Current Research
Smart Poultry Farming: Using ML and IoT to Predict Diseases in Chickens
I am leveraging machine learning and artificial intelligence to enhance poultry farming in resource-constrained areas. My research explores how audio and image data can be harnessed to develop innovative technologies that support poultry farmers, particularly in remote regions. By collaborating with poultry farmers, agricultural extension experts, and veterinary officers, I aim to create practical and scalable solutions that improve productivity, health monitoring, and overall farm management. Through this interdisciplinary approach, I seek to bridge technological gaps and empower small-scale farmers with accessible, data-driven tools to enhance their livelihoods.
Poultry-FL-IoT: Leveraging Federated Learning and IoT for Newcastle Disease Prediction in Poultry.
Poultry farming remains a vital source of income and food for rural households. It is especially common in areas with poor road infrastructure, which makes it challenging for farmers to access agricultural services when their poultry fall ill. This limitation increases the risk of disease transmission to humans and other animals. This project focuses on using chicken vocalizations to predict Newcastle disease in poultry.
Deep learning methods for poultry disease prediction using images
This study explores the use of deep-learning models to predict poultry diseases based on fecal images, focusing on their generalizability across different geographical regions. Specifically, we assess whether models trained on Tanzanian data can effectively classify poultry diseases in Malawi. We collected fecal image data from Malawi with the help of agricultural extension officers and tested deep-learning models—MobileNet, DenseNet, and ResNet—each designed for binary classification (healthy vs. sick) and multiclass classification. The evaluation revealed that binary classifiers, particularly MobileNet and DenseNet, outperformed multiclass models on unseen data, demonstrating greater robustness for cross-context disease prediction. In contrast, multiclass models performed well within their training context but showed a significant drop in accuracy when tested on Malawi data, likely due to variations in poultry breeds, disease manifestations, and environmental factors affecting fecal appearance. The study highlights the challenge of generalizing deep learning models across regions and underscores the importance of using representative datasets that capture diverse poultry breeds and environmental conditions. Overall, the research demonstrates the potential of deep learning in poultry disease detection while emphasizing the need for context-aware models and diverse training data to enhance adaptability.
Designing Context-Aware Smart Home Cameras for Marginalized Homes with Dominant Patriarchal Norms.
The proliferation of smart home security systems, such as cameras, foregrounds privacy issues between household members. These issues are dominant in cultures that promote patriarchal norms while jeopardizing women’s rights. Scholars have suggested that the introduction of smart home security systems in sub-Saharan Africa—a region with dominant patriarchal norms—exacerbates patriarchal norms and contributes to domestic violence against women. Despite this, the current design of smart home cameras has not been updated to account for privacy issues in patriarchal societies while supporting domestic security. This project aims to design, develop, and deploy context-aware smart home cameras aimed at minimizing the unintended consequences of these technologies in patriarchal societies.