Machine Learning-Based Presymptomatic Detection of Rice Sheath Blight Using Spectral Profiles

Area of Science:

• Agricultural science
• Plant pathology
• Spectroscopy

Background:

• Early detection of plant diseases is crucial for effective management.
• Rice sheath blight (ShB), caused by *Rhizoctonia solani*, significantly impacts rice production.
• Timely intervention before symptom onset enables targeted disease control strategies.

Purpose of the Study:

• To assess the efficacy of near-infrared (NIR) spectroscopy combined with machine learning for the early detection of rice sheath blight.
• To develop predictive models for identifying infected yet asymptomatic rice plants.

Main Methods:

• Collected NIR spectra from rice leaves one day post-inoculation with *R. solani*, before symptom appearance.
• Employed machine learning algorithms, including Support Vector Machine (SVM) and Random Forest, for classification.
• Utilized Sparse Partial Least Squares Discriminant Analysis (SPLS-DA) for result validation.

Main Results:

• An SVM-based model achieved 86.1% accuracy in distinguishing between mock-inoculated and inoculated plants.
• When comparing control, mock-inoculated, and inoculated plants, the best SVM model reached 73.3% accuracy.
• These models successfully identified infected plants based on spectral profiles prior to visible symptoms.

Conclusions:

• NIR spectroscopy and machine learning offer a promising approach for early diagnosis of rice sheath blight.
• This technique can potentially be developed into field tools for proactive disease management.
• Further validation through field trials is necessary to confirm practical applicability.