基于融合模型的心脏病图像检测研究
Research on Heart Disease Image Detection Based on Fusion Model

针对目前临床心脏病图像检测中存在的准确率低、效果差等问题，提出了一种融合YOLOv5和Attention的检测技术。这种方法旨在充分利用YOLOv5模型在目标检测领域的优势，同时结合Attention注意力机制，以提高对心脏病图像的识别和分类能力。据调查显示，传统的YOLOv5模型在检测心脏病图像时准确率较低且在处理复杂背景和局部特征方面存在一定的局限性，分类准确率不足80%，在实际医疗检测中误差较大。在此基础上，进一步将Attention注意力机制与YOLOv5模型相融合。具体来说，首先通过YOLOv5模型对图像进行特征提取，使用分支网络分别预测特征图中物体的位置和类别，初步得到检测结果并用于下游任务进行分类。然后，Attention机制会根据输入检测结果的特征分布自动计算出每个区域的权重，使模型在检测心脏病图像时能够聚焦于特定的部分，更好地捕捉到关键信息，从而使心脏病图像分类准确性达到98%以上。
A detection technique combining YOLOv5 and Attention is proposed to address the issues of low accuracy and poor effectiveness in current clinical heart disease image detection. This method aims to fully utilize the advantages of the YOLOv5 model in the field of object detection, while combining the Attention mechanism to improve the recognition and classification ability of heart disease images. According to a survey, the traditional YOLOv5 model has low accuracy in detecting heart disease images and has certain limitations in processing complex backgrounds and local features. The classification accuracy is less than 80%, and there is a significant error in actual medical detection. On this basis, the Attention mechanism is further integrated with the YOLOv5 model. Specifically, the YOLOv5 model is first used to extract features from the image, and a branch network is used to predict the position and category of objects in the feature map. The initial detection results are ob-tained and used for downstream task classification. Then, the Attention mechanism will automatically calculate the weight of each region based on the feature distribution of the input detection results, enabling the model to focus on specific parts when detecting heart disease images, better capturing key information, and achieving a classification accuracy of over 98% for heart disease images.