Computer Vision Meetup: Better Farming Through Embedded AI скачать в хорошем качестве

Computer Vision Meetup: Better Farming Through Embedded AI

Blue River Technology, a subsidiary of John Deere, uses computer vision and deep learning to build intelligent machines that help farmers grow more food more efficiently. By enabling robots to tell the difference between crops and weeds and then only spraying the weeds, these machines are revolutionizing agriculture’s approach to chemical usage. By outfitting tractors with perception sensors and autonomous driving capabilities, we are freeing farmers from tedious jobs like tillage so they can spend more time doing higher-value tasks. In this session we will share how we solve machine vision problems using deep learning, and some of the specific challenges we’ve addressed along the way (such as dust interference and the visual similarities between weeds and crops). We will deep dive into the tech stack, including on-premise compute, image augmentations, 8-bit quantization trade-offs and tips and tricks to improve model performance. #computervision #ai #artificialintelligence #machinevision #machinelearning #datascience #opensource Contents of this video -- 0:00 – Introduction 0:46 – Sharing Slides and Starting the Talk 1:05 – Why Agriculture Needs AI: Global Challenges 3:07 – John Deere’s Evolving Strategy: From Bigger to Smarter 4:10 – Impact at Scale: 800,000 Connected Machines 4:41 – Blue River & Deere: AI Integration in Equipment 5:14 – Introducing the See & Spray System 6:12 – See & Spray: Herbicide Savings & Field-Level Intelligence 7:07 – See & Spray: Cameras, GPUs, and Smart Nozzles 8:44 – See & Spray: ML Architecture & Weed Detection 10:11 – See & Spray: Heatmap Visualization & Customer ROI 11:24 – Tackling Dust with Custom Augmentations 13:37 – Handling Dust: Augmentation, Predictors, and Edge Cases 14:47 – Predicting Low Light and Motion Blur 15:10 – Autonomous Tractor Overview 16:04 – How Autonomy Works: Geofencing & Mobile App 17:10 – Obstacle Detection & Teleoperation System 18:13 – Dust & Vision Challenges in Tillage 18:46 – CV System Details: Depth Maps & Classification Maps 20:19 – Roadmap: Autonomy in Corn & Soybeans by 2030 21:01 – Why Agriculture Affects Everyone 22:10 – Q&A Begins 22:41 – Q1: RGB vs Multispectral Cameras 23:03 – Q2: Infrared/Ultraviolet to Remove Dust 25:20 – Q3: Training Datasets for Weed vs Crop Classification 27:57 – Q4: Handling Latency and Spray Timing 29:34 – Why 36 Cameras? Occlusion Handling & Precision --