在图像美学质量评价的发展过程中，前期有大量的科研工作用于人工设计美学特征，当前深度卷积网络的应用虽然减少了人工设计美学特征的流程，但使模型变得不可解释，这影响了人类对美学的认知。本文利用现有的可解释算法，对美学单一数值评价模型做出一定的解释。通过相关性归因方法，将原始输入图像对分类结果影响权重大的区域进行可视化，对证明模型的可靠性、发现新的美学规则具有十分重要的意义。

In the development process of image aesthetics quality evaluation, a large amount of scientific research work was used to artificially design aesthetic features in the early stage. Although the current deep convolutional network reduces the manual design process, the model becomes unexplainable, which affects the aesthetics of human beings’ cognition. Therefore, in this article, by using the existing interpretable algorithm, a certain explanation is made on the aesthetic single numerical evaluation model. This article uses the gradient attribution method to visualize the areas where the original input image has a significant influence on the classification results, which is of great significance for proving the reliability of the model and discovering aesthetic rules.