Detection of Crop Pests and Diseases on Web and Mobile Devices using Deep Learning
Funded By: African Technology Policy Studies Network (ATPS) as part of the Artificial Intelligence
for Agriculture and Food Systems (AI4AFS) Innovation Research Network
Focus Area: Pest and Disease detection
Institutions
1. University of Energy and Natural Resources, Sunyani, Ghana.
2. Ghana Developing Communities Association-NGO.
3. DIGILECT SYSTEM - SME.
Akuafo Adanfo Mobile and Web App
The University of Energy and Natural Resources has sought to develop skills needed to solve natural resource problems
through interdisciplinary research in Science, Engineering, and Agriculture. The communities in the University’s
catchment area and Ghana as a whole are highly dependent on agriculture for their livelihood. They feed their families
with food from maize, cassava, and tomatoes and sell some for financial gains. Due to the scarcity of land, the farmers
cultivate cashew plants on small pieces of land to support their families financially. However, these crops are
infested by pests and diseases every farming season, resulting in crop losses, hunger, malnutrition, low income, and
poverty. To contribute to the improvement of farmers’ livelihood, we have developed a deep learning (DL)-based
mobile (Android) and web applications to efficiently detect cassava, maize, tomatoes, and cashew pest/diseases. The DL
models were trained to reduce carbon footprint with the ability to recognize pests and diseases of these crops.
The models are embedded in a mobile app and deployed on mobile phones for use by farmers.
When the app is installed on a mobile device, the user may capture/scan a plant with the phone’s camera or pick an image
from and the phone gallery. The recognition is shown instantaneously in addition to the certainty (probability value)
of the prediction.
For further verification of high uncertain outputs (identified pest/disease with high uncertainty), the user is
alerted of the necessity to seek clarification from an expert/consultant serving as a man-in-the-loop for the system.
Due to high illiteracy rates in the farming communities, our AI system is simple and user-friendly with aa accompanying
prediction-to-voice facility that communicates the results and recommendations in English and the popular local
language “Twi” to the farmer. This is to also facilitate easy usage by the visually impaired. To achieve responsible
AI, the models are designed to be privacy preserving and robust by means of frequent security updates. Due to
low internet penetration in the catchment areas, the mobile apps do not need internet connectivity to function.
For community involvement, farmers in the Bono, Bono East, Savannah, and the Ahafo Regions were involved through
stakeholder meetings. Over 7,000 farmers were also trained to use the apps. E-kiosks werenalso setup in five communities
to provide services to illiterate farmers, the visually impaired, and those without phones and internet connectivity
at no fee. As part of gender equality and inclusion, the e-kiosks were managed by women and the disabled.
Next Steps
Support is needed to scale up the apps as follows
1. Improve the visibility of the appsthrough the help of the Ministry of Agriculture, district assemblies, the media,
our NGOs and SMEs.
2. Funding is needed to add more functionality such as farmer location, number of users, weather data, etc. to the app.
3. More data need to be collected for further training of the AI models to improve the performance of the apps.
4. Funding is needed to maintain e-kiosks attendants in all the communities to give further training to the farmers.
4. Farmers have emphasized the need for additional crops to be added to the app.
Appreciation
Special thanks to ATPS and UENR for funding and hosting this project respectively.