Classification of canola seed varieties based on multi-feature analysis using computer vision approach

Figures & data

Figure 1. Digital images of eight canola varieties

Figure 2. Range Oriented Pixel-based Segmentation (ROPS) Algorithm

Table 1. Correlation-based Feature Selection (CFS) Table

S. No.	Multi-Features
1	Histo_Mean
2	Histo_Standard_Deviation
3	Texture_Energy_Average
4	Texture_Energy_Range
5	Inertia_Average
6	Correlation Average
7	Correlation Range
8	Inverse_Diff_Average
9	Inverse_Diff_Range
10	Texture_Entropy_Range
11	Spectral_DC
12	Ring1
13	Sector1
14	Sector2
15	Sector3

Table 2. Employed ANN architecture for multi-feature dataset on canola seed varieties

Input layers	Hidden Layers	Neurons	Learning rate	Momentum	Validation Threshold	Epochs	Output Layers
15	1	10	0.3	0.2	20	500	8

Figure 3. The canola seed classification framework

Figure 4. The ANN model for eight canola seed varieties

Table 3. Confusion Matrix for canola varieties classification with ROI (256 ×256)

Canola Varieties	CON-I	CON-I1	CON-I11	Pakola	Canola Raya	Rainbow	PARC Canola Hybrid	Tarnab-III	Total Dataset	Classification Percentage
CON-I	158	10	12	4	6	2	1	7	200	79%
CON-I1	15	150	10	8	7	0	6	4	200	75%
CON-I11	20	7	160	8	3	0	9	3	200	80%
Pakola	3	6	18	162	8	0	3	0	200	81%
Canola Raya	2	9	12	7	166	0	0	4	200	83%
Rainbow	0	16	20	13	6	144	1	0	200	77%
PARC Canola Hybrid	0	11	14	4	2	1	168	0	200	84%
Tarnab-III	4	10	8	9	5	0	0	164	200	82%

Table 4. Confusion Matrix for canola varieties classification with ROI (512 ×512)

Canola Varieties	CON-I	CON-II	CON-III	Pakola	Canola Raya	Rainbow	PARC Canola Hybrid	Tarnab-III	Total Dataset	Classification Percentage
CON-I	190	5	2	1	0	1	1	0	200	95%
CON-II	1	190	5	0	3	0	0	1	200	95%
CON-III	2	3	192	1	2	0	0	0	200	96%
Pakola	1	3	2	190	0	0	0	4	200	95%
Canola Raya	1	2	4	0	190	3	0	0	200	95%
Rainbow	0	1	0	2	0	196	1	0	200	98%
PARC Canola Hybrid	0	2	4	0	3	1	190	0	200	95%
Tarnab-II1	2	1	0	0	5	2	0	192	200	96%

Table 5. A comparison between the proposed and current techniques

Source/References	Techniques	Classifiers	Accuracy
(Shifa, Naweed, Omar, Jhandir, & Ahmad, 2011)	RGB+ Infra-red and shortwave infra-red	K-Nearest Neighbor (K-NN)	98%
(Rehmani, 2015)	Texture + Spectral Features	Artificial Neural Network (ANN)	96%
Gonzales, Andreu, Potter, & Rodas, 2014)	RGB Color Features	Principle Component Analysis (PCA)	92%
Szczypinski, Klepaczko, & Zapotoczny, 2015	Texture + Shape+ Color Features	Artificial Neural Network (ANN)	86%
(Shahid, Naweed, Qadri, Mutiullah, & Rehman, 2016)	Texture Features	Artificial Neural Network (ANN)	96%
(Ropelewska, Zapotoczny, Budzynski, & Jankowski, 2017)	Texture +Geometric Features	Artificial Neural Network (ANN)	92%
Khan, Muhammad Attique 2019	Fused Features (Color + Color Histogram + Local Binary Patterns)	multi-class support vector machine (M-SVM)	98.10%
Aqib Ali et.al, 2020	Hybrid Features (histogram, texture, and spectral features)	Multilayer Perceptron	98%
Proposed Methodology	Multi Features (Histogram, Texture, Binary, RST and spectral Features)	Correlation Based Feature Selection (CFS) + Artificial Neural Network	95–98%

Figure 5. Canola Varieties Confusion Graph for ROI (256 ×256)

Figure 6. Canola varieties classification graph with ROI (256 ×256)

Figure 7. Canola Varieties Confusion Graph for ROI (512 ×512)

Figure 8. Canola varieties classification graph with ROI (512 ×512)