Petrographic recognition and classification of bioclastic carbonate thin sections based on attention mechanism

Abstract

The huge reserves of carbonate reservoirs have become an important target for oil and gas exploration and development, but their complex pore structure makes petrographic identification difficult, while the traditional identification and classification process using manual analysis of rock thin sections suffers from problems such as being susceptible to subjective factors and being time-consuming. To address these problems, this paper proposes a new method for the intelligent classification of thin sections based on an attention mechanism. Firstly, 295 original images of carbonate reservoirs in the Haffaya oil field of Iraq in the Middle East are used for training to obtain a network model that can directly use rock thin sections images as input variables for analysis of the classification network. It was then classified into seven categories according to Dunham's classification method and compared with the ResNet classification network. The results show that the combined accuracy of the method is 13.8% higher than that of the ResNet network, reaching 93.6%; it can effectively extract the relevant characteristics of rock thin sections under the condition of small sample data, and has strong applicability in complex carbonate reservoirs.