WO2024099422A1 - 一种基于UNet的面波频散质量快速评估方法 - Google Patents
一种基于UNet的面波频散质量快速评估方法 Download PDFInfo
- Publication number
- WO2024099422A1 WO2024099422A1 PCT/CN2023/130903 CN2023130903W WO2024099422A1 WO 2024099422 A1 WO2024099422 A1 WO 2024099422A1 CN 2023130903 W CN2023130903 W CN 2023130903W WO 2024099422 A1 WO2024099422 A1 WO 2024099422A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- unet
- dispersion
- surface wave
- quality
- frequency dispersion
- Prior art date
Links
- 239000006185 dispersion Substances 0.000 title claims abstract description 104
- 238000011156 evaluation Methods 0.000 title abstract description 8
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims abstract description 9
- 230000008859 change Effects 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 17
- 238000001303 quality assessment method Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 4
- 238000013135 deep learning Methods 0.000 claims description 3
- 210000004205 output neuron Anatomy 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 230000000737 periodic effect Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 8
- 238000013441 quality evaluation Methods 0.000 abstract 2
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000012216 screening Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 2
- 238000004880 explosion Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 238000003325 tomography Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 101000827703 Homo sapiens Polyphosphoinositide phosphatase Proteins 0.000 description 1
- 102100023591 Polyphosphoinositide phosphatase Human genes 0.000 description 1
- 101100012902 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) FIG2 gene Proteins 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013211 curve analysis Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
Definitions
- the invention belongs to the field of seismic wave data processing, and in particular relates to a UNet-based rapid evaluation method for seismic surface wave dispersion quality.
- the surface wave dispersion method is widely used in the fields of crustal structure detection, engineering and environmental exploration.
- Surface wave dispersion data from different periods have different sensitivities to velocity structures at different depths.
- Short-period dispersion data are sensitive to shallow structures, while long-period dispersion data are sensitive to deep structures. Therefore, surface wave dispersion curve analysis is an effective method for studying crust-mantle structure.
- inverting the dispersion curve By inverting the dispersion curve, a one-dimensional S-wave velocity structure can be obtained, and two-dimensional and three-dimensional velocity structures can be combined from the one-dimensional velocity structure.
- An important prerequisite for obtaining a reliable velocity structure through inversion is to extract a reliable dispersion curve.
- the surface wave dispersion is mainly obtained by cross-correlating seismic events or noise data obtained from two seismic stations, and then the work area is divided into multiple grids. The dispersion curve on each grid can be obtained using travel-time tomography technology.
- the extraction of surface wave dispersion mainly includes four steps (taking the double-station method as an example): 1. Filter the waveforms of the two stations, then perform cross-correlation calculation on the filtered seismic waves, and obtain the related time difference, and further calculate the seismic wave velocity between the two stations at a certain frequency; 2. Use frequency as the horizontal coordinate, velocity as the vertical coordinate, and cross-correlation value as the energy value to draw the two-dimensional velocity spectrum of the surface wave; 3. Manually select (or automatically identify by computer) the area with the maximum energy on the velocity spectrum to form a curve, namely the dispersion curve; 4.
- the present invention provides a method for rapid evaluation of surface wave dispersion quality based on UNet.
- a training set and a test set are established through actual data, and then UNet is constructed to realize automatic evaluation of surface wave dispersion quality, which greatly improves the efficiency of data processing.
- the present invention provides a method for quickly evaluating surface wave dispersion quality based on UNet, comprising the following steps:
- Step 1 Collect surface wave dispersion data, manually mark each dispersion point with a qualified label, and divide it into training set samples and test set samples;
- Step 2 Construct a UNet model for surface wave dispersion quality assessment and train the UNet model using training set samples;
- Step 3 The period of surface roll dispersion, surface roll dispersion, and the rate of change of dispersion are used as inputs of the trained UNet model to obtain the UNet prediction result of the surface roll dispersion after quality assessment.
- each training sample needs to be manually assigned a qualified region (label), and the qualified frequency Scatter points are marked as 1 and unqualified ones are marked as 0.
