Fig. 1. Actual hot melting welding process of polyethylene sheet

Fig. 2. Physical of THz-TDS system

Fig. 3. THz detection simulation diagram of polyethylene pipe

Fig. 4. THz detection waveform comparison

Fig. 5. The sample has not undergone planar processing and corresponding THz kurtosis imaging after hot fusion welding

Fig. 6. The Incomplete fusion sample before welding and the corresponding THz detection kurtosis imaging diagram

Fig. 7. The embedded object of inclusion defect sample and the corresponding THz kurtosis imaging are required

Fig. 8. Defect classification flow based on wavelet scattering network-convolution neural network

Fig. 9. Wavelet scattering framework

Fig. 10. CNN Network architecture

Fig. 11. Accuracy and loss of each iteration of the first training model

Fig. 12. Confusion matrix of test results using the first kind of training model

Fig. 13. Accuracy and loss of each iteration of the second training model

Fig. 14. Confusion matrix of test results using the second kind of training model

Fig. 15. The comparison chart between the number of defect recognition and the actual number of two training models

Fig. 16. Curve diagram of error between the number of defect recognition and the actual number of two training models