Research on the classification method of different quality dry alfalfa based on scanning electron microscopy (SEM) image texture analysis

Figures & data

Figure 1. Scanning electron microscopy imaging system schematic diagram.

Figure 2. Comparison of the images and histograms of the gray images processed by histogram equalization (a) Histogram gray alfalfa image of the original image (b) Enhanced histogram of the alfalfa image (c) Gray alfalfa image (d) Enhanced alfalfa image.

Table 1. Definition of each texture feature parameter

Parameter	Formula
Contrast	$f_1=\sum _{n=0}^{N_g-1}{n}^2\left\{\sum _{i=1}^{N_g}\sum _{j=1}^{N_g}p\left(i,j\right)\left|i-j\right|=n\right\}$
Correlation	$f_2=\frac{{\sum }_i{\sum }_j\left(i,j\right)p\left(i,j\right)-{\mu }_x{\mu }_y}{{\sigma }_x{\sigma }_y}$
Prominence of Clustering	$f_3=\sum _i\sum _j{\left(i+j-{\mu }_x-{\mu }_y\right)}^{4}p\left(i,j\right)$
Shadow of Clustering	$f_4=\sum _i\sum _j{\left(i+j-{\mu }_x-{\mu }_y\right)}^{3}p\left(i,j\right)$
Angular Second Moment	$f_5=\sum _i\sum _j{\left\{p\left(i,j\right)\right\}}^2$
Entropy	$f_6=-\sum _i\sum _{j}p\left(i,j\right)log\left(p\left(i,j\right)\right)$
Inverse Difference Moment	$f_7=\sum _i\sum _j\frac{1}{1+\left(i-j\right)}p\left(i,j\right)$
Maximal Probability	$f_8=\begin{array}{c}\mathrm{M}\mathrm{A}\mathrm{X}\\ i,j\end{array}p\left(i,j\right)$
Sum of Squares Variance	$f_9=\sum _i\sum _j{\left(i-\mu \right)}^{2}p\left(i,j\right)$
Sum of Average	$f_{10}=\sum _{i=2}^{2N_g}i{p}_{x+y}\left(i\right)$
Sum of Variance	$f_{11}=\sum _{i=2}^{2N_g}{\left(i-f_8\right)}^2{p}_{x+y}\left(i\right)$
Sum of Entropy	$f_{12}=-\sum _{i=2}^{2N_g}{p}_{x+y}\left(i\right)log\left\{p_{x+y}\left(i\right)\right\}$
Variance of Difference	$f_{13}=varianceof{p}_{x-y}$
Difference of Entropy	$f_{14}=-\sum _{i=0}^{N_g-1}{p}_{x-y}\left(i\right)log\left\{p_{x-y}\left(i\right)\right\}$

Figure 3. Cluster diagram of each different quality of dry alfalfa.

Figure 4. Score diagram of the two discriminant functions.

Table 2. Classification accuracy rate

Sample	Measured	Shade dried	Sun dried	Mold treatment	Correct percentage
Train	Shade dried	17	0	1	94.4%
	Sun dried	0	15	1	93.8%
	Mold treatment	1	1	13	86.7%
	Overall percentage	36.7%	32.7%	30.6%	91.8%
Calibration	Shade dried	3	0	0	100.0%
	Sun dried	1	6	1	75.0%
	Mold treatment	0	0	4	100.0%
	Overall percentage	26.7%	40.0%	33.3%	86.7%
Validation	Shade dried	3	0	1	75.0%
	Sun dried	0	1	0	100.0%
	Mold treatment	0	0	6	100.0%
	Overall percentage	46.7%	20.0%	33.3%	100.0%

Figure 5. Observation forecast chart.

Table 3. Classification results

Sample	Measured	Shade dried	Sun dried	Mold treatment	Correct percentage
Train	Shade dried	13	0	0	100.0%
	Sun dried	0	13	0	100.0%
	Mold treatment	0	0	12	100.0%
	Overall percentage	34.2%	34.2%	31.6%	100.0%
Calibration	Shade dried	5	0	0	100.0%
	Sun dried	0	9	0	100.0%
	Mold treatment	0	0	8	100.0%
	Overall percentage	22.7%	40.9%	36.4%	100.0%
Validation	Shade dried	7	0	0	100.0%
	Sun dried	0	3	0	100.0%
	Mold treatment	0	0	5	100.0%
	Overall percentage	46.7%	20.0%	33.3%	100.0%

Figure 6. Observation forecast chart.

Table 4. Identification results of each model

Model	Recognition(%)
PCA-ANN	93.33
LDA-ANN	100
PCA-LS-SVM	86.667
LDA-LS-SVM	100

Figure 7. PCA input results of the ANN model (a) Network training error curve (b) Schematic diagram of the neural network output value and target value comparison.

Figure 8. LDA input results of the ANN model (a) Network training error curve (b) Schematic diagram of the neural network output value and target value comparison.

Figure 9. Classification results.

Figure 10. Classification results.