Research on efficient feature extraction: Improving YOLOv5 backbone for facial expression detection in live streaming...

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marketing Analytics

Background:

• Facial expressions transmit non-verbal cues that influence human behavior and decision-making.
• In live streaming marketing, anchor facial expressions significantly impact consumer engagement and purchasing intent.
• Accurate detection of these expressions is crucial for understanding audience response and optimizing marketing strategies.

Purpose of the Study:

• To develop an efficient feature extraction network for detecting marketing anchors' facial expressions in live streaming videos.
• To enhance the YOLOv5 model for improved accuracy and reduced latency in facial expression recognition.
• To create a robust system for analyzing the impact of facial cues on consumer behavior.

Main Methods:

• A two-step cascade classifier and recycler was employed to filter video frames and generate a specialized facial expression dataset.
• GhostNet and coordinate attention mechanisms were integrated into the YOLOv5 architecture.
• The modified YOLOv5 model was trained and evaluated on a self-built dataset of anchor facial expressions.

Main Results:

• The proposed efficient feature extraction network demonstrated superior performance compared to the original YOLOv5.
• The modified model achieved significant improvements in both detection speed and accuracy.
• The integration of GhostNet and coordinate attention effectively addressed latency and enhanced recognition capabilities.

Conclusions:

• The developed YOLOv5-based network offers an efficient and accurate solution for detecting facial expressions in live streaming marketing.
• This technology can provide valuable insights into consumer responses, aiding marketers in optimizing their strategies.
• Further research can explore real-time emotion analysis and its application in personalized marketing interventions.