Limits...
Optimal Sensor Selection for Classifying a Set of Ginsengs Using Metal-Oxide Sensors.

Miao J, Zhang T, Wang Y, Li G - Sensors (Basel) (2015)

Bottom Line: The relation of the minimum numbers of sensors with number of samples in the sample set was revealed.The results showed that as the number of samples increased, the average minimum number of sensors increased, while the increment decreased gradually and the average optimal classification rate decreased gradually.Moreover, a new approach of sensor selection was proposed to estimate and compare the effective information capacity of each sensor.

View Article: PubMed Central - PubMed

Affiliation: State Key Laboratory of Industrial Control Technology, Institute of Cyber Systems and Control, Zhejiang University, Hangzhou 310027, Zhejiang, China. jiacheng@zju.edu.cn.

ABSTRACT
The sensor selection problem was investigated for the application of classification of a set of ginsengs using a metal-oxide sensor-based homemade electronic nose with linear discriminant analysis. Samples (315) were measured for nine kinds of ginsengs using 12 sensors. We investigated the classification performances of combinations of 12 sensors for the overall discrimination of combinations of nine ginsengs. The minimum numbers of sensors for discriminating each sample set to obtain an optimal classification performance were defined. The relation of the minimum numbers of sensors with number of samples in the sample set was revealed. The results showed that as the number of samples increased, the average minimum number of sensors increased, while the increment decreased gradually and the average optimal classification rate decreased gradually. Moreover, a new approach of sensor selection was proposed to estimate and compare the effective information capacity of each sensor.

No MeSH data available.


The mean standard deviation of classification accuracy with increasing number of failed sensor for LDA, SVM-RBF and KNN (k = 1, 3, 5, 7).
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4541866&req=5

sensors-15-16027-f007: The mean standard deviation of classification accuracy with increasing number of failed sensor for LDA, SVM-RBF and KNN (k = 1, 3, 5, 7).

Mentions: The optimal classification accuracy of sample set A1 = 1, 2, 3, 4, 5, 6, 7, 8, 9 was achieved with 10 (out of 12) sensors in Section 3.2. System robustness for sample set A1was investigated with P (P = 0, 1… 9) (out of 10) failed sensors. For P failed sensors, there werepossible failure combinations. The mean and standard deviation of the system classification accuracy with P failed sensors were used to evaluate system robustness. LDA, SVM-RBF and KNN (k = 1, 3, 5, 7) were employed and compared as classifier. From Figure 7, we observed that performance of LDA and SVM-RBF is critically affected by sensor failure while KNN performs much better than the other two classifiers. How to improve the robustness of a system suffering from sensor failure is a meaningful subject to which we will pay attention in future work.


Optimal Sensor Selection for Classifying a Set of Ginsengs Using Metal-Oxide Sensors.

Miao J, Zhang T, Wang Y, Li G - Sensors (Basel) (2015)

The mean standard deviation of classification accuracy with increasing number of failed sensor for LDA, SVM-RBF and KNN (k = 1, 3, 5, 7).
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4541866&req=5

sensors-15-16027-f007: The mean standard deviation of classification accuracy with increasing number of failed sensor for LDA, SVM-RBF and KNN (k = 1, 3, 5, 7).
Mentions: The optimal classification accuracy of sample set A1 = 1, 2, 3, 4, 5, 6, 7, 8, 9 was achieved with 10 (out of 12) sensors in Section 3.2. System robustness for sample set A1was investigated with P (P = 0, 1… 9) (out of 10) failed sensors. For P failed sensors, there werepossible failure combinations. The mean and standard deviation of the system classification accuracy with P failed sensors were used to evaluate system robustness. LDA, SVM-RBF and KNN (k = 1, 3, 5, 7) were employed and compared as classifier. From Figure 7, we observed that performance of LDA and SVM-RBF is critically affected by sensor failure while KNN performs much better than the other two classifiers. How to improve the robustness of a system suffering from sensor failure is a meaningful subject to which we will pay attention in future work.

Bottom Line: The relation of the minimum numbers of sensors with number of samples in the sample set was revealed.The results showed that as the number of samples increased, the average minimum number of sensors increased, while the increment decreased gradually and the average optimal classification rate decreased gradually.Moreover, a new approach of sensor selection was proposed to estimate and compare the effective information capacity of each sensor.

View Article: PubMed Central - PubMed

Affiliation: State Key Laboratory of Industrial Control Technology, Institute of Cyber Systems and Control, Zhejiang University, Hangzhou 310027, Zhejiang, China. jiacheng@zju.edu.cn.

ABSTRACT
The sensor selection problem was investigated for the application of classification of a set of ginsengs using a metal-oxide sensor-based homemade electronic nose with linear discriminant analysis. Samples (315) were measured for nine kinds of ginsengs using 12 sensors. We investigated the classification performances of combinations of 12 sensors for the overall discrimination of combinations of nine ginsengs. The minimum numbers of sensors for discriminating each sample set to obtain an optimal classification performance were defined. The relation of the minimum numbers of sensors with number of samples in the sample set was revealed. The results showed that as the number of samples increased, the average minimum number of sensors increased, while the increment decreased gradually and the average optimal classification rate decreased gradually. Moreover, a new approach of sensor selection was proposed to estimate and compare the effective information capacity of each sensor.

No MeSH data available.