An Image Dataset of Taro Leaf Blight Disease Collected from the West African Sub-Region
Description
Taro Leaf Blight, caused by the pathogen Phytophthora colocasiae, manifests as distinct necrotic spots with white sporangia bands and orange droplets on the leaves. Thriving in temperatures between 25°C to 30°C, the disease spreads rapidly through rain splash and wind-blown spray. This blight not only reduces yields but also affects the income of smallholder farmers who depend heavily on taro cultivation. In recent years, taro yields in Africa have declined due to this disease, combined with other factors such as limited input utilization and cultivation on less fertile lands. Early detection of Taro Leaf Blight is essential for effective management and prevention. Technologies such as smartphone-based apps, handheld spectrometers, drone-mounted sensors, and biosensors are being explored to enable real-time disease identification. These methods empower farmers to implement timely measures, thus minimizing yield losses and preserving crop quality. However, challenges like the financial barriers of advanced technologies and the need for technical knowledge pose limitations, especially for smallholder farmers in developing countries. A critical part of combating TLB is building robust datasets for training deep-learning models for disease detection. To this end, a meticulous data collection effort was undertaken by teams in Nigeria and Ghana. This initiative focused on capturing images of taro plants at various stages of TLB infection—Taro Early Blight, Taro Mid Blight, Taro Late Blight, and Taro Healthy. The result of the initiative is the Taro Leaf Blight Disease Image Dataset. The dataset consists of 18,248 images. The breakdown of each class is as follows: Taro-Late: 1,270 Taro-Mid: 3,370 Taro-Early: 4,864 Taro-Healthy: 8,744 Taro-Not-Early: 4,640 (Combination of Taro-Late & Taro-Mid) Each image is a JPG (RGB) of size 500x500. Support for implementation of project activities was made possible by the Research Grant (109705-001/002) by the Responsible Artificial Intelligence Network for Climate Action in Africa (RAINCA) consortium made up of WASCAL, RUFORUM and AKADEMIYA2063 provided by IDRC.