- the UNet network model has 4 layers on the left and 4 layers on the right, including a total of 15 convolutional layers, 3 pooling layers, and 3 transposed convolutional layers; the RELU activation function is selected to act on the output of the convolutional layer; and Batch_normilization is used to normalize the data after the convolution operation.
- the input layer of the UNet model includes three channels: period, surface wave dispersion, and dispersion change rate; the input layer has a total of 48 neurons, and the data of less than 48 dispersion points are filled with zero; the output layer is the label corresponding to each period point, the quality qualified label is 1, otherwise the label is 0;
- the sigmoid activation function is applied to the output layer so that the value of the final output neuron is between 0 and 1.
- x is the input.
- the subscript i represents the i-th training sample, N is the total number of training samples; D is the prediction result achieved by deep learning; label i is a manually given label, which characterizes the quality of the data at each sampling point, d i is the input dispersion data, which contains three channels; ⁇ is the hyperparameter that needs to be updated in the network back propagation.
- the present invention proposes to use UNet to quickly evaluate seismic surface wave dispersion data, establish a training set and a test set through actual data, and then construct a UNet model to realize automatic evaluation of the surface wave dispersion quality.
- the method described in the present invention is concise and efficient, and can perform rapid prediction after the model is trained, complete the processing of batch dispersion data in a short time, realize rapid evaluation of the quality of the surface wave dispersion curve, eliminate the data parts that do not meet the requirements, and retain high-quality data.
- This method can not only serve as a substitute for conventional methods, but also greatly improves the efficiency of data processing compared to traditional manual recognition methods, laying a foundation for real-time inversion of surface wave dispersion.
- Figure 1 is a schematic diagram of the distribution of stations that collect surface wave dispersion data
- FIG2 is a schematic diagram of the UNet network model structure of the present invention.
- Figure 3 is a schematic diagram of the UNet prediction results of surface wave dispersion
- FIG4 is a schematic diagram of the prediction results of the test set.
- the UNet-based rapid assessment method for surface wave dispersion quality described in the present invention comprises the following specific steps:
- the current method mainly uses the cross-correlation of earthquake events or noise data obtained from two seismic stations. Then the work area is divided into multiple grids, and the dispersion curve on each grid can be obtained using travel time tomography technology.
- the surface wave dispersion of the present invention is obtained by noise calculation. A total of 100000 square meters of surface wave dispersion data from December 2013 to March 2015 were collected. The vertical component continuous time series of (15 months) was recorded by 668 broadband and ultra-wideband portable stations (China Earthquake Array, Phase II) and 62 permanent broadband stations (as shown in Figure 1).
- the empirical Green's function was estimated by cross-correlation and other methods, and finally 4160 pairs of dispersion data were extracted for quality assessment training (Pan et al., 2019).
- Each training sample needs to be manually assigned a qualified area (label), and the qualified dispersion points are marked as 1 and the unqualified ones are marked as 0. 85% of them are used for training sets and 15% for test sets.
- the dispersion results after screening can be tomographically imaged to obtain the dispersion of the entire area.
- the selection of surface wave dispersion requires judging the quality of each frequency point of each dispersion data. Only when the data of all cycles of a dispersion curve are poor, they are all discarded; therefore, the quality assessment of surface wave dispersion is not a simple binary classification problem.
- the UNet neural network is selected to implement the quality assessment of the receiver function.
- the structure of the model is shown in Figure 2.
- the network has 4 layers on the left and right, including a total of 15 convolutional layers, 3 pooling layers, and 3 transposed convolutional layers. Since there are no negative numbers in the surface wave dispersion data, the RELU activation function is selected to act on the output of the convolutional layer.
- This activation function can improve the nonlinear ability of the network and avoid the problems of gradient explosion and gradient vanishing during training.
- Batch_normilization is used to standardize the data to further prevent the gradient vanishing or gradient explosion phenomenon, and at the same time increase the regularization effect.
- the input layer includes three channels: the period of surface wave dispersion, surface wave dispersion, and the rate of change of dispersion.
- the input layer has a total of 48 neurons, and the data with less than 48 dispersion points are filled with zero, which means that the network is suitable for quality assessment of dispersion data with less than 48 dispersion points with any period.
