Abstract: In the freshwater snail product classification and processing scenario, accurately and efficiently identifying the sexes and dead features of Cipangopaludina cahayensis is crucial for the quality classification and grading of freshwater snail products. Distinguishing between male and female individuals allows for targeted selection of high-quality parents for snail seed breeding. Timely removal of rotten dead snails is important for maintaining water quality in aquaculture and disease prevention. Currently, the methods for identifying the sexes and dead features of Cipangopaludina cahayensis mainly include: 1) distinguishing between males and females based on differences in antennae observed in their natural state; 2) distinguishing between male and female glands under strong light transmission and observing the internal shrinkage of snails after death to differentiate. However, these methods suffer from issues such as high workload, subjectivity, high time costs, low detection efficiency, and high false detection rates. In response to the demands of modernizing China's fisheries, achieving automation and intelligence in the classification of male and female individuals and dead features of Cipangopaludina cahayensis is of significant importance for improving the technological development of freshwater snail factory farming and aquatic product classification and processing. Therefore, how to achieve accurate identification and rapid detection of the sexes and dead features in the processing of snail products is a pressing issue that needs to be addressed in the automation of quality classification and grading operations for Cipangopaludina cahayensis. Cipangopaludina cahayensis has a relatively late development in automated aquaculture compared to other aquatic organisms, with limited targeted research on intelligence. Additionally, existing algorithmic literature focuses solely on the detection of the phenotypes of freshwater snail shells, neglecting considerations such as model lightweighting, unbalanced detection accuracy, and real-time detection speed. The detection effectiveness of the algorithms for distinguishing the sexes and dead features of Cipangopaludina cahayensis still remains inadequate. In summary, this study adopts the YOLOv8n model as the base model and proposes a Cipangopaludina cahayensis male-female and death feature detection algorithm based on AP2O-YOLOv8. This research aims to provide a theoretical foundation and reference for the automation and intelligence of processes such as quality classification and grading of Cipangopaludina cahayensis products. In terms of model design, this study introduces the P2 layer for small target detection, incorporates larger-scale feature maps containing more information about snail target positions and inter-class local features, and combines the ASF-YOLO structure and C2f-OREPA module to further enhance the algorithm's multi-scale feature fusion capability and real-time detection speed. This approach allows the model to have higher detection performance while being more lightweight and efficient. The improved algorithm in this article integrates three enhancement schemes. Compared to the original YOLOv8, its precision (P), recall (R), and mean average precision at IOU 0.5 have increased by 2.1%, 2.6%, and 5.6% respectively. The parameter size has decreased from 2.9MB to 2.1MB, a reduction of 27.6%. The frames per second (FPS) have increased from 180 to 226, a 25.6% improvement. The AP2O-YOLOv8 model proposed in this article for the detection of male and female Cipangopaludina cahayensiss, as well as their vital status, significantly enhances the detection accuracy of different features of Cipangopaludina cahayensiss compared to the original benchmark model. Simultaneously, it effectively reduces the complexity of the model, greatly increasing real-time detection speed. This study provides new ideas and methods for the classification and detection of male and female, live and dead Cipangopaludina cahayensiss, helping further advance the automation and intelligence upgrade of the quality classification and processing process of Cipangopaludina cahayensiss.

Key words: Cipangopaludina chinensis, male and female snail, dead snail, YOLOv8, ASF-YOLO, small target detection head, C2f-OREPA

摘要： 淡水螺产品分类加工环节中，对中华田螺雌雄及死活个体进行准确的品质分类至关重要。针对现有目标检测算法在该类型分类作业任务中精度不足、参数量过多、检测速度低的问题，本研究提出一种基于AP2O-YOLOv8的中华圆田螺雌雄及死亡特征检测算法。该算法通过引入P2层小目标检测头，从而提升网络对于田螺细微特征的检测精度。其次，通过结合ASF-YOLO结构，充分强化网络的多尺度特征融合能力。此外，将主干网络的C2f模块替换为C2f-OREPA模块，使网络复杂的结构重参数转为单卷积层，有效减少模型的推理成本。试验结果表明，在该数据集上AP2O-YOLOv8算法的mAP0.5为93.2%，参数量为2.1MB，FPS为226。相较原YOLOv8n的mAP提升了5.6%，参数量降低了27.6%，FPS提升了27.6%，在提升检测精度和实时检测速度的同时还降低了模型部署难度。本研究为中华圆田螺雌雄及死亡特征分类检测提供新的思路和方法，有助于进一步推动实现中华圆田螺品质分类加工环节的自动化及智能化技术升级。

关键词: 中华圆田螺, 雌雄田螺, 死亡田螺, YOLOv8, ASF-YOLO, 小目标检测头, C2f-OREPA