BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//events//Events//EN
CALSCALE:GREGORIAN
X-WR-CALNAME:Nov 5 - Visual AI Agriculture Meetup
REFRESH-INTERVAL;VALUE=DURATION:PT1H
X-PUBLISHED-TTL:PT1H
BEGIN:VEVENT
UID:event-g5w98bd6@ontown.app
DTSTAMP:20261008T195825Z
DTSTART:20261105T170000Z
SUMMARY:Nov 5 - Visual AI Agriculture Meetup
LOCATION:San Francisco\, United States
DESCRIPTION:AI Professionals San Francisco Machine Learning & Agentic AI\nJ
 oin our virtual meetup to hear talks from experts on cutting-edge topics a
 t the intersection of agriculture and AI.\n\nDate\, Time and Location\n\nN
 ov 05\, 2026\n9:00 AM - 11:00 AM PST\nOnline. Register for the Zoom! (http
 s://voxel51.com/events/visual-ai-agriculture-meetup-november-5-2026)\n\nTa
 lks will include:\n\nU-Net Framework for Micro-Scale Surface Damage Segmen
 tation in High-Resolution Soybean Seed Imagery\n\nAccurate detection of su
 rface-level seed damage is critical for soybean seed quality assurance\, y
 et automated micro-scale damage detection in high-resolution imagery remai
 ns an open challenge due to extreme spatial variability in damage scale\, 
 severe class imbalance across damage types\, and the computational demands
  of processing ultra-high resolution agricultural imagery at scale. This r
 esearch addresses the semantic segmentation of fine-grained soybean seed s
 urface defects\, such as wrinkles\, dark spots\, and general surface damag
 e\, in 6048 × 4024 pixel images where target damages can be as small as 1
 8 × 18 pixels\, using an augmented dataset of 77\,000 individual seed ima
 ges.\n\nTo overcome the difficulties of micro-scale detection\, spatial sp
 arsity\, and class confusion\, we propose a two-stage training and dual mo
 del inference framework built on an optimized U-Net architecture with a Re
 sNet34 encoder. In the first stage\, a damage specialist model is trained 
 using weighted loss functions and a class-balanced approach that prioritiz
 es damage classes.\n\nIn the second stage\, transfer learning is applied t
 o initialize a healthy seed specialist model from Stage 1 weights\, with r
 ebalanced class weights and tile probabilities that identify healthy seeds
 . During inference\, a confidence-gated damage filter suppresses low-confi
 dence predictions\, and healthy seed labels are assigned only when the spe
 cialist model's confidence exceeds that of the damage model.\n\nThe two sp
 ecialist models achieve validation accuracies of 94% and 98.53%\, respecti
 vely\, and the combined inference system successfully detects and localize
 s all three damage categories across unseen test images under conditions o
 f extreme spatial variability and class imbalance. Qualitative evaluation 
 confirms close alignment of predicted boundaries with ground truth annotat
 ions across all damage types\, including dark spots.\n\nThese results demo
 nstrate that confidence-gated dual model inference can reduce class confus
 ion in imbalanced micro-scale segmentation\, advancing the feasibility of 
 fine-grained automated seed quality inspection at the scale.\n\nAbout the 
 Speaker\n\nSaurav Upadhyaya (https://www.linkedin.com/in/sauravupadhyaya/)
  is an AI/ML researcher with a Master's degree in Computer Science\, whose
  work spans agricultural AI\, conversational systems\, and public health\,
  consistently translating advanced technology into tangible real-world imp
 act.\n\nFrom camera to drone: wildfire detection and integration at the ed
 ge\n\nWildfires are getting larger and more expensive every year. But caug
 ht early\, a fire is a small job for a small team\; caught late\, nothing 
 stops it. The cameras to catch them are already installed\, so the work is
  not necessary more sensors but making the ones we have act.\n\nThis talk 
 walks the whole chain: a small vision model running on the camera that fla
 gs smoke\, an integration layer that scores the alert against terrain\, we
 ather and history to decide whether it matters\, and a drone dispatched to
  confirm before anyone commits a crew. We will cover what has to run at th
 e edge and why\, how small the model can get before it stops seeing smoke\
 , and what it takes to trust a detection nobody has looked at yet.\n\nAbou
 t the Speaker\n\nMaxime Carriere (https://voxel51.com/events/www.linkedin.
 com/in/maxime-carriere-871756161) is co-founder of Kernwerk in Berlin\, wh
 ere he works on compressing AI models to run on cheap embedded hardware.\n
 \n5\,000 Flights to Answer One Question: Can a Drone Sample as Well as a H
 uman?\n\nEnvironmental testing still begins with a person walking onto a s
 ite with a shovel. We built a system that turns a plain-language brief int
 o a sampling campaign\, flies it\, and returns soil\, water and vegetation
  samples with GPS\, timestamps and chain of custody intact: 5\,000 flights
 \, roughly 9\,000 miles and 570 flight hours over a 185-acre site in Roata
 n\, much of it under canopy where GPS degrades and clearances are tight.\n
 \nI will walk through the vision and planning stack\, including multispect
 ral and thermal canopy sensing\, terrain and land-cover mapping\, and the 
 constraint solver that clears a route against airspace and battery budget 
 before anything leaves the ground. Then the harder problem: evidentiary co
 mparability.\n\nA drone-collected result means nothing to a regulator unle
 ss it agrees with a hand-collected split under the same method\, so I will
  show the dual-collection protocol we designed to test exactly that…
URL:https://ontown.app/e/g5w98bd6-nov-5-visual-ai-agriculture-meetup/
END:VEVENT
END:VCALENDAR