- the output layer is the label corresponding to each periodic point.
- the quality qualified label is 1, otherwise the label is 0.
- the sigmoid activation function is applied to the output layer so that the value of the final output neuron is between 0 and 1.
- x is the input.
- the subscript i represents the i-th training sample, N is the total number of training samples; D is the prediction result achieved by deep learning; label i is a manually given label that characterizes the quality of the data at each sampling point, d i is the input dispersion data, which contains three channels; ⁇ is the hyperparameter that needs to be updated in the network back propagation.
- the back propagation algorithm Adam (Kingma and Ba, 2014) was used to update the weights of the neurons, with a learning rate of 0.001. According to the decreasing curve of the objective function with the number of iterations, 70 iterations were selected to ensure that the objective function was sufficiently reduced.
- the data to be predicted is prepared in the format of the training set input, including the period, dispersion data and the time derivative of dispersion; then it is imported into the input layer of UNet to quickly obtain the quality assessment result.
- the quality assessment value of each period point is in the range of 0 to 1.
- a threshold for example, 0.8
- the dispersion value will be retained, otherwise the dispersion value will be deleted.
- the higher the value the higher the quality requirement of the network.
- a dispersion curve with better quality can be obtained, and the poorer quality part is deleted.
- Figure 3 shows the quality assessment results of three surface wave dispersion data, where a to d show the overall curves respectively.
- the evaluation results are good quality, good quality in the middle part, good quality in the back section, and poor quality of the overall curve.
- the upper figure of each sub-figure shows the comparison results of the dispersion data with good prediction quality and the original dispersion data, and the lower figure shows the prediction results.
