A novel key information classification method using hybrid feature extraction and a multi-level deep network for...

Area of Science:

• Engineering
• Signal Processing
• Machine Learning

Background:

• High-precision monitoring of large electromechanical equipment during transport is challenging.
• Extracting features from non-stationary signals in noisy environments is difficult.
• Traditional methods lack stability and effective high-dimensional feature classification.

Purpose of the Study:

• To propose a key information classification approach for precise transportation status monitoring.
• To overcome limitations in traditional feature extraction, classification, and stability.
• To enhance structural health and fault monitoring of electromechanical equipment.

Main Methods:

• Developed a multi-level deep network (MLDN) integrating hybrid feature extraction.
• Employed improved feature mode decomposition and a multidimensional criterion (SKEC) for signal reconstruction.
• Utilized a convolutional neural network-bidirectional long short-term memory architecture and relevance vector machine for classification.

Main Results:

• Achieved superior recognition performance and strong noise robustness.
• Demonstrated high-precision and sparsity-enhanced state feature classification.
• Attained an average accuracy of 99.08% for key information classification.

Conclusions:

• The proposed hybrid feature extraction and MLDN approach effectively addresses challenges in equipment transport monitoring.
• The method offers a robust solution for complex, non-stationary signal analysis.
• This technique significantly improves the accuracy and reliability of transportation status monitoring.