Automated plant identification using artificial neural network and support vector machine

Figures & data

Figure 1. Shadow removing pre-processing. Thin streak of shadow can still be seen in (c), which is then removed in (d).

Figure 2. Image segmentation process.

Figure 3. Selected shape features. (a) Area, (b) perimeter and (c) major and minor axes.

Table 1. Features extracted for input vectors.

Features	No.
Shape features	5
Area, perimeter, eccentricity, minor axis, major axis
Moment features	9
Moment invariants
Central moments
Center of mass
Co-occurrence matrix	64
Histogram of oriented gradients	80
Total number of features	158

Table 2. Sample allocation of both machine learning algorithms.

Model algorithm	ANN	SVM
Training	42 (70%)	48 (80%)
Validation	6 (10%)	–
Testing	12 (20%)	12 (20%)
Total	60 (100%)	60 (100%)

Table 3. Classification results of (a) artificial neural network model and (b) support vector machine.

(a)	Actual	Ficus benjamina	Ficus pellucidopunctata	Ficus sumatrana	(b)	Actual	Ficus benjamina	Ficus pellucidopunctata	Ficus sumatrana
	Predicted					Predicted
	Ficus benjamina	3 (75.0%)	0	1 (25.0%)		Ficus benjamina	3 (75.0%)	0	1 (25.0%)
	Ficus pellucidopunctata	0	4 (100%)	0		Ficus pellucidopunctata	1 (25.0%)	4 (100%)	0
	Ficus sumatrana	1 (25.0%)	0	3 (75.0%)		Ficus sumatrana	0	0	3 (75.0%)

Note: Outputs of models (rows) were compared with the target species (columns).

Figure 4. The receiver operating characteristics (ROC) curve of both models.

Table 4. Accuracy and area under curve (AUC) comparison of both models.

		Area under curve (AUC)
Models	Accuracy	F. benjamina	F. pellucidopunctata	F. sumatrana	Total
Artificial neural network	83.3%	0.9375	1.0000	0.7188	2.6563
Support vector machine	83.3%	0.7188	0.8750	0.9063	2.5001
Difference	0	0.2187	0.1250	−0.1875	0.1562