Identification of Rock Cracks in Outcrop Areas Based on Improved DeepLab V3+

The study of reservoir fractures has always been a key focus in oil and gas exploration and development, and the most intuitive method is to observe rock fractures in outcrop areas. Manual surveying has low efficiency and strong subjectivity, so identifying rock fractures in outcrop areas has always been a difficult task. In response to this difficulty, this article utilizes drone oblique photography technology to significantly improve surveying efficiency, and proposes an improved DeepLabv3+ semantic segmentation model to achieve accurate segmentation of rock fractures. The experimental results showed that the MAP (Mean Average Precision) of the improved DeepLabv3+ model reached 83.77%, which was 12.44% higher than the original DeepLabv3+ model; MIoU (Mean Intersection over Union) has increased from 68.21 % to 78.84%; The accuracy has increased from 74.83% to 91.66%; The parameter count is only 15.14M. The experimental results show that the improved DeepLabv3+ model in this paper exhibits superior segmentation performance in rock fracture segmentation tasks in field outcrop areas, and has good mobile deployment performance.