Vision based Rice Genotype Identification using Machine Learning Techniques

Author: Chandrika G., Juliet Hepziba S., Arumugam Pillai M., Kavitha Pushpam A. and Vijayalakshmi R.

Download PDF

Abstract

Images of ten rice genotypes consisting of Chithrakar, Kuliyadichan, TRY 4, ADT 53, ACK 14090, ACK 15004, ADT 45, ASD 19, IR 64, Bhavani were selected and subjected to various image processing machine learning tools for vision based classification. Using Grain analyzer, morphological features of rice seed such as length, breadth, thickness, geometric mean diameter, sphericity, surface area, weight and area were estimated. The mean data were subjected to PCA analysis in STAR software to reduce the dimensionality. The trait such as length, surface area, geometric mean diameter, area and weight contributed significant variation as they possessed positive values in both PC1 and PC2. The predicted variables of PCA and visual, textural, spectral characteristics of rice seed image obtained from Image Analyser (LEICA) were subjected to various image processing process. The processed images were fed into the machine tools viz., Partial Least Square Regression (PLS) and Support Vector Machine (SVM) for vision based classification. Totally 2000 images were taken and 80 percent images were used for training the model, 20 percent images were kept for testing the model. By comparing accuracy, precision, recall and F1 score of both the methods, PLS gives better performance than the SVM classifier. By using these classifiers, genotypes could be identified based on morphological features, visual characteristics and textural characteristics, as the accuracy and prediction are reliable.

Keywords

Rice seeds, Principal Component Analysis, Machine learning tools, Partial Least Square Regression, Support Vector Machine classifier

Conclusion

Ten rice genotypes used in this study were classified efficiently by using machine learning techniques. By comparing accuracy, precision, recall and F1 score of both the methods, PLS gives better performance than the SVM classifier. Traits such as length, surface area, geometric mean diameter, area and weight were the highly contributing trait for presence of variation among the genotypes. By using these classifiers, genotypes could be identified based on morphological features, visual characteristics and textural characteristics, as the accuracy and prediction are reliable. Thus, Machine learning tools is an alternative approach to identify the varietal purity based on seed morphological features and images.

References

-

How to cite this article

Chandrika G., Juliet Hepziba S., Arumugam Pillai M., Kavitha Pushpam A. and Vijayalakshmi R. (2023). Vision based Rice Genotype Identification using Machine Learning Techniques. Biological Forum – An International Journal, 15(8a): 285-291.