- the straight line is the artificially given label, and the dotted line with a vertical bar represents the UNet prediction result. The closer to 1, the better the quality, and the length of the vertical bar represents uncertainty. It can be seen that the area with a large UNet prediction value corresponds to a relatively continuous and smooth dispersion curve part, which is a qualified standard dispersion curve shape.
- This method can quickly evaluate the quality of surface wave dispersion curves in a short time, remove the data that does not meet the requirements, and retain high-quality data.
- This method can not only replace the manual selection of dispersion curves, but also greatly improve the processing efficiency of surface wave dispersion data.
Landscapes
- Geophysics And Detection Of Objects (AREA)
Abstract
本发明属于地震波数据处理领域，公开了一种基于UNet的地震面波频散质量快速评估方法，其首先采集面波频散数据，对每个频散点人为标记合格标签，并将其分成训练集样本和测试集样本；然后构建面波频散质量评估的UNet模型，并利用训练集样本对UNet模型进行训练；最后将周期、面波频散、以及频散的变化率作为训练好的UNet模型的输入，得到质量评估后的面波频散的UNet预测结果。本发明所述的方法能够在短时间内完成批量频散数据的处理，不仅能够作为传统方法的替代，且相对于传统的人工识别方法极大地提升了数据处理的效率，为面波频散的实时反演奠定基础。
Description
本发明属于地震波数据处理领域，具体是涉及一种基于UNet的地震面波频散质量快速评估方法。
面波频散方法被广泛地应用于地壳结构探测、工程与环境勘探领域。来自不同周期的面波频散数据对不同深度的速度结构具有不同的敏感性，短周期的频散数据对浅部结构敏感而长周期的频散对深部的结构敏感；因此，面波频散曲线分析是研究壳幔结构的有效方法。通过反演频散曲线，可以得到一维S波速度结构，二维和三维的速度结构可以由一维速度结构组合起来。通过反演得到可靠的速度结构的重要前提是提取可信赖的频散曲线。目前主要通过两个地震台站获得的地震事件或者噪声数据进行互相关方法来获得面波频散，然后将工区划分为多个网格，使用走时层析成像技术可以得到每个网格上的频散曲线。
面波频散的提取主要包含四个步骤(以双台法为例)：1.对两个台站的波形进行滤波，然后对滤波后的地震波行进行互相关计算，并且得到相关时差，并进一步计算出某个频率下两个台站之间的地震波速度；2.以频率为横坐标，以速度为纵坐标，以互相关值为能量值，绘制出面波的二维速度谱；3.人工挑选(或者计算机自动识别)出速度谱上能量最大区域，形成一条曲线，即频散曲线；4.人工质量筛选，由于计算误差，或者原始波形的信噪比较低等问题，通常需要对得到的频散曲线进一步筛选，得到光滑的，速度值符合基本地质规律的频散曲线。在上述的步骤3中，由于存在计算不稳定，精度较低等问题，或者噪声的存在，提取的相速度可能存在不连续现象，且频散提取的过程需要人机交互进行频散拾取，这不可避免地引入了主观因素；某些自动频散拾取技术也可能导致部分周期相速度不合理的情况。
发明内容
为解决上述技术问题，本发明提供了一种基于UNet的面波频散质量快速评估方法，通过实际数据建立起训练集和测试集，然后构建了UNet实现对面波频散质量进行自动评估，极大地提升了数据处理的效率。
本发明所述的一种基于UNet的面波频散质量快速评估方法，步骤为：
步骤1、采集面波频散数据，对每个频散点人为标记合格标签，并将其分成训练集样本和测试集样本；
步骤2、构建面波频散质量评估的UNet模型，并利用训练集样本对UNet模型进行训练；；
步骤3、将面波频散的周期、面波频散、以及频散的变化率作为训练好的UNet模型的输入，得到质量评估后的面波频散的UNet预测结果。
进一步的，步骤1中，每个训练样本需要人为指定合格的区域(标签)，将合格的频
散点标记为1，不合格的标记为0。
进一步的，所述UNet网络模型左右各4层，一共包含15个卷积层，3个池化层，3个转置卷积层；选取RELU激活函数作用于卷积层的输出；卷积操作之后使用Batch_normilization对数据进行标准化。
进一步的，所述UNet模型的输入层包括三个通道：周期，面波频散，以及频散的变化率；输入层一共48个神经元，不足48个频散点的数据充零；输出层为每个周期点对应的标签，质量合格标签为1，否则标签为0；
为了能够更好的使输出层与标签匹配，对输出层作用于sigmod激活函数，使得最终输出神经元的值在0～1之间，sigmod激活函数的数学表达为：
σ(x)＝1/(1+exp(-x))
σ(x)＝1/(1+exp(-x))
其中，x为输入。
