基于注意力U-Net模型的露头裂缝自动识别方法

应新疆塔里木油田深层两大类复杂油藏项目的要求，采用多传感器联合的野外地质露头采集技术与多尺度数字露头构建技术，建立了基于多尺度-多信息三维数字露头的综合地质知识库，其中一项重要的工作是深度学习在岩性、缝洞等方面的自动识别与研究。基于无人机拍摄的裂缝图像分辨率高且复杂的特点，在图像预处理阶段，首先对原始图像进行了切割，并做了灰度化处理以降低计算量。采用在医学领域内识别图像效果显著的U-Net网络模型，同时在传统U-Net模型的跳跃连接阶段引入注意力机制以提高目标特征的关注能力，带来模型性能的提升。使用处理过的真实岩石裂缝图像，完成了模型的训练和测试。实验结果表明：改进之后的网络模型裂缝识别效果良好，精确率可以达到88%,较原始网络提高了2.37%,损失在0.110左右，F₁分数能够达到85以上。

An automatic identification method of outcrop crack based on attention U-Net model

At the request of two types of complex reservoir projects in the depth of a domestic oilfield, the multi-sensor combined field geological outcrop collection technology and multi-scale digital outcrop construction technology are used to establish a comprehensive geological knowledge base based on multi-scale-multi-information three-dimensional digital outcrop, one of the important tasks is the automatic identification and research of deep learning in lithology, seam and other aspects. Based on the high resolution and complexity of the crack images taken by the drone, the original images were cut in the image preprocessing stage and grayscaled to reduce the amount of computation. Using a network model based on U-Net structure, the attention mechanism is introduced in the jump connection stage of the traditional U-Net model to improve the grasping ability of the target features and improve the performance of the model. Using the processed images of real rock cracks, the model was trained and tested. The experimental results show that the improved network model has a good crack identification effect, the accuracy can reach 88%, which is 2.37% higher than that of the original network, the loss is about 0.110, and the F1 score can reach more than 85.