Automated Model Hardening with Reinforcement Learning for On-Orbit Object Detectors with Convolutional Neural Networks

On-orbit object detection has received extensive attention in the field of artificial intelligence (AI) space research. Deep-learning-based object-detection algorithms are often computationally intensive and rely on high-performance devices to run. However, those usually lack space-qualified versions, they can hardly meet reliability requirement if directly deployed a satellite platform, due software errors induced by environment. In this paper, we evaluated impact space-environment-induced through large-scale fault injection tests. Aside from silent data corruption (SDC), propose an extended criterial SDC-0.1 better quantify effect transient faults algorithms. Considering that bit-flip error could cause severe result many cases, novel automated model hardening with reinforcement learning (AMHR) framework solve problem. AMHR searches for error-sensitive kernels convolutional neural network (CNN) trial deep deterministic policy gradient (DDPG) agent fine-grained modular-level redundancy increase tolerance CNN-based detectors. Compared other selective methods, achieved lowest rates various detectors tremendously improve mean average precision (mAP) SSD detector 28.8 presence multiple errors.