Key points are not available for this paper at this time.
Approximately 30% of natural disasters worldwide occur during nighttime or under low-light conditions. Nighttime search and rescue faces critical challenges due to the low visibility, temperature drop, and other complex environmental factors, where real-time efficiency directly impacts the survival probability of trapped individuals. Traditional methods such as carpet-style manual search and unimodal visible light or infrared nighttime search and rescue approaches suffer from limitations, including the time consumption, limited coverage, and poor perceptual capability. While the development of unmanned aerial vehicles (UAVs) offers a platform solution, their reliance on offline data processing leads to information latency, failing to meet the real-time requirement in complex nighttime environments. Moreover, ground-based complex models are often constrained by the limited computational resources available on UAVs for online processing. In this paper, to address these issues, we propose the spiking multimodal fusion detector (SMFD) framework, which is a low-light and infrared multimodal UAV remote sensing framework for low-power, nighttime rescue, inspired by the brain’s spiking mechanism. For the first time, a multimodal nighttime rescue dataset (MNRD) was constructed, covering multiple disaster scenarios, such as hypothermia, earthquake, fire, flood, landslide, and fall.A brain-inspired low-power online processing method based on a spiking neural network (SNN) was designed. Compared with advanced existing approaches, the proposed method achieves comparable detection accuracy and improves the energy efficiency of the deployed system by 5.11 × , thereby addressing the technical bottleneck of precise and efficient nighttime rescue operations.
Liu et al. (Mon,) studied this question.