SU-YOLO: Spiking Neural Network for Efficient Underwater Object Detection

TL;DR

SU-YOLO presents a fast, energy-efficient Spiking Neural Network for underwater object detection by integrating spike-based denoising, Separated Batch Normalization, and CSPNet-inspired residual blocks within a YOLO-based framework. The approach enables direct end-to-end training of a fully spiking detector, achieving mAP$_{0.5}$ of 0.788 on URPC2019 with 6.97M parameters and 2.98 mJ energy consumption, outperforming existing SNNs and matching lightweight ANNs. Key innovations include SpikeDenoiser (binary, integer-addition denoising on feature maps), SeBN (time-step aware normalization tailored for residuals), and SU-Block (CSPNet-based residual blocks to mitigate spike degradation). The results demonstrate strong detection performance with low energy usage on underwater datasets, suggesting significant potential for neuromorphic approaches in resource-constrained marine robotics; future work will implement hardware measurements and broaden applicability to other datasets.

Abstract

Underwater object detection is critical for oceanic research and industrial safety inspections. However, the complex optical environment and the limited resources of underwater equipment pose significant challenges to achieving high accuracy and low power consumption. To address these issues, we propose Spiking Underwater YOLO (SU-YOLO), a Spiking Neural Network (SNN) model. Leveraging the lightweight and energy-efficient properties of SNNs, SU-YOLO incorporates a novel spike-based underwater image denoising method based solely on integer addition, which enhances the quality of feature maps with minimal computational overhead. In addition, we introduce Separated Batch Normalization (SeBN), a technique that normalizes feature maps independently across multiple time steps and is optimized for integration with residual structures to capture the temporal dynamics of SNNs more effectively. The redesigned spiking residual blocks integrate the Cross Stage Partial Network (CSPNet) with the YOLO architecture to mitigate spike degradation and enhance the model's feature extraction capabilities. Experimental results on URPC2019 underwater dataset demonstrate that SU-YOLO achieves mAP of 78.8% with 6.97M parameters and an energy consumption of 2.98 mJ, surpassing mainstream SNN models in both detection accuracy and computational efficiency. These results underscore the potential of SNNs for engineering applications. The code is available in https://github.com/lwxfight/snn-underwater.

Figures (15)

• Figure 1: Comparison of detection performance and energy consumption across various models on URPC2019. The blue dots represent ANN models, while the green and red dots indicate SNN models.
• Figure 2: Diagram of SU-YOLO and the structure of its modules. In the main structural diagram, narrower graphics represent feature maps with reduced dimensions.
• Figure 3: Schematic diagram illustrating the processing flow and effects of the SpikeDenoiser module.
• Figure 4: Feature maps extracted from backbone layers when using ResNet or CSPNet in SNN.
• Figure 5: Impact of pooling and activation sequence on output.