A Real-Time Oil Content Analysis Method of Cuttings Based on Deep Learning

Abstract

In the oil exploration site, logging technology is usually used to obtain geological data, and the obtained cuttings are analyzed by manual identification. Due to the different experience of logging personnel and their different cognition of the results, only qualitative analysis can be conducted. In this paper, a real-time analysis method of oil content in cuttings based on deep learning is proposed. Firstly, the cuttings image is collected under the fluorescent lamp, and then the semantic segmentation network MobileNetV3 UNet (M3-UNet) is used to segment the oil-bearing region of cuttings image automatically. Finally, the real-time quantitative analysis of oil content in cuttings is realized. In order to make the proposed algorithm be applied to the oil exploration site, the proposed algorithm is deployed on the Jetson edge calculation equipment to realize the portable oil content in cuttings analysis equipment. In the coding stage of the proposed M3-UNet, MobileNetV3 is used as the backbone network to extract the features from the input cuttings image, which can reduce the requirements on the performance of logging equipment while maintaining the segmentation accuracy. The experimental results show that the Intersection over Union (IoU), F1-score and Pixel Accuracy (PA) of M3-UNet segmentation in the test dataset are all more than 9% higher than U-Net and MobileNetV2 UNet (M2-UNet). The segmentation speed of the cuttings image on Jetson equipment reaches 21FPS, meeting the requirements of real-time analysis of oil content of the cuttings image on the logging site.

Keywords

• Deep learning
• oil exploration
• U-Net
• cuttings image
• image segmentation