ZHANG Xiaoyan1, LI Yan2, LU Bibo1*, HOU Guangshun3, XING Zhifeng3, YANG Xiaopeng4
(1. School of Computer Science and Technology, Henan Polytechnic University, Jiaozuo 454003, China;
2. College of Computer Science and Technology, Jiaozuo Metallurgical building Materials Senior Technical School, Jiaozuo 454003, China;
3. Institute of Resources and Environment, Henan Polytechnic University, Jiaozuo 454003，China;
4. School of Surveying and Land Information Engineering, Henan Polytechnic University, Jiaozuo 454003,China)
Abstract: Oolite is a special kind of sedimentary particles, and its distribution density, grain size and other information can intuitively reflect the water depth and hydrodynamic conditions of the formation environment. It has important geological significance. In geology, oolitic rock specimens are usually ground into rock thin sections which are then observed under the microscope by professionals for obtaining the estimated values of oolitic content, roundness, grain size and so on. This method has various shortcomings including the large amount of computation, high cost, long period and large manpower input, etc. Moreover, this method is greatly affected by subjective factors, with relatively different results obtained by different experts. To address the above problems, in this paper, we have proposed a deep learning-based oolite intelligent detection and feature statistics method. The YOLOv5 detection model is mainly adopted for detecting micrographs of oolite rock thin sections, and a lightweight SE-Net channel attention mechanism module is then added into the backbone part of the YOLOv5 network for improving the detection performance. Then, the NMS method is replaced by the DIoU-NMS method for improving the problem of missing detection when the oolite distribution is crowded in the image. The experimental results show that final accuracy of the improved algorithm reached 98.8%, which was 1.3% higher than that of the original algorithm. Finally, quantitative statistics and analysis of the detection results have been carried out by using the image processing technology, and histograms of the statistical results of the oolite content, roundness information and grain size in images have been obtained. The great convenience is provided to geological staffs for carrying out related works.
Keywords: deep learning; improved YOLOv5; oolitic detection; attention mechanism; statistical analysis
