Publication:

Multimodal CNN–LSTM Framework for Real-Time Maize Disease Detection

Maize diseases present a major challenge to agricultural productivity and food security, particularly in low-resource settings in sub-Saharan Africa. Timely detection plays an important role in reducing yield losses and enabling effective farm management. This research introduces and validates a multimodal machine learning–based system for real-time maize disease detection in Bomet County, Kenya. The system integrates maize leaf image data, environmental sensor data, and farmer-reported observations to develop a hybrid Convolutional Neural Network–Long Short-Term Memory (CNN–LSTM) model designed to automatically identify and categorize maize diseases. A mixed-methods research design was adopted, combining machine learning experiments with surveys and interviews involving farmers and agricultural officers. The findings revealed that Maize Lethal Necrosis (MLN) was the most prevalent disease (41%), followed by Gray Leaf Spot (33%) and Northern Leaf Blight (26%). Environmental variables such as humidity and temperature demonstrated strong associations with disease occurrence. The proposed multimodal CNN–LSTM framework integrates maize leaf images, environmental sensor data, and farmer observations, achieving an accuracy of 94.2%, which outperforms conventional image-only CNN models (87.5%) and environmental-data based LSTM models (81.3%). Additionally, 78% of farmers reported faster disease diagnosis using the developed system. The findings demonstrate that the proposed system supports real-time maize disease detection through an edge-enabled architecture, enabling deployment on mobile devices and facilitating practical intelligent system integration in agricultural environments.