Abstract

Underwater object detection is critical for oceanic research and industrialsafety inspections. However, the complex optical environment and the limitedresources of underwater equipment pose significant challenges to achieving highaccuracy and low power consumption. To address these issues, we propose SpikingUnderwater YOLO (SU-YOLO), a Spiking Neural Network (SNN) model. Leveraging thelightweight and energy-efficient properties of SNNs, SU-YOLO incorporates anovel spike-based underwater image denoising method based solely on integeraddition, which enhances the quality of feature maps with minimal computationaloverhead. In addition, we introduce Separated Batch Normalization (SeBN), atechnique that normalizes feature maps independently across multiple time stepsand is optimized for integration with residual structures to capture thetemporal dynamics of SNNs more effectively. The redesigned spiking residualblocks integrate the Cross Stage Partial Network (CSPNet) with the YOLOarchitecture to mitigate spike degradation and enhance the model's featureextraction capabilities. Experimental results on URPC2019 underwater datasetdemonstrate that SU-YOLO achieves mAP of 78.8% with 6.97M parameters and anenergy consumption of 2.98 mJ, surpassing mainstream SNN models in bothdetection accuracy and computational efficiency. These results underscore thepotential of SNNs for engineering applications. The code is available inhttps://github.com/lwxfight/snn-underwater.