An Automated Framework Based on UAV and Convolutional Neural Networks for the Detection of Xylella fastidiosa Disease in Olive Trees

This chapter presents an integrated framework detecting trees affected by Xylella fastidiosa disease using UAVs and convolutional neural networks. This framework involves drones that can collect an adequate number of olive leaf images to detect the presence of disease symptoms. To this aim, a simulation environment has been developed within the Robot Operating System (ROS) framework, including Gazebo, Rviz, and MoveIt platforms. Then, a suitable algorithm is proposed. The inspection process involves a simulated drone driven by a workstation performing all the necessary computations. A communication protocol is agreed upon with four codes used by the drone to forward requests to the workstation, which will respond to the ROS topics explicitly created. The algorithm consists of the following steps. Firstly, the drone takes off and reaches the centre of the area to be monitored. Then, it identifies the position of the trees by processing the aerial images and autonomously plans the main trajectory to visit them all. Each time a tree is reached, its canopy is thoroughly inspected to collect an adequate number of images to detect the presence of Xylella disease by a post-processing analysis using neural networks. Several neural networks are trained to compare results and determine the best solution. The presented results show the effectiveness of the proposed approach.