Prediction method of human defecation based on informer audio data augmentation and improved residual network

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Signal Processing

Background:

• Defecation care for disabled individuals presents significant health management challenges.
• Current methods for predicting defecation are either physically harmful or intrusive.
• Bowel sounds are identified as a potential indicator for defecation intention.

Purpose of the Study:

• To develop a non-intrusive method for predicting human defecation using deep learning and bowel sound analysis.
• To address the limitations of existing defecation forecasting techniques.
• To improve the quality of life and health management for disabled patients.

Main Methods:

• Utilized a wavelet domain-based Wiener filter to denoise bowel sound data.
• Extracted integrated features from bowel sounds in time, frequency, and time-frequency domains using statistical analysis, FFT, and WPT.
• Developed an audio signal expansion algorithm based on the Informer model to augment limited bowel sound data.
• Designed an improved one-dimensional residual network (1D-IResNet) for defecation classification.

Main Results:

• The proposed bowel sound augmentation strategy effectively increased data sample size and diversity.
• The 1D-IResNet model demonstrated accelerated training speed on the augmented dataset.
• Achieved a high classification accuracy of 90.54% and an F1 score of 83.88%.
• Demonstrated good classification stability with high performance metrics.

Conclusions:

• Deep learning-based bowel sound recognition is a viable and effective method for human defecation prediction.
• The developed AI system offers a promising, non-intrusive solution for defecation care, particularly for disabled patients.
• The study highlights the potential of leveraging physiological sounds for predictive health monitoring.