Development of a Crop Disease Detection App Using Image Recognition for Farmers in Benue State

Agricultural productivity in Benue State, the North Central zone's food basket, is significantly constrained by undiagnosed crop diseases that cause substantial yield losses. This study develops a mobile application for crop disease detection using convolutional neural network (CNN) image recognition, targeted at smallholder farmers in Benue State. The application is built on a TensorFlow Lite model trained on a dataset of 8,000 labelled crop disease images sourced from open agricultural repositories. The Android front-end was developed in Kotlin, with an offline inference capability critical for areas with limited connectivity. The research methodology combines the agile development model with machine learning model evaluation cycles. Field testing involved 60 farmers across Makurdi, Otukpo, and Gboko local government areas, who photographed crop samples and compared app diagnoses to those of extension officers. The model achieved an overall accuracy of 84.7 percent across ten crop disease categories, including cassava mosaic and yam rot. The study notes that image quality in low-light farm conditions reduces detection accuracy and recommends image preprocessing enhancements in subsequent versions. The research advances the application of AI-powered software engineering in Nigerian smallholder agriculture, offering a cost-effective alternative to infrequent extension officer visits. Keywords: crop disease detection, image recognition, CNN, Benue State, mobile agriculture