Exploring the Convolutional Neural Networks Architectures for Quadcopter Crop Monitoring

Utilizing Unmanned Aerial Vehicles (UAVs) equipped with Convolutional Neural Networks (CNNs) has become pivotal in revolutionizing crop monitoring. This synergy offers unparalleled advantages in precision agriculture by enabling real-time, high-resolution data collection and analysis. UAVs facilitate rapid and cost-effective aerial surveillance, capturing intricate details of crop health, pest infestations, and growth patterns. CNNs, a subset of deep learning, enhance image processing, allowing for automated and accurate identification of anomalies. This integration optimizes resource management, reduces environmental impact, and enhances overall crop yield predictions. The combination of UAVs and CNNs is a transformative approach, empowering farmers with actionable insights for informed decision-making in modern agriculture. The paper is part of the current trend of UAVs use in agriculture, with the main purpose of evaluating the results obtained for crop monitoring by training three different architectures of CNNs, namely the simple sequence, SqueezeNet and GoogleNet. The research was carried out on a number of four identified crops prevalent in the studied area and allows, based on the results obtained, to choose the most suitable variant of CNN taking into account training precision, classification accuracy, initial costs and in-site use.