A spatiotemporal attention-based neural network to evaluate the route risk for unmanned aerial vehicles

摘要

Route risk evaluation is crucial for planning safe routes when unmanned aerial vehicles (UAVs) perform missions in hostile environments. The purpose of route risk evaluation is to fuse the intents, capabilities, and opportunities of the enemy to predict the damage to UAV in the future. Opportunities depend on the current as well as historical situation of the battlefield and are difficult to estimate effectively for the existing risk evaluation models. We propose a novel spatiotemporal attention-based evaluation network (STAEN) to automatically evaluate the route risk. In particular, the spatiotemporal attention values provided by the STAEN can reflect the opportunities of the enemy to threaten the UAV, which helps to understand the spatiotemporal evolutions of the situations on different routes. In addition, the network can automatically focus on the key route segments and defense subsystems in different evolution stages, to evaluate the route risk more accurately. The validity and interpretability of the evaluation network are verified by simulation experiments.

关键词