An effective classifier based on convolutional neural network and regularized extreme learning machine.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Convolutional Neural Networks (CNNs) are effective for feature extraction but can be computationally intensive.
• Regularized Extreme Learning Machines (RELMs) offer fast training and good generalization but may struggle with complex feature learning.
• Combining these approaches aims to leverage their respective strengths for enhanced classification performance.

Purpose of the Study:

• To introduce and evaluate a novel hybrid classifier, CNN-RELM, for improved classification accuracy.
• To assess the performance of CNN-RELM against standalone CNN and RELM models.
• To demonstrate the efficiency and learning capabilities of the proposed CNN-RELM classifier.

Main Methods:

• A hybrid model, CNN-RELM, was developed by integrating a CNN with a RELM.
• The CNN component was trained using gradient descent to achieve a target accuracy.
• The fully connected layer of the CNN was replaced by a RELM optimized via a genetic algorithm.

Main Results:

• The CNN-RELM classifier demonstrated superior performance compared to individual CNN and RELM models.
• Experimental results on various face databases validated the feasibility and effectiveness of CNN-RELM.
• The hybrid model exhibited advantages from both CNN and RELM, leading to easier learning and faster testing.

Conclusions:

• CNN-RELM presents a feasible and effective classification approach by synergistically combining CNN and RELM.
• The proposed method achieves better performance and efficiency than its constituent models.
• This hybrid classifier offers a promising direction for advanced pattern recognition tasks.