HOME PREDICTION RICE VARITIES ALGORITHM STATISTICS DEVELOPERS CONTACT US HELP iRSVPred 2.0
iRSVPred 2.0 | ALGORITHM

Note: The complete dataset and all the scripts are available at: https://github.com/tbgicgeb/iRSVPred-2.0



Work - Flow Diagram



Figure: Flowchart depicting the methodology used for the development of iRSVPred 2.0 web server and an Android-based mobile application.

Figure: Flowchart depicting the methodology used for the development of iRSVPred 2.0 web server and an Android-based mobile application.



InceptionResNetV2 model CNN architecture with different layers.

Model architecture



The distribution of the images into the original dataset used for training and evaluation of AI-based models.

The distribution of the images into the original dataset used for training and evaluation of AI-based models.

The distribution of original and augmented images into a final dataset used for training and evaluation of AI-based models.

The distribution of original and augmented images into a final dataset used for training and evaluation of AI-based models.



Predicted variety
1121 1509 1637 1718 1728 BAS-370 CSR-30 DHBT-3 PB-1 PB-6 Unknown
Actual variety 1121 49 0 0 1 0 0 0 0 0 0 0
1509 0 50 0 0 0 0 0 0 0 0 0
1637 0 0 40 0 0 0 0 0 7 3 0
1718 0 0 0 50 0 0 0 0 0 0 0
1728 5 0 0 0 45 0 0 0 0 0 0
BAS-370 0 0 0 0 0 50 0 0 0 0 0
CSR-30 0 0 0 0 0 0 50 0 0 0 0
DHBT-3 0 0 0 0 0 0 0 50 0 0 0
PB-1 0 0 0 0 0 0 0 0 50 0 0
PB-6 0 0 0 0 0 0 0 0 0 50 0
Unknown 0 0 0 0 0 0 0 0 0 0 142
Evaluation metrics Precision 91 100 100 98 100 100 100 100 88 94 100
Recall 98 100 80 100 90 100 100 100 100 100 100
Accuracy 99 100 98.5 99.9 99.2 100 100 100 99 99.5 100

A confusion matrix showing precision and recall of an AI model (deployed on iRSVPred 2.0) with original images (n=642) in an external dataset.

Predicted variety
1121 1509 1637 1718 1728 BAS-370 CSR-30 DHBT-3 PB-1 PB-6 Unknown
Actual variety 1121 1710 16 0 59 15 0 0 0 0 0 0
1509 13 1757 0 27 3 0 0 0 0 0 0
1637 0 0 1590 0 20 0 3 0 135 52 0
1718 84 60 0 1648 8 0 0 0 0 0 0
1728 32 30 0 17 1704 0 6 0 11 0 0
BAS-370 0 0 0 0 0 1800 0 0 0 0 0
CSR-30 7 4 0 0 0 0 1747 3 30 9 0
DHBT-3 0 0 0 0 0 0 0 1800 0 0 0
PB-1 0 0 0 0 2 0 14 0 1784 0 0
PB-6 0 0 4 0 0 0 9 1 13 1773 0
Unknown 0 0 0 0 0 0 0 0 2 0 5110
Evaluation metrics Precision 93 94 100 94 97 100 98 100 90 97 100
Recall 95 98 88 92 95 100 97 100 99 98 100
Accuracy 99 99.4 98.7 99.9 99.4 100 99.6 99.98 99.1 99.6 99.99

A confusion matrix showing precision and recall of an AI model (deployed on RSVPred 2.0) with augmented images (n=23,112) in an external dataset.





Comparative analysis between iRSVPred and iRSVPred 2.0



S. No. Model name Images used in training Images used in internal validation Training accuracy Validation accuracy Validation loss
1 MAM-I 61632 15408 98 93.4 0.24
2 MAM-II 61632 15408 100 97.7 0.16
3 Inception_ResNet_V2 73944 18504 99.9 96.5 0.135

Overall performance of iRSVPred 2.0 prediction model “InceptionResNetV2” with respect to previous iRSVPred models MAM-I and MAM-II.

S. No. Class Number of Images (iRSVPred) Number of Images (iRSVPred 2.0)
1 1121 1500 1800
2 1509 1500 1800
3 1637 1500 1800
4 1718 1500 1800
5 1728 1500 1800
6 BAS-370 1500 1800
7 CSR-30 1500 1800
8 DHBT-3 1500 1800
9 PB-1 1500 1800
10 PB-6 1500 1800
11 Unknown 4260 5112

The class-wise distribution of original and augmented images in the external validation datasets from iRSVPred and iRSVPred 2.0.



S. No. Class No. of images used for validation Multiple augmentation model (251 epochs) accuracy (%) Multiple augmentation model (502 epochs) accuracy (%) Average of MAM-I and MAM-II Inception_ResNet_V2 model accuracy (%)
1 1121 50 74 72 73 99
2 1509 50 92 92 92 100
3 1637 50 76 84 80 98.5
4 1718 50 26 62 44 99.9
5 1728 50 86 74 80 99.2
6 BAS-370 50 98 100 99 100
7 CSR-30 50 86 64 75 100
8 DHBT-3 50 72 98 85 100
9 PB-1 50 86 90 88 99
10 PB-6 50 80 64 72 99.5
11 Unknown 50 100 100 100 100

The comparative performance of iRSVPred models with external validation dataset of original 642 images in terms of class-wise accuracy.

S. No. Class Multiple augmentation model (251 epochs) accuracy (%) Multiple augmentation model (502 epochs) accuracy (%) Average of MAM-I and MAM-II Inception_ResNet_V2 model accuracy (%)
1 1121 61.4 68.53 64.96 99
2 1509 85.33 88.2 86.8 99.4
3 1637 59.07 77.47 68.27 98.7
4 1718 38.07 64.47 51.27 98.9
5 1728 64.73 55.6 60.2 99.4
6 BAS-370 89.87 96.13 93 100
7 CSR-30 87.53 65.67 76.6 99.6
8 DHBT-3 83.73 95.4 89.6 99.98
9 PB-1 78.27 83.47 80.87 99.1
10 PB-6 69.87 59.53 64.7 99.6
11 Unknown 99.51 99.27 99.4 99.9

The comparative performance of iRSVPred models with external validation dataset of original and augmented images.




The graphical representation of the average accuracy of iRSVPred MAMs with respect to prediction accuracy of InceptionResNetV2 model (iRSVPred 2.0) for external validation dataset.

The graphical representation of the average accuracy of iRSVPred MAMs with respect to prediction accuracy of InceptionResNetV2 model (iRSVPred 2.0) for external validation datasets of original and augmented images.



×