ColCo

Cocoa cultivation, essential for the chocolate industry and small farming communities, plays a significant role in Colombia's economy. According to the International Cocoa Organization (ICCO), in 2020 Colombia ranked tenth in the world in terms of annual cocoa production. Colombia harvests cacao year-round, and the cacao production year runs from October to September. Most producers farm an average of three hectares of cacao.

Cacao diseases cause a major threat to production which impacts yields and farmer livelihoods. These diseases develop due to environmental conditions, requiring farmers to effectively manage and sustain healthy crops to ensure quality. Below, you will find the most common diseases affecting cacao pods.

Colombian cacao production has been facing significant challenges. With 16.7 million hectares, these suitable lands play a vital role in the country's cacao cultivation. However in recent years climate-related factors and disease outbreaks have severely impacted productivity. This section highlights visual data representing the impact of these issues.

In 2023, cacao production in Colombia dropped to 59, 831 MT (metric tons), which is a 13% decrease from 2021. Due to constant climate changes and disease outbreaks such as frosty pod rot, this has led to a decline in cocoa production. The excessive rainfalls in Colombia has increased the spread of frosty pod rot, leading to poor soil quality and growing conditions.

A study by Carlos E. González-Orozco and Allende Pesca analyzed nearly 5,000 cacao farms, identifying four key regions influenced by climate. High humidity and specific temperature ranges, especially in the Pacific region of Nariño, creates conditions for fungal diseases such as frosty pod rot, affects cacao growth stages, and increases moisture retention in the soil and air.

Regions in northern Colombia receive the highest solar radiation (Fig. 2A), while southern Andean areas face significant soil temperature fluctuations (Fig. 2B). The hyper-humid Pacific regions, including Chocó and Valle del Cauca (Fig. 3 sub-region 4c), and areas bordering Chocó-Darién (Fig 3sub-region 2c), experience high humidity, increasing the risk of frosty pod rot. The central province (Fig 3 sub-region 2b) is a transitional zone, shifting from hot and dry to wet and humid, impacting cacao growth.

This project focuses on Granja Luker farm, Palestina. It integrates machine learning, geospatial analytics, and sensor networks to transform disease management and support sustainable farming practices. A deep learning model is used to identify diseases from images for cocoa pods, while geolocation data helps uncover reginal patterns influenced by weather conditions. Additionally sensor nodes collect soil an environmental data, providing insight that will improve the accuracy of our disease predictions

To detect visible symptoms of cocoa pod diseases, we are currently using deep learning models such as convolutional neural network (CNN) and Random Forest Classifier (RFC). The key steps involved are:

• Dataset Overview: The  Enfermedades Cacao YOLOv4  dataset, obtained from Kaggle, features images of cocoa pods, both healthy and unhealthy. The diseases included in these dataset are Fito and Monilia. This dataset is utilized to train our Machine learning model for accurate disease detection and classification.
• Image Preprocessing: Techniques such as contrast enhancement, noise reduction and augmentation (rotation, scaling, cropping) are applied to improve the dataset quality

• Feature Extraction & Classification: When using CNN, it analyzes images to detect features such as color irregularities, texture changes, and lesions. This catches early-stage detection of diseases such as discoloration or fungal growth. A RFC is used to make the final disease classification.
• Geotagging: Data is geotagged and integrated with GIS tools to create interactive maps highlighting areas of high disease risk. The images were captured by our team members in Colombia. Additionally, when a user takes and submits a picture, their location is recorded, helping the model understand the context of the disease based on climate and environmental conditions.

GPS modules and hardware sensors are deployed in cocoa farms to collect critical environmental data.

• Sensor Deployment: A network of sensors per acre is strategically placed to monitor key variables such as temperature, humidity, and soil moisture.
• Data Collection: Sensors continuously relay data to a centralized system, enabling real-time monitoring of environmental conditions that contribute to disease risk.
• Predictive Modeling: Using the collected data, machine learning models analyze patterns to predict potential disease hotspots. For instance, high humidity combined with low soil pH is strongly correlated with black pod disease.

We plan to integrate a map that highlights critical zones using a two-tier color-coded system to represent varying levels of weather-related risk. Red will indicate high-risk areas, yellow will signify low-risk zones, and other areas will represent safe regions with no immediate danger from weather conditions. This visual representation aims to provide users with clear and intuitive insights into the severity of weather impacts in different areas, enabling better decision-making and preparedness.

Our next steps for this project focus on enhancing the accuracy of our machine learning models by integrating advanced image processing techniques to enhance our results. With our team traveling to Colombia, we were able to gather invaluable data, including geotagged coordinates and images, which will help refine the model’s precision. Furthermore, setting up the sensors and collecting data would further improve the predictions made by the models.

Meet the team! We are students from California State University, Northridge, working together to use the latest technologies to meet growing demands and solve real-world challenges.