Carbonate texture identification using multi-layer perceptron neural network

This study proposes an approach to identifying carbonate rocks depositional texture utilizing more than 1250 thin-section photos as a data set from shallow-water carbonates of the Kruja zone (Albania). The proposed technique uses a numerical methodology based on digitized images of thin sections. According to the Dunham classification, it permits the user to recognize the carbonate texture automatically. The carbonate rock should not be affected by post-depositional processes. This technique uses as input gray-scale digital images taken from thin sections. Image processing can extract (as output) a set of 22 numerical parameters measured in the entire image, including white areas representing calcite cement. A multi-layer perceptron (MLP) neural network inputs these numerical parameters and produces, as output, the estimated depositional texture. We utilized 654 images of thin sections to evaluate this technique to train the neural network. We used 348 pictures taken from the same data set and 250 from the Burizana section (Kruja zone, Albania) to test the technique. The proposed method has shown 90.5 and 91.3% accuracy in identifying the automatically depositional texture using digitized images on the 348 and 250 test sets. Based on the good results obtained, this technique can be extending not only in identifying carbonate rocks texture but in every type of rock, getting fast and correct results. Python was used as a computer programming language for image processing and displaying images.