摘要： 对于海产生物科考与捕捞等行业，在海上远洋的船只在利用水下机器人进行海产生物的捕捞与识别时，由于通信带宽受限，计算资源有限，而采用轻量化网络模型可以更好地适应这样的条件。为此，提出了一种改进YOLOv5的海产生物目标检测模型。首先，引入高效的轻量化卷积模块Gsconv（Group Shuffle Convolution），对模型主体进行缩减；然后改进损失函数，引用 -giou损失函数进行优化，提升预测框回归精度；再引L1-norm正则化剪枝，裁剪不必要的通道以及相关的卷积核；最后采用L2知识蒸馏，将模型精度调整到接近剪枝前的水平。结果显示，与原有基线模型YOLOv5n相比，改进后的模型计算量下降了53%，参数量下降了51%。所提出的改进算法在保持模型性能的同时保证了轻量化处理的有效性。

关键词: 海产生物目标检测, YOLOv5n, Gsconv, 模型剪枝, 知识蒸馏

Abstract: In marine biology research and fishing industries, ships using underwater robots for capturing and identifying marine organisms may face limited communication bandwidth and computing resources, making lightweight network models a better fit for such conditions.To address this issue, a modified YOLOv5 model for marine organism detection was proposed, incorporating improvements in the neck network, model pruning, and knowledge distillation techniques. By leveraging the Gsconv lightweight convolution module to replace standard convolutions in the YOLOv5n Neck section, the model was effectively reduced. Additionally, a novel α-giou loss function was adopted to enhance bounding box regression accuracy. L1-norm regularization pruning was applied to eliminate unnecessary channels and associated convolutional kernels based on weight coefficients. Finally, retraining and L2 knowledge distillation were employed to fine-tune the model accuracy close to pre-pruning levels. Experimental results demonstrated a 53% reduction in computational load and a 51% decrease in parameters compared to the original YOLOv5n baseline network. The proposed algorithm ensures the effectiveness of lightweight processing while maintaining model performance, offering a promising approach for lightweight handling of marine organism recognition models for underwater robots.

Key words: Marine biological target detection, YOLOv5n, Gsconv, Model pruning, Knowledge distillation