Identification of Common Knife LIBS Spectra Using 1D-CNN Combined with Data Augmentation

Tao Zhang; Chunyu Li; Chuanzhao Li

doi:10.3788/LOP241095

Laser & Optoelectronics Progress, Volume. 62, Issue 3, 0330002(2025)

Identification of Common Knife LIBS Spectra Using 1D-CNN Combined with Data Augmentation

Tao Zhang^*, Chunyu Li, and Chuanzhao Li

Author Affiliations

Institute of Criminal Investigation, People's Public Security University of China, Beijing 100038, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Schematic of experimental setup

Download full size

Fig. 2. Principle of 1D-CNN

Download full size

Fig. 3. Original spectra of samples. (a) Sample number 40; (b) sample number 100

Download full size

Fig. 4. PCA flowchart

Download full size

Fig. 5. Interpretation rate and cumulative interpretation rate of the top 400 principal components

Download full size

Fig. 6. Accuracy curves and loss function curves

Download full size

Fig. 7. Synthetic data of sample number 40. (a) Normal distribution method; (b) linear combination method

Download full size

Fig. 8. Synthetic data of sample number 100. (a) Normal distribution method; (b) linear combination method

Download full size

Fig. 9. Average prediction accuracy

Download full size

Table 1. Parameter settings of 1D-CNN

View table

Table 1. Parameter settings of 1D-CNN

Neural network layer	Model parameter
Input layer	Preprocessed LIBS data
Convolutional layer 1	Kernel_size is 1×3，stride is 1，filters are 64，Relu
Pooling layer 1	Max pooling，pool_size is 1×2，strides are 2
Convolutional layer 2	Kernel_size is 1×3，stride is 1，filters are 128，Relu
Pooling layer 2	Max pooling，pool_size is 1×2，strides are 2
Convolutional layer 3	Kernel_size is 1×3，stride is 1，filters are 256，Relu
Pooling layer 3	Max pooling，pool_size is 1×2，strides are 2
Convolutional layer 4	Kernel_size is 1×3，stride is 1，filters are 512，Relu
Pooling layer 4	Max pooling，pool_size is 1×2，strides are 2
Convolutional layer 5	Kernel_size is 1×3，stride is 1，filters are 512，Relu
Pooling layer 5	Global max pooling
Fully connected layer 1	128 neurons
Dropout layer 1	Randomly dropping out 20% of the neurons
Fully connected layer 2	64 neurons
Dropout layer 2	Randomly dropping out 20% of the neurons
Output layer	138 neurons

Table 2. Prediction accuracy of normal distribution method
View table
Table 2. Prediction accuracy of normal distribution method
Expansion amount Prediction accuracy
Max Min Average
0 0.9674 0.9457 0.9602
40 0.9710 0.9638 0.9686
80 0.9783 0.9710 0.9746
120 0.9819 0.9710 0.9746

Table 3. Prediction accuracy of linear combination method
View table
Table 3. Prediction accuracy of linear combination method
Expansion amount Prediction accuracy
Max Min Average
0 0.9674 0.9457 0.9602
40 0.9457 0.9023 0.9215
80 0.9565 0.9348 0.9481
120 0.9493 0.9239 0.9348

Tools

Get Citation

Copy Citation Text

Tao Zhang, Chunyu Li, Chuanzhao Li. Identification of Common Knife LIBS Spectra Using 1D-CNN Combined with Data Augmentation[J]. Laser & Optoelectronics Progress, 2025, 62(3): 0330002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites