG-16) model as the feature extractor within a Siamese network architecture.
• Developed a novel channel correlation loss function to capture inter-channel relationships in feature maps, improving generalization.
• Employed the nearest neighbor algorithm for the final classification of electronic components.

Main Results:

• The proposed Siamese network achieved high classification accuracy even with limited training samples.
• The method demonstrated robustness in distinguishing between electronic components with visually similar appearances.
• Significant reduction in the cost and effort associated with collecting training data was observed.

Conclusions:

• The Siamese network-based approach effectively addresses the challenge of electronic component classification with few samples.
• The integration of an improved VGG-16 and channel correlation loss enhances model performance and generalization.
• This method offers a practical solution for improving electronic waste management and recycling efficiency.