Early detection of rice leaf blast using leaf reflectance spectroscopy and machine learning

Overview

Rice leaf blast (RLB) is a destructive fungal disease that threatens rice production worldwide. Early detection, before visible symptoms appear, is critical for timely intervention, yet conventional scouting relies on visual assessment that often misses the disease at its most treatable stages. This solution applies leaf reflectance spectroscopy together with machine learning to identify RLB infection during the asymptomatic, early, and mild phases of disease development. By detecting subtle biochemical shifts in rice leaves, the approach enables growers, breeders, and crop protection services to act sooner and reduce yield losses.

Technical specifications

Core approach:

• Measures leaf reflectance across the full 350–2500 nm range, capturing visible, near-infrared, and shortwave-infrared signatures
• Tracks biochemical changes linked to RLB progression, starting with chlorophyll content, followed by anthocyanin and leaf water content
• Extracts sensitive spectral features including single-band reflectance, pigment-sensitive spectral indices, and wavelet coefficients from the visible and near-infrared regions
• Uses an integrated machine learning and feature selection framework to classify healthy versus infected leaves

Validation to date:

• 2018 experiment: 26 paired infected and healthy leaf samples covering asymptomatic to early infection stages
• 2019 experiment: 45 paired infected and healthy leaf samples covering early to mild infection stages
• Most discriminating features identified: visible reflectance, pigment-sensitive spectral indices, and visible–NIR wavelet coefficients

Future validation:

• Apply the machine learning framework to time-series spectral-biochemical data from artificial inoculation trials
• Quantify classification accuracy across the asymptomatic-to-mild disease timeline
• Deliver novel phenotyping and machine learning tools for early RLB detection within a one-year study period

Technology readiness level

The technology has progressed through controlled experiments across two growing seasons with independent leaf samples, confirming that pigment-driven spectral features can distinguish infected from healthy leaves at early and mild stages. The next phase focuses on integrating machine learning classification, validating accuracy over the full asymptomatic-to-mild timeline, and packaging the outputs into deployable phenotyping tools. The team is actively seeking partners in leaf spectroscopy and machine learning, as well as funding for key personnel, to complete validation and advance toward practical deployment for rice disease management.