[ Instrument network instrument research and development ] Artificial neural networks (ANN) are developing rapidly in data-driven natural and information science research fields, such as image graphics, materials, biology and medicine, astronomy and geography, and earth sciences. In exploration geophysics, most of these studies can be regarded as visual image classification or segmentation problems. For example, geologists use seismic reflection data images to classify underground sedimentary units or oil and gas reservoirs and identify discontinuous geological structures such as faults, fractures, or salt bodies.
![]()
Artificial neural networks can learn morphological patterns in such images, and many of them are based on current convolutional neural networks (CNN), which is specifically designed for image-related tasks in computer vision. Compared with visual images, seismic reflection signals are essentially different: sparse signal polarity changes and limited bandwidth. The seismic response of geological features also differs in wave propagation path, frequency, amplitude, and direction of polarity. Therefore, the study of seismic interpretation based on data-driven ANN is a complex mapping problem of high-dimensional sparse signals.
Geng Zhi, a postdoctoral fellow at the Institute of Geology and Geophysics, Chinese Academy of Sciences, and co-supervisor and researcher Wang Yanfei, proposed a data-driven, seismic data classification, automatic search neural network architecture (SeismicPatchNet, SPN). Assuming that the key signal features embedded in the exploration seismic data can be captured by ANN, the parameters described are less than the CNN architecture. Researchers designed conceptual signal patches with specific seismic amplitude sequences. Taking marine gas hydrates as an example, these signals are similar to the key seismic reflections of hydrates. Various complex damage methods are considered to be applied to the above signals to generate Active dataset of the specific CNN architecture used for searching.
The neural network architecture maintains polarity information by decomposing the network core to reduce the amount of parameters and summarizing the opposite sampling features. The application of high-performance graphics processing unit (GPU) obtains the final network architecture SPN through inverse problem regularization modeling and random search algorithms . This research constitutes the first data-driven CNN designed with efficient computing capabilities, aiming to interpret seismic data end-to-end from the perspective of sparse signal processing.
The researchers found by comparing the international standard neural network model that the neural network architecture parameter storage capacity of the study is about 0.5% of the VGG-16 architecture; the prediction speed of SPN is nearly 18 times faster than ResNet-50, and it is identifying submarine gas hydrate resources The index shows advantages in terms of BSR. The saliency mapping shows that the architecture proposed in this study captures key features and shows the prospect of end-to-end interpretation of multiple seismic data sets at low computational cost.
Relevant research results were published on Nature Communications with the title Automated design of a convolutional neural network with multi-scale filters for cost-efficient seismic data classification. The research work is funded by the original innovation project from 0 to 1 of the Chinese Academy of Sciences, the key deployment project of the Institute of Geology and Earth, and the national key research and development plan project.
Mazda Android Car Dvd ,New
HD 1024X600 capacitive touch screen: The same touch screen found on your smart
phone,more responsive than ever
Mazda Android Car Dvd,Compatible with steering wheel controls:
skip to the next song, adjust the volume or switch another channel, all without
having to take your hands off the steering whell, making your journey safe and
more enjoyable.
Mazda Android Car Dvd,Support Full HD 1920X1080P: support various
formats including DAT, MPG, MPEG, VOB, MP4, MOV, AVI, DVD video, ASF,WMV, MKV,
TS, RM, RMVB, and video coding formats such as H.264, Xvid, MPEG2, MPEG4, VC-1,
WMV9, MPEG1, H.263, Divx.
Mazda Android Car Dvd can fit this car as follow:
|
KD-8069
|
CX-9
2012-2013
|
|
KD-7007
|
CX-7
2012-2013
|
|
KD-7005
|
CX-5
2012-2013
|
|
KD-7003
|
MAZDA 3
2004-2009
|
|
KD-8002
|
MAZDA 2
2010-2012
|
|
KD-8003
|
MAZDA 3
2009-2012
|
|
KD-8005
|
Mazada
5 2009-2012
Premacy 2009-2012
|
|
KD-8001
|
MAZDA
6 2008-2012
Mazda6 Ruiyi 2008-2012
Mazda6 Ultra 2008-2012
|
![1]( "1")
Mazda Android Car Dvd
Mazda Android Car Dvd,Mazda 5 Android Car Dvd,Mazda 3 Android Car Dvd,Mazda Android 4.1 Car Dvd
SHEN ZHEN KLYDE ELECTRONICS CO., LTD ,