Aerial Landscape Recognition via Multi-Input Neural Network

Throughout the last decade, the advancements in the hardware allow use for wider applications of the unmanned aerial vehicles (UAV). UAVs feature significant advantages in autonomous aerial landscape mapping and recognition (ALR) over traditional methods due to their high level of operationality and mission repeatability, along with a simple alteration of e.g., on board remote sensors. ALR system based on convolutional neural networks is proposed. The system is designed with real-time capabilities. Data classification based on histogram and Gabor filter is explored on commercially available aerial images. The research roadmap designed to offload the dependency of the process on flight testing to improve the cost-efficiency of the development is proposed as well.