Tianyu Liu

Tianyu Liu

Auto-generated profile, based on publicly available information

Tianyu Liu


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 July 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
CGCanzhi Guo
Canzhi Guo
TLTianyu Liu
Tianyu Liu
TYTianxue Yang
Tianxue Yang
ZZZhongqiang Zhang
Zhongqiang Zhang
GCguanggui cheng
guanggui cheng
Industrial ultrasonic Total Focusing Method (TFM) imaging has become a mainstream non-destructive testing technology for workpiece internal defects in petrochemical, nuclear power and aerospace fields due to its high resolution and full-range focusing performance. Nevertheless, TFM images are plagued by intense speckle noise and low signal-to-noise ratio(SNR) of defect edges, which hinders the distinction between defect features and background. These issues cause existing deep learning-based detection models to suffer from missed and false detections, with limited detection accuracy and generalization capability. To address these issues, this paper presents FDE-YOLO, an improved model for defect detection of industrial ultrasonic TFM images. A dual-branch Feature Edge Enhancement and Calibration (FEEC) module is embedded into the YOLO11n backbone to enhance weak defect edges and suppress speckle noise. Meanwhile, the Feature Cleanliness Coefficient (FCC) is proposed to quantitatively evaluate the noise suppression effect of heat maps. Dynamic upsampling is introduced into the neck network to mitigate the loss of defect details with low SNR. Furthermore, the Efficient Multi-scale Attention (EMA) module is integrated to strengthen the perception of small targets, amplify defect-related feature responses and suppress background noise interference. Experimental results on a self-built dataset show that FDE-YOLO achieves a Precision of 0.9227, Recall of 0.8256 and mAP@0.5 of 0.9399. Its GFLOPs rise slightly from 6.3 to 7.7, and the inference speed reaches 344.31 FPS, fully satisfying industrial real-time detection requirements. This work provides a reliable solution for TFM-based defect detection and promotes the intelligent upgrading of industrial non-destructive testing.