Neural Network Based Model Reference Adaptive Control of Quadrotor UAV for Precision Agriculture

This paper aims to design a neural network-based model reference intelligent adaptive control for quadrotor UAV and implement a machine learning approach to recognize the severity level of yellow wheat rust for precision agriculture. Yellow wheat rust is a fungal disease that can cause massive destruction in wheat production and quality. Obtaining accurate data from large-scale crops and detecting those diseases based on specific standards via visual inspection become labor-intensive, time-consuming, and sensitive to human error. Addressing these issues involves deploying a quadrotor for data acquisition and training a cutting-edge Convolutional Neural Network (CNN) for image analysis. Since existing control techniques are computationally intensive and show poor performance in tolerating unmatched uncertainty, the proposed controller is designed by using a nested control approach. In this control architecture, feedforward neural networks are trained to estimate position controller parameters online, whereas recurrent neural networks are trained to estimate the model and control the attitude of the quadrotor. Then, Xception CNN is trained by using a transfer learning approach. To verify controller performance, numerical simulations have been conducted in various scenarios. The results show that the designed controller has high tracking precision, robustness, and enhanced antidisturbance ability in nominal scenarios and the presence of matched and unmatched uncertainties, and the retrained model achieves an accuracy of 97.28%. Therefore, the suggested controller is a promising quadrotor control technique, and the retrained Xception model can be used for detecting the severity level of yellow wheat rust.