Robotic and mechatronic systems for crop condition monitoring based on multispectral data

Abstract

This study investigated the use of robotic and mechatronic systems for monitoring the condition of agricultural crops based on multispectral data. A structural and functional diagram of the system was proposed, combining an unmanned aerial platform, a flight controller, electric actuators, inertial and navigation sensors, a stabilization system, optical and multispectral observation devices, and intelligent data analysis tools. To assess the condition of vegetation, spectral indices and semantic segmentation models were used, based on a U-shaped convolutional neural network, an enhanced third-version deep segmentation architecture, and a transformer-based segmentation model in its base configuration. An experimental study was conducted on real multispectral data, distinguishing between healthy and stressed vegetation, rust-infected vegetation, and soil. The best overall results were obtained for the third-version deep segmentation architecture with improvements: the average intersection and union coefficient was zero point four six one seven, and the harmonic measure of accuracy and completeness was zero point six one one eight. The practical significance of the obtained results lies in enabling the spatial localization of problem areas in crops and supporting the adoption of agronomic decisions in precision farming systems.