进一步的，UNet训练的损失函数需要极小化向量差的二范数：
其中，下标i表示第i个训练样本，N为训练样本的总数；D为由深度学习实现的预测结果；labeli为人为给定的标签，表征每个采样点数据的质量，di为输入频散数据，其包含三个通道；θ为网络反向传播中需要更新的超参数。
本发明所述的有益效果为：本发明提出了利用UNet来对地震面波频散数据进行快速评估，通过实际数据建立起训练集和测试集，然后构建了UNet模型实现对面波频散质量进行自动评估。本发明所述的方法简洁高效，可以在模型训练好之后进行快速预测，在短时间内完成批量频散数据的处理，实现对面波频散曲线的质量进行快速评估，剔除掉不符合要求的数据部分，保留下优质的数据。该方法不仅能够作为常规方法的替代，且相对于传统的人工识别方法极大地提升了数据处理的效率，为面波频散的实时反演奠定基础。
图1是采集面波频散数据的台站分布示意图；
图2是本发明所述的UNet网络模型结构示意图；
图3是面波频散的UNet预测结果示意图；
图4是测试集的预测结果示意图。
为了使本发明的内容更容易被清楚地理解，下面根据具体实施例并结合附图，对本发明作进一步详细的说明。
本发明所述的一种基于UNet的面波频散质量快速评估方法，具体步骤如下：
1.数据准备
为了得到面波频散数据，目前主要通过两个地震台站获得的地震事件或者噪声数据进行互相关。然后将工区划分为多个网格，使用走时层析成像技术可以得到每个网格上的频散曲线。本发明的面波频散通过噪声计算得到，一共采集了2013年12月至2015年3月
(15个月)的垂直分量连续时间序列，由668个宽带和超宽带便携式台站(中国地震台阵，二期)和62个永久宽带台站记录(如图1所示)。通过互相关等方法估计经验格林函数，最后提取出4160对频散数据用于质量评估的训练(Pan et al.，2019)。每个训练样本需要人为指定合格的区域(标签)，将合格的频散点标记为1，不合格的标记为0。其中的85％用于训练集，15％用于测试集。最后可以将筛选之后的频散结果进行层析成像，获得整个区域的频散。
2.UNet构建
面波频散的挑选需要对每一个频散数据的每一个频点质量进行判断，只有当某条频散曲线所有周期的数据都较差，才全部舍去；因此，面波频散的质量评估不是一个简单的二分类问题。这里选用UNet神经网络来实现接收函数的质量评估，模型的结构如图2所示。网络左右各4层，一共包含15个卷积层，3个池化层，3个转置卷积层。由于面波频散数据中没有负数，选取RELU激活函数作用于卷积层的输出，该激活函数能够提高网络的非线性能力，同时避免了训练过程中梯度爆炸和梯度消失问题。卷积操作之后使用了Batch_normilization对数据进行标准化，以进一步防止梯度消失或梯度爆炸现象，同时可以增加正则化效果。
输入层包括三个通道：面波频散的周期，面波频散，以及频散的变化率。输入层一共48个神经元，不足48个频散点的数据充零，这意味着网络适用于具有任意周期的低于48个频散数据的质量评估。输出层为每个周期点对应的标签，质量合格标签为1，否则标签为0。为了能够更好的使输出层与标签匹配，对输出层作用于sigmod激活函数，使得最终输出神经元的值在0～1之间，sigmod激活函数的数学表达为：
σ(x)＝1/(1+exp(-x))
σ(x)＝1/(1+exp(-x))
其中，x为输入。
UNet训练的损失函数，需要极小化向量差的二范数：
这里下标i表示第i个训练样本，N为训练样本的总数；D为由深度学习实现的预测结果；labeli为人为给定的标签，表征每个采样点数据的质量，di为输入频散数据，其包含三个通道；θ为网络反向传播中需要更新的超参数。
在训练过程中，使用反向传播算法Adam(Kingma and Ba，2014)来更新神经元的权值，学习率为0.001。根据目标函数随迭代次数的下降曲线，选择迭代70次以保证目标函数充分下降。
3.质量预测
模型训练好之后，将需要预测的数据按照训练集输入的格式准备好，即包括周期，频散数据以及频散的时间导数；然后导入UNet的输入层，可以快速得到质量评估结果，每个周期点的质量评估值位于0～1区间内。此时可以设定一个阈值(例如0.8)，高于该阈值保留频散值，反之删去频散值，值越高表明网络对质量的要求更高。最后可以得到质量较好的频散曲线，删去了质量较差的部分。
图3展示了三个面波频散数据的质量评估结果，其中a～d图分别展示了整体曲线
质量较好，中间部分质量较好、后段质量较好、整体曲线质量较差的评估结果。每个子图的上图表示预测质量较好的频散数据以及原始频散数据对比结果，下图表示预测结果，平直折线为人为给定标签，带竖棒的虚线表示UNet预测结果，越接近于1表示质量越好，竖棒长度表示不确定性。可以看出，UNet预测值较大的区域对应着相对连续，平滑的频散曲线部分，是合格的标准频散曲线形态。根据测试结果，设定阈值0.8可以较好地筛选出合格的频散曲线，即预测结果大于0.8被认为质量合格。总体上来看，UNet预测结果与人为给定的数据吻合良好，说明了UNet筛选面波频散曲线的有效性。但是在图3中，面波频散数据的变化趋势较大，存在较大的不确定性，UNet预测整段数据不符合质量要求，虽然与人为给定标签存在差异，但是也可以认为UNet的预测结果是合理的。图4显示了全部测试集的预测结果，其中，图4(a)为原始数据，图4(b)为UNet的预测结果，可以看出，UNet筛选后的频散曲线去掉了大部分高频的不连续曲线，整体上表现得更加平滑，渐变，更加符合实际情况。
