Abstract: Hydraulic engineering projects are mostly located in regions of high mountains and deep gorges with complicated geological conditions. Evaluation of the integrity of rock mass in the dam site and reservoir area is significant to the construction, but traditional evaluation, often completed manually, is laborious and of high cost. In this study, a knowledge transfer method is applied to achieve a deep feature ensemble of rock core images; then an intelligent evaluation model is developed for hydraulic rock mass integrity based on Weighted Support Vector Machine (WSVM). We compare the evaluation results obtained using a single deep model and a deep feature ensemble method, and find that the latter improves model performance better, raising the accuracy by more than 5%. We also compare WSVM with SVM and other machine learning methods. The results prove that WSVM is more effective in the intelligent evaluation of hydraulic rock mass integrity. It realizes automatic and intelligent analysis of the integrity evaluation to a certain extent, and provides a new method for geological survey and hydraulic construction.

Key words: core image, rock mass integrity evaluation, deep feature ensemble, deep learning, machine learning