Keynote Talk: Dina Machuve - End-to-End Deep Learning for Chicken Disease Diagnosis

Описание: In this keynote, Dina Machuve explores how deep learning can support poultry farmers by diagnosing chicken diseases directly from images. The talk walks through CNNs, data collection workflows, annotation, modelling approaches, and real-world deployment. It also highlights early results and future directions, including multimodal models.

Timestamps:
00:00 – The Challenge
00:56 – The Motivation
01:23 – Convolutional Neural Networks (CNNs) Intro
03:33 – Convolutional Layer
05:55 – Convolutional Neural Network
09:09 – Activation Functions
10:15 – What the hidden layer “sees”
10:55 – What the network “assembles”
11:32 – Inception
11:54 – Application in the field
12:25 – The Workflow
12:50 – Step 1: Image Data Collection at Farms
16:41 – Step 2: Annotation
17:04 – Lab-labelled data: PCR Diagnostics
18:12 – Dataset: Chicken Droppings Images
18:37 – Step 3: Workflow
19:26 – Poultry Diseases Diagnostics
20:08 – Model Performance Results: Classification
22:47 – Modelling – Object Detection (YOLOv8)
23:20 – Modelling – Vision Transformers
24:37 – Deployment
29:45 – Next: Multimodal
31:08 – Our Progress
32:39 – Q&A

What You’ll Learn:
How CNNs and deep learning can be applied to agricultural health

How image data is collected and labelled for disease detection

Performance of models such as YOLOv8 and Vision Transformers

Real-world challenges from farm to deployment

Future multimodal approaches for more accurate disease diagnosis

#DeepLearningIndaba2025 #AIinAgriculture #PoultryFarming #DeepLearning #ComputerVision #YOLOv8 #VisionTransformers #AIforGood

Machuve explores how deep learning can support poultry farmers by diagnosing chicken diseases directly from images. The talk walks through CNNs, data collection workflows, annotation, modelling approaches, and real-world deployment. It also highlights early results and future directions, including multimodal models.
Timestamps:
00:00 – The Challenge
00:56 – The Motivation
01:23 – Convolutional Neural Networks (CNNs) Intro
03:33 – Convolutional Layer
05:55 – Convolutional Neural Network
09:09 – Activation Functions
10:15 – What the hidden layer “sees”
10:55 – What the network “assembles”
11:32 – Inception
11:54 – Application in the field
12:25 – The Workflow
12:50 – Step 1: Image Data Collection at Farms
16:41 – Step 2: Annotation
17:04 – Lab-labelled data: PCR Diagnostics
18:12 – Dataset: Chicken Droppings Images
18:37 – Step 3: Workflow
19:26 – Poultry Diseases Diagnostics
20:08 – Model Performance Results: Classification
22:47 – Modelling – Object Detection (YOLOv8)
23:20 – Modelling – Vision Transformers
24:37 – Deployment
29:45 – Next: Multimodal
31:08 – Our Progress
32:39 – Q&A

What You’ll Learn:
How CNNs and deep learning can be applied to agricultural health

How image data is collected and labelled for disease detection

Performance of models such as YOLOv8 and Vision Transformers

Real-world challenges from farm to deployment

Future multimodal approaches for more accurate disease diagnosis

#DeepLearningIndaba2025 #AIinAgriculture #PoultryFarming #DeepLearning #ComputerVision #YOLOv8 #VisionTransformers #AIforGood