Hyperspectral and lidar-based maize tar spot detection and severity prediction platform

Overview

This research platform leverages high-throughput remote sensing and proximal imaging technologies to predict the onset and quantify the severity of maize tar spot, a destructive fungal disease. By combining drone-based hyperspectral VNIR+SWIR and LiDAR measurements with cell phone-based proximal imaging, the platform enables rapid, accurate disease assessment across large diversity panels of maize genotypes. The approach supports the identification, validation, and advancement of tar spot-resistant maize breeding lines, addressing a critical need for the seed industry and growers.

Technical specifications

• Multi-modal remote sensing: Co-aligned hyperspectral VNIR+SWIR and LiDAR data capture both biochemical changes (spectral reflectance) and structural changes (stalk integrity) in infected maize canopies.
• Drone-based data collection: Aerial platforms enable scalable field-level phenotyping across hundreds of genotypes.
• Proximal imaging: Cell phone-grade RGB imagery is used to train machine learning and deep learning models to automatically identify, annotate, and measure tar spot lesions, enabling percent severity assessment from farmer- or scout-submitted images.
• Machine learning pipelines: Models are trained to distinguish tar spot-specific spectral signatures from general photosynthetic decline (e.g., chlorophyll-driven NDVI changes), improving diagnostic specificity.
• Genetic mapping integration: Accurate severity measurements feed directly into genetic mapping studies to identify resistance loci and validate breeding lines.
• Introgression and doubled haploid pipelines: Resistance alleles are being introgressed into elite varieties, and doubled haploid populations are being developed for further genetic validation.

Technology readiness level

The platform is currently at TRL 4–5. Year-one validation with 500 genotypes (RGB and multispectral imagery) demonstrated a mild relationship with tar spot severity, primarily driven by general photosynthetic decline rather than tar spot-specific signatures. Year-two data collection (co-aligned hyperspectral VNIR+SWIR and LiDAR) is complete, with processing and analysis underway. The proximal cell phone imaging dataset is being annotated, with machine learning-based lesion identification and severity quantification in development. A predictive cell phone app for farmer or scout use is envisioned as a future deployment requiring additional collaborators. Introgression of tar spot resistance into elite varieties and creation of doubled haploid populations are already funded and in progress.