Project Overview
Plant diseases can significantly reduce crop yields and hurt agricultural productivity. While inspecting plants visually is important for early diagnosis, having experts check every field manually is time-consuming and expensive.
Recent advances in computer vision make it possible to detect plant diseases using smartphone cameras. However, moving from clean laboratory photos to outdoor fields brings real challenges: changing sunlight, cluttered backgrounds with soil and weeds, overlapping leaves, and motion blur [1].
Tutorial Goal
This tutorial explores how computer vision can go beyond basic single-image classification. We will look at mobile data capture, CNN classification, YOLO object detection, pixel-level segmentation for calculating disease severity, DeepSORT video tracking, and practical mobile deployment trade-offs.
Why Use Video and Multi-View Analysis?
Most plant disease projects rely on a single photo of a leaf. In real-world conditions, that single image can easily fail:
- Overcoming glare and shadows: Bright sunlight or shadows can hide disease spots or make healthy leaves look sick. Video captures multiple angles, helping find clearer views.
- Combining predictions across frames: Using techniques like majority voting across multiple video frames helps filter out occasional blurry or incorrect predictions.
- Tracking individual leaves: Tracking algorithms like DeepSORT can follow a specific leaf as the camera moves, preventing the system from counting the same leaf multiple times [5].
What You Will Learn
In this tutorial, you will explore:
- Mobile Capture & Preprocessing: How lighting, blur, and color adjustments affect model inputs [2], [3].
- CNN Classification & Transfer Learning: How convolutional networks identify disease categories and why transfer learning is essential for smaller datasets.
- YOLO Object Detection: How bounding boxes locate multiple infected leaves in cluttered outdoor scenes [1].
- Segmentation & Severity Estimation: Using U-Net and DeepLabV3+ to measure the exact percentage of diseased leaf area [4].
- Video Analysis & Tracking: How DeepSORT tracks leaves over time and how majority voting combines multi-frame predictions.
- Plant Doctor Case Study: Examining real experimental results from Montagut Marques et al. (2025) and discussing why real-time video processing remains challenging [5].
- Deployment Trade-Offs: Comparing on-device mobile processing with cloud-based options.
Tutorial Roadmap
Mobile Capture & Preprocessing
Outdoor lighting factors, camera angles, color spaces, and noise reduction.
Plant Disease Classification
CNN architectures, transfer learning, and the limitations of single global labels.
Object Detection with YOLO
Locating multiple leaves with bounding boxes in natural foliage.
Segmentation & Severity Estimation
Measuring pixel-level damage with DeepLabV3+ and a Python severity calculation example.
Video & Multi-View Analysis
DeepSORT leaf tracking and majority voting across 10 video frames.
Case Study: Plant Doctor
Analyzing the hybrid YOLOv8 + DeepSORT + DeepLabV3+ system and its 50-minute processing time.
Deployment, Evaluation & Quiz
Edge vs. cloud architectures, research labs, an author interview, and a 5-question quiz.
Annotated References
5 annotated research papers with synopses, methodology, and direct links.
Tutorial Modules
Capture & Preprocessing
Explore how lighting, angles, and background clutter affect field photos and how preprocessing helps.
CNN Classification
Learn how convolutional networks and transfer learning categorize diseased leaves.
YOLO Object Detection
Find out how YOLO locates multiple leaves with bounding boxes in cluttered natural scenes.
Segmentation & Severity
Create pixel-level disease masks and calculate damage percentages with Python.
Video Analysis & Tracking
Examine how DeepSORT tracks leaves over time and how majority voting combines frame predictions.
Case Study: Plant Doctor
Review real results from Montagut Marques et al. (2025), including the 50-minute processing time.
Deployment & Quiz
Compare edge and cloud processing, read the expert interview, and test your knowledge with a quiz.
Annotated References
Review the 5 academic research papers supporting this tutorial with summaries and DOI links.
Module Narration
A 1–2 minute audio summary introducing the research scope and objectives of this tutorial.