36results about How to "Achieve precise extraction" patented technology

The invention discloses a method for full automatic extraction of water remote sensing information in a coastal zone and belongs to the field of methods for full automatic extraction of remote sensing information. The method comprises the steps of cutting a remote sensing image, roughly extracting water information and finely extracting water information, wherein scale conversion is performed twice in the process; the scale conversion performed at the first time is the scale conversion from bottom to up for converting an image element to an object; and the scale conversion performed at the second time is the scale conversion from up to bottom for converting the whole area to a local part. Compared with the prior art, the invention realizes a method combining geological knowledge with datamining in a scale conversion frame. The method is characterized by zero sample and zero parameter, and can be automatically operated completely. The method provided by the invention is suitable for various environments in a coastal zone of each sea area, has better stability, has higher precision of extraction results, has better extracting completeness and continuousness for detailed informationthan the traditional method, and can be directly applied to maintaining and updating the remote sensing thematic information of all-level national foundation geographic information databases.

The present invention provides a kind of microseismic effective signal detection method combining U-net network and DenseNet network, comprising: (1) generating original data set; (2) calibration data set; (3) constructing MSNet network; (4) adjusting Network parameters; (5) Calibrate the first arrival point after signal learning. The present invention also provides a microseismic effective signal detection system combining U-net network and DenseNet network, including: original data set generation module, data set calibration module, MSNet network construction module, network parameter adjustment module, after learning point calibration module. The present invention is based on U-net network learning and extraction of signal features of CNN network, combined with the DenseNet network, to achieve the purpose of further optimizing the learning and extraction of signal features. The invention has the characteristics of deeper feature extraction and finer signal segmentation in microseismic monitoring, and can be widely used in the field of underground state monitoring.

The invention provides a method for extracting elevation control points of building areas by satellite laser altimetry. Accurately extract Gaussian sub-waveform characteristic parameters and overall fitting; use high-resolution images or digital surface model (DSM) data, and Gaussian sub-wave half-width value parameters to select laser altimetry data containing contour plane layers within the spot range; Establish the corresponding relationship between the contour plane layer and its reflected Gaussian wavelet; use the precise distance of the center position of the sub-waveform characteristic parameters obtained by decomposition to locate the center of the spot, and calculate the absolute elevation of the corresponding contour plane layer; combine with high-resolution images , making remote sensing image elevation control points. The invention uses the full-waveform large-spot laser altimetry data as the remote sensing image elevation control point to expand the field from flat terrain to building areas and other areas containing contour plane layers.

The present invention relates to the field of visual recognition, and proposes a blind road detection method based on color pattern analysis and semantic segmentation, aiming to solve the problem that blind road detection is sensitive to changes in illumination, color and texture, and is not suitable for blind road areas in complex environments Accurate extraction of technical issues. The method includes: obtaining the color probability feature map of the blind road corresponding to the image to be detected; obtaining the probability feature map of the position of the blind road corresponding to the image to be detected by using a pre-built Gaussian diffusion model; The local texture feature map of the blind road, and obtain the feature fusion image according to the above blind road color probability feature map, the blind road position probability feature map and the above blind road local texture feature map, and perform semantic segmentation on the above feature fusion image to obtain the above blind road area in the image to be detected . The invention can quickly and accurately detect blind road areas.