Facial expression recognition method based on PSA-YOLO network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Face expression recognition is crucial for human-computer interaction.
• Existing methods often face challenges with speed and accuracy.
• The need for efficient and precise facial expression analysis is growing.

Purpose of the Study:

• To develop an improved face expression recognition method.
• To enhance both the speed and accuracy of facial expression identification.
• To introduce the PSA-YOLO model for advanced facial expression analysis.

Main Methods:

• Developed a lightweight backbone network (PSA-CSPDarknet-1) integrating Focus structure and pyramid squeeze attention (PSA) mechanism.
• Incorporated a spatial pyramid convolutional pooling module to improve deep feature map analysis.
• Utilized the α-CIoU loss function for bounding box regression to boost recognition accuracy.

Main Results:

• Achieved a significant reduction in running time from 1,800 ms to 200 ms.
• Demonstrated accuracy improvements of 3.11%, 2.58%, and 3.91% on JAFFE, CK+, and Cohn-Kanade datasets, respectively.
• Validated the model's effectiveness and applicability on standard facial expression datasets.

Conclusions:

• The PSA-YOLO method offers a substantial improvement in face expression recognition speed and accuracy.
• The integrated attention mechanisms and spatial pyramid pooling contribute to enhanced performance.
• The proposed model shows strong applicability and potential for real-world facial expression analysis systems.