Laser Spot Focusing State Discrimination Algorithm Based on Time Series Prediction

Ziwen Ren; Huaiguang Liu; Wei Sun

doi:10.3788/LOP241779

Laser & Optoelectronics Progress, Volume. 62, Issue 6, 0615015(2025)

Laser Spot Focusing State Discrimination Algorithm Based on Time Series Prediction

Ziwen Ren^1,2、*, Huaiguang Liu^1,2, and Wei Sun^1,2

¹Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering, Wuhan University of Science and Technology, Wuhan 430081, Hubei , China

²Key Laboratory of Metallurgical Equipment and Control Technology, Ministry of Education, Wuhan University of Science and Technology, Wuhan 430081, Hubei , China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(15)

Fig. 1. Laser autofocus device model

Download full size

Fig. 2. Overall framework of PCLT-Net

Download full size

Fig. 3. PMA-ConvNeXt model structure

Download full size

Fig. 4. ConvNeXt module structure

Download full size

Fig. 5. SE attention mechanism structure

Download full size

Fig. 6. ECA attention mechanism structure

Download full size

Fig. 7. LSTM unit structure

Download full size

Fig. 8. Temporal attention mechanism model

Download full size

Fig. 9. Laser autofocus platform. (a) Focus alignment mechanism; (b) laser spot imaging system

Download full size

Fig. 10. A set of image sequences. (a) Six frames arranged in time order; (b) label image

Download full size

Fig. 11. Comparison of training loss and training accuracy. (a) Training accuracy of Focus-Lite; (b) Training accuracy of Focus-Med; (c) Training accuracy of Focus-Extreme; (d) training loss of Focus-Lite; (e) training loss of Focus-Med; (f) training loss of Focus-Extreme

Download full size

Table 1. Dataset collection
View table
Table 1. Dataset collection
Dataset Degree of jitter Duration /s Total frame
Focus-Lite Light 46 1047
Focus-Med Medium 44 1015
Focus-Extreme Severe 76 1751

Table 2. Dataset enhancement
View table
Table 2. Dataset enhancement
Dataset Number of focus sequences Number of out-of-focus sequences Total number of sequences
Focus-Lite 572 470 1042
Focus-Med 751 693 1444
Focus-Extreme 739 984 1723

Table 3. Accuracy comparison before and after adding attention mechanism

View table

Table 3. Accuracy comparison before and after adding attention mechanism

Model	Focus-Lite		Focus-Med		Focus-Extreme
Model	×	√	×	√	×	√
ShuffleNetV2-LSTM	84.9	85.6	86.2	87.3	80.5	81.2
MobileNetV3-S-LSTM	84.3	86.9	88.6	89.8	84.2	86.1
EfficientNet-B0-LSTM	87.8	91.6	89.2	90.3	84.8	86.6
PMA-ConvNeXt-LSTM	91.3	93.2	90.0	91.2	85.6	88.7

Table 4. Comparison of parameters, prediction time and accuracy of different network models

View table

Table 4. Comparison of parameters, prediction time and accuracy of different network models

Model	Parameter quantity /MB	Focus-Lite			Focus-Med			Focus-Extreme
Model	Parameter quantity /MB	Training time /s	Testing time /s	Accuracy / %	Training time /s	Testing time /s	Accuracy / %	Training time /s	Testing time /s	Accuracy / %
ShuffleNetV2-LSTM-TA	1.25	30.4	0.05	85.6	37.7	0.05	86.3	48.3	0.06	81.2
MobileNetV3-S-LSTM-TA	0.93	12.3	0.05	86.9	19.5	0.05	88.8	22.8	0.05	83.1
EfficientNet-B0- LSTM-TA	4.01	48.2	0.05	91.6	54.6	0.06	89.3	57.8	0.07	86.6
PCLT-Net	27.94	11.6	0.05	93.2	13.6	0.05	90.2	17.6	0.06	86.7

Tools

Get Citation

Copy Citation Text

Ziwen Ren, Huaiguang Liu, Wei Sun. Laser Spot Focusing State Discrimination Algorithm Based on Time Series Prediction[J]. Laser & Optoelectronics Progress, 2025, 62(6): 0615015

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites