Material Identification of Particles in Space-Borne Electronic Equipments Based on Principal Component Analysis - Intelligent Computing for Sustainable Energy and Environment - page 467

Tabl e 1. SVM output with different top number of principal components

PCs SVs Margin Error rate

1 496(99.2%) 0.000001 45.4%

2 424(84.8%) 0.000001 38.8%

3 446(35.7%) 0.000067 27.3%

4 52(10.4%) 0.001007 17.6%

5 46(9.2%) 0.008764 13.1%

6 31(6.2%) 0.087357 10.2%

7 37(7.4%) 0.087546 11.3%

8 37(7.4%) 0.087822 13.6%

9 39(7.8%) 0.088081 14.2%

Tabl e 2. SVM output using PCA features

Classifier output

Wire Al Tin Accuracy

Wire 237 5 8 94.8%

Al 5 227 18 90.8%

Tin 4 225 21 90.0%

Tabl e 3. SVM output using features directly

Classifier output

Wire Al Tin Accuracy

Wire 228 13 9 91.2%

Al 13 205 32 82.0%

Tin 10 210 30 84.0%

5Con lu on

In this paper, a material identification method based on principal component
analysis has been investigated for the particle remainders on space-borne elec-
tronic equipments. Some valuable conclusions have been drawn as follows:
(1) Nine features have been identifiedinthetimeandfrequencydomains.
PCA is then used for further feature extraction.
(2) The first six principal components are used as the inputs to the SVM model
and they give a better classification performance than directly using the nine
features to build the SVM model. The accuracy of particle material identification
is above 90% in the experiments.
(3) In view of the dimensionality reduction and noise reduction capabilities
of PCA and generalization ability of SVM, the proposed particle material iden-
tification method can be extended to other sealed devices.