AI-Guided Drone-GPS Detection of Fall Armyworm Pest and Symptoms in Maize Farm

Abstract

The invasion of the fall armyworm into African farms has posed a significant threat to maize production, resulting in huge economic losses for farmers and threatening food security on the continent. Most farmers have tackled this menace using chemical methods, but excessive use of insecticides carries environmental and health risks. Early detection of this pest will facilitate control and limit pesticide use through precision spraying. This study exploits all the features of the pest and its symptoms to train a robust deep learning model that can identify the presence and symptoms of this pest on maize plants. Images of fall armyworm and its symptoms on maize farms were gathered across three locations in Nigeria. The YOLO (You Only Look Once) algorithm was used to train the deep learning model until consistent and fair performance metrics were obtained. From the developed model, a mobile app was created in Android Studio, while another version of the model was deployed on a laptop in a Python environment. A demonstration farm was set up for maize plantation. The mobile app showed excellent performance in detecting the pest and its symptoms. To further automate the detection process, a drone (DJI technology) was used to scan the farm in waypoint mode, and its recordings were linked to the model on the laptop with satisfactory detection and even counting of the number of detections obtained. GPS locations of the detection spots were collated in real time. This development enables capacity for on-the-spot precision spraying as a more economical approach to pest eradication.