Fall Armyworm diagnosis

Abstract

The Fall Armyworm Diagnosis System is a mobile-first solution developed to support Ugandan farmers and agricultural officers in identifying, monitoring, and managing Fall Armyworm (FAW) infestations. The system leverages artificial intelligence, geolocation, weather integration, and expert-driven community support to address the persistent challenges of late-stage pest detection, improper pesticide use, and lack of localized outbreak data. It is optimized for rural environments where internet connectivity and technical literacy may be limited. The platform features a machine learning model based on TensorFlow Lite and MobileNet architecture, capable of analyzing maize leaf images and classifying them into four stages: Healthy, Eggs, Frass, or Larval Damage. A pre-screening module ensures that only valid maize leaf images are processed. Stage-specific treatment recommendations including herbicide names, dosages, and safe application instructions are generated automatically. The system also integrates GPS-based geolocation and reverse geocoding to map detections to districts, enabling the visualization of outbreak patterns through color-coded maps and supporting regional pest surveillance. In addition to detection, the system provides real-time weather data using the OpenWeatherMap API to help users optimize pesticide application timing. Users can view historical detections, generate downloadable PDF reports, and access a dynamic analytics dashboard that visualizes infestation trends, detection stage distributions, and regional outbreak comparisons. A community support module enables authenticated users to post crop-related concerns, attach images, and receive advice from verified agricultural experts. This report documents the entire system lifecycle from the definition of system requirements and specifications to architectural design, backend development using Flask and SQL, and mobile interface implementation using Flutter. It also covers inspection, testing, and validation procedures across multiple modules, including input processing, detection logic, map rendering, and performance under rural network conditions. The outcomes demonstrate that the system offers an effective, user-friendly, and intelligent platform for pest detection and management. By combining machine learning, geospatial data, weather forecasts, and expert knowledge, the Fall Armyworm Diagnosis System enhances farmers’ ability to respond to infestations early, apply treatment correctly, and contribute valuable data for national agricultural planning. Its scalable and modular design also offers potential for adaptation to other crop pests and regions.