Wide Field‐of‐View Multiscale Noncontact Photoacoustic Intelligent Defect‐Detection Algorithm

Jijing Chen; Yihan Pi; Yixuan Pang; Hao Zhang; Kaixuan Ding; Ying Long; Jiao Li; Zhen Tian

doi:10.3788/CJL240877

Chinese Journal of Lasers, Volume. 51, Issue 21, 2109002(2024)

Wide Field‐of‐View Multiscale Noncontact Photoacoustic Intelligent Defect‐Detection Algorithm

Jijing Chen¹, Yihan Pi¹, Yixuan Pang¹, Hao Zhang¹, Kaixuan Ding¹, Ying Long¹, Jiao Li^1,2、*, and Zhen Tian^1,2、**

Author Affiliations

¹School of Precision Instrument and Opto-Electronics Engineering, Tianjin University, Tianjin 300072, China

²Georgia Tech Shenzhen Institute, Tianjin University, Shenzhen 518067, Guangdong , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

$Modeling of the elasto-optical process in flip-chip samples. (a) Elasto-optical model of an intrinsic silicon substrate coated with an aluminum film; (b) elasto-optical model of an intrinsic silicon substrate without an aluminum coating; (c) relationship between reflectivity variations ΔR and refractive index mismatch Δn at the boundary, for different δnmaterial values in the coated substrate; (d) relationship between reflectivity variations ΔR and refractive index mismatch Δn at the boundary, for different δnmaterial values in the uncoated substrate$

Fig. 1. Modeling of the elasto-optical process in flip-chip samples. (a) Elasto-optical model of an intrinsic silicon substrate coated with an aluminum film; (b) elasto-optical model of an intrinsic silicon substrate without an aluminum coating; (c) relationship between reflectivity variations $Δ R$ and refractive index mismatch $Δ n$ at the boundary, for different $δ n_{m a t e r i a l}$ values in the coated substrate; (d) relationship between reflectivity variations $Δ R$ and refractive index mismatch $Δ n$ at the boundary, for different $δ n_{m a t e r i a l}$ values in the uncoated substrate

Download full size

Fig. 2. Construction and characterization of the NINC-PAM system. (a) Schematic diagram of the NINC-PAM system; (b) optical-mechanical joint scanning image of a tungsten wire mesh in water; (c) resolution characterization results of NINC-PAM

Download full size

Fig. 3. Schematic diagrams of different improvement modules in Chip-YOLO. (a) Small object detection (SOD) layer; (b) SPPF module improved with LSKA; (c) RepGFPN

Download full size

Fig. 4. Architecture diagram of the multi-scale defect-detection algorithm Chip-YOLO

Download full size

Fig. 5. Preparation process of the flip-chip sample. (a) Spin coating; (b) dehydration; (c) photolithography; (d) development;

Download full size

Fig. 6. Process of establishing a dataset of delamination defects in the flip-chip samples. (a) Imaging result from NINC-PAM;

Download full size

Fig. 7. Intelligent detection results of various algorithms based on NINC-PAM (All images share a common scale bar, and all scale bars are set at 270 μm). (a) Chip-YOLO; (b) YOLOv3; (c) YOLOX; (d) YOLOv7; (e) YOLOv8; (f) Faster R-CNN; (g) corresponding bright-field microscopy image of the detection region

Download full size

Table 1. Thermal and acoustic parameters of different materials in flip chips

View table

Table 1. Thermal and acoustic parameters of different materials in flip chips

Material	Speed of sound /（m·s^-1）	Thermal diffusivity /（m²·s^-1）	Beam size of excitation /μm	Thermal relaxation time /ns	Pressure relaxation time /ns
Silicon	8433^［16］	8.8×10^-5［17］	3	102.27	0.36
Aluminium	6420^［18］	9.8×10^-5［19］	3	91.97	0.47

Table 2. Statistical results of defect sizes in defect datasets

View table

Table 2. Statistical results of defect sizes in defect datasets

Size

Defect pixel area /

（pixel×pixel）

Actual area of the defect /

（pixel×pixel）

Number of defects

Proportion of defects /%

Small

<32×32

15109

50.86

Medium

32×32 to 96×96

8924

30.04

Large

>96×96

5673

19.10

Table 3. Performance comparison of different object detection algorithms on the validation dataset
View table
Table 3. Performance comparison of different object detection algorithms on the validation dataset
Algorithm AP /% Number of parameters /MB Amount of computation （GFLOPs）
YOLOv3 49.6 61.5 154.9
YOLOX 53.6 25.3 73.8
YOLOv7 55.4 36.9 104.7
Faster R-CNN 55.9 41.5 91.4
Chip-YOLO 60.1 12.4 39.8

Table 4. Performance comparison of the algorithm after introducing various enhancement modules

View table

Table 4. Performance comparison of the algorithm after introducing various enhancement modules

Object detection algorithm	AP /%	Number of parameters /MB	Amount of computation （GFLOPs）
YOLOv8	56.8	10.6	28.8
YOLOv8+SOD	58.1	10.1	36.9
YOLOv8+SOD+ LSKA	58.7	11.2	37.8
YOLOv8+SOD+ LSKA+RepGFPN （Chip-YOLO）	60.1	12.4	39.8

Table 5. Intelligent detection results statistics of different algorithms

View table

Table 5. Intelligent detection results statistics of different algorithms

Algorithm	Maximum confidence / arb. units	Minimum confidence / arb. units	Average confidence / arb. units	Confidence greater than 50% /%	Inference time /ms
Chip-YOLO	0.95	0.31	0.62	41	12.2
YOLOv3	0.80	0.20	0.49	20	28.3
YOLOX	0.86	0.21	0.50	19	16.2
YOLOv7	0.86	0.21	0.54	29	12.1
YOLOv8	0.92	0.24	0.56	34	8.1
Faster R-CNN	0.95	0.36	0.63	42	42.4

Tools

Get Citation

Copy Citation Text

Jijing Chen, Yihan Pi, Yixuan Pang, Hao Zhang, Kaixuan Ding, Ying Long, Jiao Li, Zhen Tian. Wide Field‐of‐View Multiscale Noncontact Photoacoustic Intelligent Defect‐Detection Algorithm[J]. Chinese Journal of Lasers, 2024, 51(21): 2109002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites