Real time feedback :

Abstract

Uganda is one of the countries in Africa which has an agro-based economy that feeds the population and other foreign countries through exports. Proper crop disease diagnosing tools and information about agricultural practices is key to the growth and development in the agricultural sector. In order to help farmers, get real time answers on queries about agronomic practices and also get accurate diagnosis about their crops, we have built an android application using a dataset collected from farmers in eastern Uganda and some parts of northern Uganda. This application can answer queries on land preparation, seed selection, planting, crop maintenance, harvesting and good storage practices on any of the three major crops in Uganda that’s cassava, maize and beans. It can also help farmers diagnose cassava crop diseases with an accuracy of 94%. This application is active and available 24/7. The Q&A component was built using machine learning and natural language processing. In building the model, an S. BERT transformer model was used and even with a Bleu score accuracy of 60%, the system provides useful and clear feedback to agronomic queries in any of the three crops mentioned above. With such an application, farmers can progress towards easier information about farming related practices, crop disease diagnosis and hence a better agricultural output.