Deep Learning Approach to Cassava Disease Detection Using EfficientNetB0 and Image Augmentation

Cassava, a vital crop in the Philippines and other tropical regions, is highly susceptible to various diseases that drastically reduce its yield. Traditional inspection methods for detecting these diseases are manual, time-consuming, expensive, and prone to inaccuracies. While recent advances enable improved detection, many approaches focus primarily on leaves and stems, overlooking tubers—one of the most critical parts of the plant. Since tubers are the harvested portion of the cassava and a direct source of food and income, early disease detection in this part is crucial for preventing severe yield losses. Furthermore, symptoms often manifest in the tubers before becoming visible in other parts, making their monitoring essential for timely intervention. To address these challenges and improve accuracy, we employed EfficientNetB0 and data augmentation techniques to enhance disease detection across multiple parts of the cassava plant. The developed system integrates a Raspberry Pi 4B with a camera module LCD screen enclosed in a 3D-printed casing for ease of use by farmers, and this showed detection accuracies of 94% for leaves, 90% for stems, and 92% for tubers. The system’s reliability was validated with <i>p</i>-values at a 0.05 significance level. By reducing the need for expensive manual inspections, the system offers a robust solution for early disease detection, particularly in the tubers, to mitigate yield losses. Its proven accuracy and practical design support better disease management practices, thereby improving crop health while enhancing food security and supporting the livelihoods of cassava farmers.