基于岩石新鲜面图像与孪生卷积神经网络的岩性识别方法研究

Lithology Recognition Based on Fresh Rock Images and Twins Convolution Neural Network

目前岩性识别多基于人工判别方法,需要一定的专业背景和丰富的判别经验。该文提出基于岩石新鲜面图像与孪生卷积神经网络结构的深度学习岩性自动识别方法,兼顾岩石数据的全局图像信息和局部纹理信息。首先利用孪生卷积神经网络中的子通道提取岩石图像的全局和局部特征信息,再将特征信息融合以构建统一描述子,最后根据描述子信息识别岩性。选取野外拍摄的岩石图像作为模型验证数据,通过专家命名构建深度学习样本库对模型进行验证和分析。实验结果表明,该文提出的基于AlexNet孪生卷积神经网络对岩石数据的适用性较强,对岩性的识别精度达89.4%,能很好地区分岩石类型。

Lithology recognition is mostly based on artificial identification methods,and it needs professional background and rich experience.An automatic deep learning lithology recognition method based on fresh rock images and twins convolution neural network structure is proposed in this paper,which takes full advantages of the global and local context information of rock data.The proposed method first extracts global and local feature information from rock images by sub-channels of twins convolution neural network,then fuses the extracted features to construct a unified descriptor,and finally recognizes rock lithology according to the descriptor.The rock images taken in the field are selected as experimental data,the model is validated and analyzed by using a deep learning sample bank named by experts.The results show that the proposed method based on AlexNet twins convolution neural network is more applicable to rock data.The lithology recognition accuracy achieves 89.4%.The model proposed in this paper can distinguish rock types very well and has certain practical value.