研究目的
To propose a new DEMON spectrum extraction method using empirical mode decomposition (EMD) for ship noise classification, where the band number and width are automatically determined.
研究成果
The new DEMON spectrum extraction method using EMD automatically determines band number and width, replacing traditional bandpass filters. In tests with five ship types, a feedforward neural network achieved 91.6% correct classification, demonstrating effectiveness. Future work could involve using all IMFs for enhanced classification accuracy.
研究不足
The method relies on the first IMF for demodulation, which may not capture all modulation effects; other IMFs could be explored for improved accuracy and robustness. The bandwidth and number of IMFs are data-dependent, which might vary with different noise characteristics.
1：Experimental Design and Method Selection:
The study uses empirical mode decomposition (EMD) to decompose ship noise signals into intrinsic mode functions (IMFs) as an adaptive alternative to traditional bandpass filtering in DEMON analysis. A feedforward neural network is employed for classification based on extracted DEMON spectra.
2：Sample Selection and Data Sources:
Five kinds of ship noise signals are used, with details provided in Table 1 (e.g., lengths from 132 s to 684 s, sampling frequency 44100 Hz). Signals are divided into segments of 10 seconds with 9-second overlaps for analysis.
3：List of Experimental Equipment and Materials:
Not specified in the paper.
4：Experimental Procedures and Operational Workflow:
The signal is decomposed using EMD into IMFs. The first IMF is selected for demodulation due to its dominant modulation effect. The envelope is extracted (methods like absolute value or square operation are mentioned), and FFT is applied to obtain the DEMON spectrum. Features (frequency and power of significant lines) are input into a feedforward neural network for classification.
5：Data Analysis Methods:
Scaled conjugate gradient method is used for neural network training, with cross-entropy as the performance metric. Data is split into training, validation, and test sets. Performance is evaluated using percentage correct classification and confusion matrix.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
