Performance Comparison of Different Methods for Miner Emotion Recognition

A dataset constructed for a study proposing a deep learning-based bimodal framework to recognize unsafe emotions exhibited by miners during operations. The dataset contains five categories of miners' unsafe emotions and was used to train a model achieving a mean recognition accuracy of 85.56%. The dataset was last updated on May 15, 2026.

Use Cases

• Benchmarking emotion recognition models based on the described five-category classification task.
• Comparing unimodal vs. bimodal (speech and facial expression) feature fusion strategies.
• Evaluating the generalization ability of models using the three described feature enhancement strategies.
• Testing AI model performance in real-world industrial settings based on the preliminary field test mentioned.

Strengths

• Model trained on the data achieved a reported mean recognition accuracy of 85.56%.
• The dataset was used in a preliminary field test in a real mining environment, suggesting some real-world validation.

Limitations

• The dataset is very small at 5.5 KB, indicating extremely limited scope.
• Row count and column-level documentation are absent; field semantics must be inferred after download.
• Data may reflect geographic or environmental bias inherent to the specific mining conditions tested.

Provenance

Source

figshare, author Ying Lu.

Collection Method

Constructed for network training of a proposed deep learning model.

Freshness

Last updated 2026-05-15 17:37:07; freshness should be verified.

Geography

Likely related to mining environments, but specific geography is not stated.