Geological symbol recognition on geological map using convolutional recurrent neural network with augmented data

Abstract

Geological maps contain rich geological knowledge, such as faults, structures, minerals, etc. Automatically and accurately recognition of geological symbols is the basis step for understanding geological maps and constructing geological knowledge connections between maps and texts. Due to the diverse combinations of symbols, complex background and color noise interference, symbol subscripts in geological maps directly affect the accurate recognition of geological symbols. In order to solve the above problems, this paper proposes a three-stages of the framework based on deep learning to recognize symbols in geological maps. The framework contains dataset automatic construction, convolutional recurrent neural network (CRNN) model training, and geo-symbol index construction. First, we propose a method to generate a base character-based training dataset that can generate geological map legend datasets of arbitrary length and different color backgrounds; second, we train a variable-length image text recognition optical character recognition (OCR) model CRNN and conduct comparative experiments to verify the effectiveness of our proposed recognition framework. Finally, the stage of geo-symbol index construction establishes the corresponding index list of geological symbols and corresponding descriptions for converting the recognized geological symbols into corresponding names and finally output the result. We performed experimental validation and analysis on our automatically generated dataset. The experimental results show that the accuracy of our algorithm recognition reaches 94%, which verifies the effectiveness of our proposed algorithm.