Abstract: In the process of modern hydraulic engineering management, daily inspections and special surveys generate a large number of on-site images. However, images captured by multiple teams are often uploaded without classification, leading to information overload and making it difficult to effectively utilize historical images during data analysis. To address the low efficiency of manual classification, this paper proposes an intelligent archiving framework based on an improved YOLOv8 detector, enabling automated processing from object detection to image classification. A dedicated dataset containing 12 types of typical hydraulic engineering targets is constructed. The EMA attention mechanism is integrated to enhance multi-scale feature representation, and the SIoU loss function is adopted to optimize bounding box regression accuracy. Experimental results show that the improved model achieves mAP50-95 of 0.6243 and mAP50 of 0.7932, representing improvements of 9.00% and 8.83% over the baseline model, respectively. Leveraging the structured information, such as target categories, positions, and scales, derived from detection outputs, the method enables multi-dimensional automatic archiving of images according to engineering components and work scenarios, achieving a classification accuracy of 95.38% on a test set of 3,348 images. The detection-driven archiving (DDA) approach provides an effective technical pathway for intelligent management of hydraulic engineering image data.