在水产养殖业中，鱼类疾病的诊断与治疗对防止疾病蔓延和减少经济损失至关重要。为解决淡水鱼类细菌性疾病检测的问题，提出了一种基于改进YOLOv8算法的鱼类疾病检测方法。该方法首先在骨干网络中引入了EMA（Efficient Multi-Scale Attention）注意力机制，不仅增强了特征提取能力，而且通过创新的多尺度特征提取和跨空间学习架构，在降低计算复杂度的同时保持了高精度的特征表达。此外，还在Neck层中采用GSConv（Grouped Shifted Convolution）替换了传统的卷积操作，降低了模型复杂度，进一步提升了检测速度，同时确保了检测精度不受影响。结果显示：该方法相较于原始YOLOv8模型提升了2.1个百分点的检测精度，相较于其他现有模型也有显著的性能提升。研究表明：该方法可以应用于鱼类疾病检测防治场景，为鱼类病害检测提供技术支撑。[] [] 

With the development of deep learning technologies, object detection has become an important task in computer vision and has been widely applied in various fields. The YOLO (You Only Look Once) series of models, known for their efficient and fast inference capabilities, have become mainstream in the field of object detection and are widely used in various domains. In the aquaculture industry, diagnosing and treating fish diseases is crucial for preventing the spread of diseases and reducing economic losses. To address the problem of bacterial disease detection in freshwater fish, this paper focuses on the application of the YOLOv8 model in object detection and explores the role of data augmentation in improving model performance. Firstly, the basic principles and architecture of YOLOv8 are introduced, and the improvements of this model over previous YOLO versions are analyzed in detail, including the advantages of its network structure and optimization algorithms. Next, this paper proposes a fish disease detection method based on an improved YOLOv8 algorithm. This method incorporates the EMA (Efficient Multi-Scale Attention) attention mechanism into the backbone network, which not only enhances the feature extraction capability but also improves multi-scale feature extraction and cross-space learning architecture. This innovation reduces computational complexity while maintaining high-precision feature representation. Additionally, the GSConv (Grouped Shifted Convolution) operation is adopted in the Neck layer to replace traditional convolution operations, which reduces model complexity and further enhances detection speed without compromising accuracy. Experimental results show that this method achieves a 2.1 percentage point improvement in detection accuracy on our self-built freshwater fish disease dataset compared to the original YOLOv8 model. It also demonstrates significant performance improvement over other existing models. This method can be applied to fish disease detection and prevention scenarios, providing technical support for fish disease detection.