Cocoascan: AI-Powered Monitoring of Infectious Diseases impacting Cocoa Pods in Latin America. 

Cocoa trees in Latin America are susceptible to a variety of infectious diseases, and once a pod is infected it can infect an entire crop. 

Cocoascan is an innovative AI-powered tool designed to detect visual indicators of the presence of three of the top infectious diseases that affect cocoa pods – Witches Broom, Frosty Pod and Black Pod. 

With just a quick snap of your smartphone camera, Cocoascan identifies signals of infectious disease, and then offers resources for mitigation strategies for your crop.

How Does Cocoascan Work? 

Using cutting-edge AI technology, Cocoascan evaluates an image of a cocoa pod on your farm and provides valuable insights. Here’s how it works: 

1. Snap a Photo: Take a clear picture of an individual cocoa pod using your smartphone.
2. AI Analysis: Cocoascan’s sophisticated AI system processes the image, assessing key features such as coloration and shape.
3. Insights: Detects presence of visual indicators of 3 infectious diseases- Witches Broom, Frosty Pod and Black Pod
4. Resources: After receiving insights, easily access video content to learn more about pests and diseases in your region of the world. It’s a fast, seamless way to quickly detect if a pod might be infected with one of the top 3 diseases that can impact cocoa crops in Latin Ameria.

Why Is Monitoring Cocoa Pods for Disease so Important? 

Nearly 40% of the world’s cocoa crop is lost to infectious disease each year. In Latin America, much of that crop loss can be attributed to 3 infectious diseases: 

1. Witches Broom: a fungal disease caused by the pathogen Monilipohthora perniciosia, can cause abnormal growths on tree branches and significantly reduce pod production
2. Frosty Pod: a fungal disease caused by the pathogen Moniliophthora roreri, can cause pods to be highly misshaped, developing black lesions, and eventually overcome by a thick white covering of the fungus, ruining the beans inside.
3. Black Pod: a major disease primarily caused by the fungus-like (or, oomycete) pathogen Phytophthora palmivora, can create rapidly spreading black lesions that lead to pods rotting on the tree, ruining the beans inside. The same pathogen can also attack the trunk of the tree and eventually kill the tree!

If a disease is detected early, farmers may be able to take steps to quickly mitigate the spread of the disease to other pods on the tree, as well as help prevent the spread to other trees in the field. While these three diseases may produce similar symptoms on pods, the underlying biology of each pathogen is very different! If a farmer wants to use the proper control strategies to fight the disease, they need to understand what pathogen is causing the outbreak. 

Cocoascan is just one tool to help farmers in their overall Integrated Pest Management strategy to maximize pod production and yields. 

Behind the Scenes: How Cocoascan’s AI Works 

Cocoascan leverages an ensemble of AI models designed for detection and classification of infectious disease in a cocoa pod.  These models were trained using different types of neural networks, including approaches that specialize in identifying objects within images and distinguishing between different instances of those objects. 

Models were trained on over 3,000 images of cocoa pods collected in the field, and were then compared to nearly 800 validation images to analyze performance. Both models achieved a 95.5 F1 score after the training process.  

How Reliable Is It? 

Cocoascan’s AI models have been rigorously tested to ensure accuracy and reliability: 

1. Extensive Dataset: The AI was trained on 3,000 images of cocoa pods with labeled classifications verified by Integrated Pest Management and Plant Pathology experts from the Mars Cocoa Plant Science R&D Team.
2. High Accuracy Metrics: The model achieved a 95% accuracy-in-range score
3. Real-World Testing: The model was tested under varied lighting, angles, and environments within a particular region to ensure reliable performance in real-life conditions.