这说明了本发明所提出的利用CNN评估面波频散质量是可行的，该方法可以在短时间内对面波频散曲线的质量进行快速评估，剔除掉不符合要求的数据部分，保留下优质的数据。该方法不仅可以替代人工挑选频散曲线，且极大的提高了面波频散数据的处理效率。
以上所述仅为本发明的优选方案，并非作为对本发明的进一步限定，凡是利用本发明说明书及附图内容所作的各种等效变化均在本发明的保护范围之内。
Claims (5)
- 一种基于UNet的面波频散质量快速评估方法，其特征在于，所述方法步骤为：步骤1、采集地震面波频散数据，对每个频散点人为标记合格标签，并将其分成训练集样本和测试集样本；步骤2、构建面波频散质量评估的UNet模型，并利用训练集样本对UNet模型进行训练；；步骤3、将面波频散的周期、面波频散、以及频散的变化率作为训练好的UNet模型的输入，得到质量评估后的面波频散的UNet预测结果。
- 根据权利要求1所述的一种基于UNet的面波频散质量快速评估方法，其特征在于，步骤1中，每个训练样本需要人为指定合格的标签，将合格的频散点标记为1，不合格的标记为0。
- 根据权利要求1所述的一种基于UNet的面波频散质量快速评估方法，其特征在于，所述UNet模型左右各4层，一共包含15个卷积层，3个池化层，3个转置卷积层；选取RELU激活函数作用于卷积层的输出；卷积操作之后使用Batch_normilization对数据进行标准化。
- 根据权利要求3所述的一种基于UNet的面波频散质量快速评估方法，其特征在于，所述UNet模型的输入层包括三个通道：周期，面波频散，以及频散的变化率；输入层一共48个神经元，不足48个频散点的数据充零；输出层为每个周期点对应的标签，质量合格标签为1，否则标签为0；为了能够更好的使输出层与标签匹配，对输出层作用于sigmod激活函数，使得最终输出神经元的值在0～1之间，sigmod激活函数的数学表达为：
σ(x)＝1/(1+exp(-x))其中，x为输入。 - 根据权利要求4所述的一种基于UNet的面波频散质量快速评估方法，其特征在于，UNet训练的损失函数需要极小化向量差的二范数：
其中，下标i表示第i个训练样本，N为训练样本的总数；D为由深度学习实现的预测结果；labeli为人为给定的标签，表征每个采样点数据的质量，di为输入频散数据，其包含三个通道；θ为网络反向传播中需要更新的超参数。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211411023.2 | 2022-11-11 | ||
| CN202211411023.2A CN115932959A (zh) | 2022-11-11 | 2022-11-11 | 一种基于UNet的面波频散质量快速评估方法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2024099422A1 true WO2024099422A1 (zh) | 2024-05-16 |
Family
ID=86649869
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2023/130903 WO2024099422A1 (zh) | 2022-11-11 | 2023-11-10 | 一种基于UNet的面波频散质量快速评估方法 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN115932959A (zh) |
| WO (1) | WO2024099422A1 (zh) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115932959A (zh) * | 2022-11-11 | 2023-04-07 | 电子科技大学长三角研究院(湖州) | 一种基于UNet的面波频散质量快速评估方法 |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490219A (zh) * | 2019-07-02 | 2019-11-22 | 中国地质大学（武汉） | 一种基于纹理约束的U-net网络进行地震数据重建的方法 |
| CN111626355A (zh) * | 2020-05-27 | 2020-09-04 | 中油奥博(成都)科技有限公司 | 一种基于Unet++卷积神经网络的地震数据初至拾取方法 |
| CN111766625A (zh) * | 2020-07-06 | 2020-10-13 | 中国科学技术大学 | 一种基于深度学习的地震背景噪声频散曲线的提取方法 |
| US20210311218A1 (en) * | 2018-08-10 | 2021-10-07 | University Of Houston System | Surface wave prediction and removal from seismic data |
| CN115184998A (zh) * | 2022-07-01 | 2022-10-14 | 长安大学 | 一种基于改进U-net神经网络的瑞利波频散曲线自动提取方法 |
