Automatic detection of borehole breakout for image logs using YOLO algorithm

In-situ stress estimation is crucial for subsurface stability, carbon sequestration, and optimizing petroleum resource development, particularly in hydraulic fracturing. Borehole breakout serves as a key stress indicator, but their manual interpretation is time-consuming and subjective. This study proposes an automatic borehole breakout detection model for image logs using you only look once (YOLO) algorithm. The model is trained using image data from Well A2, incorporating domain-knowledge based preprocessing and data augmentation to enhance breakout detection accuracy. The model is evaluated under two scenarios: without data augmentation (Case 1) and with data augmentation (Case 2). Case 2 achieves a 78.81 % improvement in average precision, significantly outperforming Case 1. To validate its performance, the trained model is tested on Wells A2 and B1, comparing its results with expert analysis. The results confirm that Case 2 provides more reliable breakout detection, minimizing error across both wells. Additionally, the proposed model automatically computes breakout geometry, including azimuth, depth, opening angle, and length in under 0.1 s per 3-m image log, offering significantly more efficiency than manual expert analysis performed by experts.