Terahertz defect detection method based on significant spectral feature fusion and spiking neural network

Abstract

Composite insulation equipment is widely used in power systems, and the detection of internal material interface defects is an important and difficult task in power grids. Terahertz wave can help detect internal faults in equipment in a timely manner, but the analysis of large amounts of data has brought about significant labor costs. In order to improve the efficiency of Terahertz wave detection, this paper proposes a Terahertz wave defect detection method based on spectrum feature fusion and spiking neural network. By performing multiple wavelet basis transformations on terahertz wave timing waveforms, and then performing feature fusion extraction on the data through a self encoder incorporating spatial and channel attention mechanisms, the differences between different defect detection waveforms are expanded. Then, spiking neural networks are used to classify the feature fused spectra to obtain defect detection results. Compared with other typical models, this model performs better in terms of accuracy and training costs.