Event-Based Classification of Defects in Civil Infrastructures with Artificial and Spiking Neural Networks

Small Multirotor Autonomous Vehicles (MAVs) can be used to inspect civil infrastructure at height, improving safety and cost savings. However, there are challenges to be addressed, such as accurate visual inspection in high-contrast lighting and power efficiency for longer deployment times. Event cameras and Spiking Neural Networks (SNNs) can help solve these challenges, as event cameras are more robust to varying lighting conditions, and SNNs promise to be more power efficient on neuromorphic hardware. This work presents an initial investigation of the benefits of combining event cameras and SNNs for the onboard and real-time classification of civil structural defects. Results showed that event cameras allow higher defect classification accuracy than image-based methods under dynamic lighting conditions. Moreover, SNNs deployed into neuromorphic boards are 65-135 times more energy efficient than Artificial Neural Networks (ANNs) deployed into traditional hardware accelerators. This approach shows promise for reliable long-lasting drone-based visual inspections.