Energy-efficient craters detection based on spiking neural network using digital elevation models

Craters are the primary topographic features on celestial bodies and play a vital role in deep space exploration missions. Recently, artificial neural networks (ANNs) have excelled in crater detection. However, when compared to ANNs, spike neural networks (SNNs) have the advantage of low power consumption, making them more suitable for resource-limited deep space environments. To the best of our knowledge, this is the first spiked-based crater detection model. Firstly, two conversion learning algorithms based on cluster neurons and vary-time windows neurons are proposed to address the non-differentiability problem of SNNs. Secondly, a calibration mechanism is introduced to correct the conversion loss between ANNs and SNNs. Thirdly, the experimental results demonstrate that the proposed scheme exhibits sufficient precision, while reducing the energy consumption cost by more than 10 times compared to the previous method. Additionally, this method is expected to offer new insights for segmentation tasks in resource-constrained environments.