Development of an Unmanned Aerial Vehicle incorporating a Convolutional Neural Network for Weed Detection

Abstract

This paper introduces the development of an Unmanned Aerial Vehicle (UAV) as an efficient asset that incorporates a high-performing pre trained architecture for weed classification and detection in real-time. The UAV includes a modular sprayer and embedded systems. Weeds provide a continuous threat to the cultivation of agricultural plants and reduce yield, as well as increase waste. To address the issue, an autonomous system and intelligent sprayer are described to spot weeds or undesirable plants and spray herbicides on desired places. The effectiveness of five various pre-trained architectures (AlexNet, GoogleNet, VGG19, DenseNet201, and ResNet50) for image classification and prediction are investigated, and the top performing model was determined. The best-performing trained model, ResNet 50, achieved a training accuracy of 98.57 percent and a validation accuracy of 100 percent at 25 epochs.