Rock Image Automatic Annotation based on Deep Neural Network

Abstract

The automatic image annotation is a technology that uses deep neural network to extract features from the input image, and finally converts it into text description. It is currently one of the hot researches of explainable artificial intelligence. The thin section image of rock is an important data source in the field of geological exploration and oil and gas development. It can reflect the microstructure inside the rock. However, in actual research, if you only rely on manual observation to extract features from hundreds of rock images, it will affect the research efficiency and take time and effort. This paper uses GoogLeNet model in the deep convolutional neural network to implement the feature extraction of rock images, and also uses the LSTM model in the deep recurrent neural network to convert the image features into text features, and finally converts the rock images into text descriptions. The automatic image annotation technology of rock images based on deep neural networks not only realizes a step of converting rock images into text, but also assists scholars in the field of petroleum geology to realize efficient research.