| CN115932959A (zh) * | 2022-11-11 | 2023-04-07 | 电子科技大学长三角研究院(湖州) | 一种基于UNet的面波频散质量快速评估方法 |
-
2022
- 2022-11-11 CN CN202211411023.2A patent/CN115932959A/zh active Pending
-
2023
- 2023-11-10 WO PCT/CN2023/130903 patent/WO2024099422A1/zh unknown
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210311218A1 (en) * | 2018-08-10 | 2021-10-07 | University Of Houston System | Surface wave prediction and removal from seismic data |
| CN110490219A (zh) * | 2019-07-02 | 2019-11-22 | 中国地质大学（武汉） | 一种基于纹理约束的U-net网络进行地震数据重建的方法 |
| CN111626355A (zh) * | 2020-05-27 | 2020-09-04 | 中油奥博(成都)科技有限公司 | 一种基于Unet++卷积神经网络的地震数据初至拾取方法 |
| CN111766625A (zh) * | 2020-07-06 | 2020-10-13 | 中国科学技术大学 | 一种基于深度学习的地震背景噪声频散曲线的提取方法 |
| CN115184998A (zh) * | 2022-07-01 | 2022-10-14 | 长安大学 | 一种基于改进U-net神经网络的瑞利波频散曲线自动提取方法 |
| CN115932959A (zh) * | 2022-11-11 | 2023-04-07 | 电子科技大学长三角研究院(湖州) | 一种基于UNet的面波频散质量快速评估方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115932959A (zh) | 2023-04-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805269B (zh) | 一种基于lstm循环神经网络拾取震相到时的方法 | |
| WO2022160771A1 (zh) | 基于自适应多尺度特征提取模型的高光谱图像分类方法 | |
| WO2024099422A1 (zh) | 一种基于UNet的面波频散质量快速评估方法 | |
| CN111723329B (zh) | 一种基于全卷积神经网络的震相特征识别波形反演方法 | |
| CN110069886A (zh) | 基于vmd和cnn的电缆早期故障识别与分类方法 | |
| CN107506692A (zh) | 一种基于深度学习的密集人群计数与人员分布估计方法 | |
| CN109163997B (zh) | 一种基于声谱图深度学习的岩石表面强度测定方法 | |
| WO2023123568A1 (zh) | 一种探地雷达图像人工智能识别方法及装置 | |
| CN111160176A (zh) | 基于融合特征的一维卷积神经网络地面雷达目标分类方法 | |
| CN109711401A (zh) | 一种基于Faster Rcnn的自然场景图像中的文本检测方法 | |
| CN107609573A (zh) | 基于低秩分解和空谱约束的高光谱图像时变特征提取方法 | |
| CN114595732B (zh) | 基于深度聚类的雷达辐射源分选方法 | |
| CN108897042A (zh) | 有机质含量地震预测方法及装置 | |
| CN112732092B (zh) | 基于双视图多尺度卷积神经网络的表面肌电信号识别方法 | |
| CN111580151A (zh) | 一种基于SSNet模型的地震事件到时识别方法 | |
| CN108596044B (zh) | 基于深度卷积神经网络的行人检测方法 | |
| CN113687433A (zh) | 一种基于Bi-LSTM的大地电磁信号去噪方法及系统 | |
| CN109409644A (zh) | 一种基于改进的c4.5算法的学生成绩分析方法 | |
| Li et al. | Magnetotelluric noise suppression via convolutional neural network | |
| Arunsuriyasak et al. | Investigation of deep learning optimizer for water pipe leaking detection | |
| CN105787045A (zh) | 一种用于可视媒体语义索引的精度增强方法 | |
| CN110673210B (zh) | 一种地震原始数据信噪比定量分析评价方法 | |
| CN109978832B (zh) | 一种基于边缘重建的双绞线绞距检测方法 | |
| CN103577877B (zh) | 一种基于时频分析和bp神经网络的船舶运动预报方法 | |
| CN115171211A (zh) | 信道状态信息的动作和位置的联合估计方法 |