Arham Sethi
Built WERM — a novel framework for predicting weed emergence risk from field imagery and thermal data. Published at IEEE CIBCB 2026 in Greece, one of the largest computational biology conferences in the world.
Where Arham Started
His Background
- • 11th grader at The Shishukunj International School, Indore
- • Self-taught AI/ML since age 13, sparked by first encounter with ChatGPT
- • Already researching robust CNNs for real-world image conditions
- • Presented at Young Scientist India national science fair
- • Strong Python and deep learning experience
- • Interested in using AI for satellite/drone imagery analysis
His Goals
- • Publish a peer-reviewed research paper
- • Win a science fair competition
- • Apply AI to solve real agricultural problems
- • Build a competitive profile for college admissions
- • Reach intellectual depth in AI research
The Problem He Wanted to Solve
Weed emergence is a persistent agricultural challenge, with increasing herbicide resistance demanding a shift from reactive to predictive management. Existing systems can detect weeds post-emergence but no system could estimate WHERE emergence risk is highest BEFORE weeds are visible. Arham wanted to build a framework that predicts pre-emergence weed risk using field imagery and thermal data — helping farmers act before the problem starts.
The Research
Arham developed WERM (Weed Emergence Risk Mapping) — a novel framework that estimates pre-emergence weed risk by fusing four drivers: soil moisture, residue cover, structural disturbance, and growing degree days. The system uses ResNet-18 for moisture and residue estimation from RGB field images, GLCM texture analysis, and NASA POWER API thermal data, all combined through a log-space multiplicative fusion model.
Weed Emergence Risk Mapping (WERM): A Log-Space Multiplicative Framework for Pre-Emergence Risk Estimation from Field Imagery and Thermal Accumulation
No existing system could predict where weed emergence risk is highest before weeds are visible
Log-space weighted geometric aggregation of moisture (ResNet-18), residue, texture (GLCM), and thermal accumulation (NASA POWER API)
333 RGB field images, 61-image geotagged grid, 30 ground-truth soil samples validated by farmers with 20-30 years experience
94.5% texture dominance in spatial variance, ground-truth correlation ρ = 0.879, robust to perturbations (ρ ≥ 0.990)
Real-World AI for Agriculture
What makes Arham's research exceptional is its real-world validation. This isn't a theoretical exercise — he worked with actual farmers with 20-30 years of experience to validate his framework against ground-truth soil samples. The WERM framework uses a multiplicative fusion model rather than additive, meaning a deficit in any single driver (moisture, temperature, etc.) can suppress germination prediction regardless of other factors — mirroring how biology actually works. High-risk zones identified by WERM corresponded to areas of heaviest weed emergence in the preceding 2-3 seasons.
Field Images
Texture Dominance
Ground Truth ρ
Robustness ρ
The Outcome
Accepted at IEEE Conference on Computational Intelligence in Bioinformatics and Computational Biology
IEEE CIBCB 2026, Greece
2026
Weed Emergence Risk Mapping (WERM)
Accepted for Presentation & Publication
Self-taught AI enthusiast researching robust CNNs, presented at Young Scientist India but no peer-reviewed publication
Published a novel agricultural AI framework at IEEE CIBCB in Greece — validated with real farmers and ground-truth data
The Bigger Picture
Years old — Arham built a validated agricultural AI framework that real farmers can use, published at a major IEEE conference
IEEE CIBCB 2026 — presenting his research on the international stage alongside PhD researchers
No existing system predicted pre-emergence weed risk spatially — Arham's WERM framework is the first to do it
Ready to Start Your Research Journey?
Join the YRI Fellowship and work with expert mentors to conduct original research, get published, and build a profile that stands out.
Apply Now