Autonomous Cognitive GPR Based on Edge Computing and Reinforcement Learning

Abstract

Autonomous cognitive ground penetrating radar (ACGPR), carried by drones or other robotic platforms, may perform robust and accurate subsurface object detection and recognition in varying environments based on real-time data processing and decision making. However, limited system computing resources and intelligence generating capability pose significant challenges for the operations of such systems. To address these challenges, in this paper we propose an ACGPR enabled by edge computing (EC) and reinforcement learning. Specifically, an edge computing based system architecture is presented to utilize edge resources for real-time intelligence generation. A reinforcement learning approach is developed as the decision-making model for the ACGPR to adaptively adjust its operational parameters. Simulation results show the accuracy and efficacy of the proposed ACGPR system. The framework also provides insight into the design of autonomous cognitive industrial Internet of things (IoT) supported by edge computing and machine learning.