基于深度学习的露头地层剖面裂缝自动提取

At present, the main targets of oil and gas exploration and development in China are gradually shifting to carbonate oil and gas resources and shale oil and gas resources, and carbonate reservoirs are mainly fracture-vuggy reservoirs. The extraction of fractures from outcrop formation profiles can provide intuitive and vivid results for the study of fracture-vuggy reservoirs. The most common traditional research on outcrop fractures is time-consuming and labor-intensive manual interpretation.Therefore, an intelligent fracture extraction technology based on deep learning is proposed.Based on the Mask R-CNN algorithm, the online augmentation strategy and attention mechanism are combined to realize automatic fracture extraction.In order to prove the error tolerance ability of the improved Mask R-CNN algorithm proposed in this study, comparison experiments were designed for training sets containing different degrees of error labels.The results showed that the accuracy of the model decreased greatly when the proportion of error labels reached 30%.Then, through the quantitative analysis of ablation experiments, different improvements can improve the accuracy of the algorithm. The results show that the accuracy of the model is improved after the introduction of attention mechanism and online enhancement.Then, the accuracy of manual extraction was compared with OTSU method, regional growth algorithm, UNet algorithm and DeepLabv3+ algorithm.The results show that the linear features extracted by the proposed method are more accurate and complete, and the accuracy is more than 97%, which is higher than other comparison algorithms, and verifies the effectiveness of the improved Mask R-CNN method.Finally, the method was applied to extract fractures from outcrop profiles in the Ebian Xianfeng area, southwest Sichuan Basin. The parameters of fracture length, density, inclination and spacing were analyzed statistically, and the fracture distribution characteristics were quantitatively analyzed.The feasibility of the improved Mask R-CNN method was verified. This method provides a basis for speeding up the comprehensive study of petroleum exploration and target evaluation, reservoir prediction and automatic development of outcrop profile parameter characterization.