Rice Tungro Disease Detection Based on Light Absorption Assessment Using Machine Learning

Abstract:

This study aimed to determine the infrared and visible absorbance of rice (Oryza sativa) leaf samples infected with tungro at wavelengths 565nm, 645nm, 890nm, and 950nm. In this process, the samples were analyzed noninvasively without damage due to the absence of physical contact to monitor early signs of rice leaf infection. Specifically, the study aimed to (1) extract the absorbance value from the spectral response of the rice leaf; (2) train the data set and develop the classification model using machine learning; (3) design and develop the prototype; and (4) test, calibrate and evaluate the prototype. A graphical user interface was used to acquire data. The acquired data trained two pbkp_redictive models to identify rice tungro infection: artificial neural network and support vector machine algorithm. Of the two pbkp_redictive algorithms trained, ANN showed higher accuracy than SVM. Based on the three (3) evaluations, ANN has 90%, 90%, and 100% accuracy, while SVM has 80%, 80%, and 100%. The highest accuracy was attained using all four (4) LEDs instead of one. Nonetheless, the 950nm IR LED also showed excellent accuracy. Also, the average absorbance of rice leaf with tungro infection using all four (4) LEDs and 950nm IR LED were 0.258 and 0.267, respectively, while the average absorbance of rice leaf without tungro infection using all four (4) LEDs and 950nm IR LED were 0.210 and 0.223, respectively. The low absorbance average value of the rice leaf without tungro infection occurred due to the light being more reflected than absorbed.

Año de publicación:

2022

Keywords:

• Artificial Neural Network
• Support Vector Machine
• tungro infection
• remote sensing

Fuente:

scopus