A shale gas exploitation platform detection and positioning method based on YOLOv5

Abstract

Shale gas is one of the most critical components of unconventional oil and gas resources. Remote sensing image contains rich spatial information. How to use remote sensing images to quickly and efficiently obtain the target position information is one of the hot spots in current remote sensing image researches. In this paper, a deep learning based method is used to study the detection and positioning of shale gas exploitation platform in Fuling area. In order to improve the detection accuracy and recall rates, the loss function is adjusted. CIoU, which contains the scale, the distance between the target and anchor, the overlap rate and the length-width ratio of predict box, is used as the loss function to calculate bounding boxes. By testing, CIoU was used to calculate bounding box losses, which improved the Precision, Recall and AP of YOLOv5x model by 2.6%, 1.6% and 1.2%, respectively. In addition, this paper presents a method for target detection in large-scale remote sensing images. The main idea of this method is cropping-predicting-filtering-stitching. By testing, the method that we proposed can automatically obtain the location of the shale gas exploitation platform quickly and accurately from the whole remote sensing image. This method makes it possible to predict the whole remote sensing image and realize the detection and positioning of the shale gas exploitation platform in one step.