CN111695529A - X-ray source detection method based on human skeleton key point detection algorithm - Google Patents
X-ray source detection method based on human skeleton key point detection algorithm Download PDFInfo
- Publication number
- CN111695529A CN111695529A CN202010544700.2A CN202010544700A CN111695529A CN 111695529 A CN111695529 A CN 111695529A CN 202010544700 A CN202010544700 A CN 202010544700A CN 111695529 A CN111695529 A CN 111695529A
- Authority
- CN
- China
- Prior art keywords
- ray source
- network
- source detection
- wide
- observation image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Abstract
The invention discloses an X-ray source detection method based on a human skeleton key point detection algorithm, which belongs to the technical field of discovery and identification of a celestial source and comprises the following steps: s10, constructing an X-ray source detection network based on the convolutional neural network; the X-ray source detection network comprises a positioning network, a local maximum extraction algorithm, a key point region feature extraction algorithm and an intensity prediction network which are sequentially arranged; s20, training the X-ray source detection network by using the wide-field observation telescope observation image training sample; and S30, inputting the observation image of the wide-view-field astronomical telescope to be identified into the trained X-ray source detection network to obtain the flow intensity predicted value of the observation image of the wide-view-field astronomical telescope to be identified. The method can eliminate noise from the observation image and extract data information of the source, supports the extension source, and has the source resolution capability of similar distances; meanwhile, the positioning accuracy and the recall rate on the test data can reach more than 99 percent.
Description
Technical Field
The invention belongs to the technical field of discovery and identification of celestial sources, and particularly relates to an X-ray source detection method based on a human skeleton key point detection algorithm.
Background
The existing X-ray source detection algorithm is mainly a sliding cell method, after a sliding cell is established, a pixel corresponding to the center position of the cell is used as a signal value to scan an observed image, meanwhile, a pixel at the position (8 neighborhoods) adjacent to the center point is marked as noise for calculating the signal-to-noise ratio, and if the signal-to-noise ratio calculated by the current sliding cell is larger than a given threshold value, the current position is marked as an X-ray source candidate position. And then subtracting the candidate positions of the X-ray source and a part of random background in the image, marking as a cheese (cheese) image, filling the subtracted positions by using an interpolation algorithm (usually bilinear interpolation) to generate a smooth background image, and finally repeating the scanning of the first step, wherein the interpolated background image is used for representing the noise intensity when calculating the signal to noise ratio, thereby giving an X-ray source list.
The sliding cell method is fast in execution speed and does not depend on the prior assumption, but the core problem is that the center pixel is used as a candidate source, so that the method is only friendly to point sources and high in confidence only on bright sources, and the method does not have good identification capability on many complex extended sources.
Therefore, there is an urgent need for an X-ray source detection method that can extract data information of a source from an observation image by eliminating noise, can support extended sources, and has a source resolution capability with a close distance.
Disclosure of Invention
The invention aims to provide an X-ray source detection method which can eliminate noise from an observed image and extract data information of a source, can support an extended source and has the source resolution capability close to the distance, and adopts the following technical scheme:
an X-ray source detection method based on a human skeleton key point detection algorithm comprises the following steps:
s10, constructing an X-ray source detection network based on the convolutional neural network; the X-ray source detection network comprises a positioning network, a local maximum extraction algorithm, a key point region feature extraction algorithm and an intensity prediction network which are sequentially arranged;
s20, training the X-ray source detection network by using a wide-field observation telescope observation image training sample; the training process comprises the following steps:
s21, inputting the wide-view-field observation image training sample of the astronomical telescope into the positioning network, and outputting a position probability heat map by the positioning network;
s22, inputting the position probability heat map into a local maximum extraction algorithm to obtain X-ray source coordinates;
s23, inputting the X-ray source coordinates into a key point region feature extraction algorithm to obtain an X-ray source region local map;
s24, inputting the regional local graph into the intensity prediction network to obtain a traffic intensity prediction value;
s25, when the flow intensity prediction precision meets the requirement, the training of the X-ray source detection network is completed;
and S30, inputting the observation image of the wide-view-field astronomical telescope to be recognized into the trained X-ray source detection network to obtain the predicted value of the flow intensity of the observation image of the wide-view-field astronomical telescope to be recognized.
Further, the key point region feature extraction algorithm is a human skeleton key point detection algorithm.
Further, the positioning network is a ResNet residual network, and the ResNet residual network comprises 4 ResBlock modules which are sequentially arranged.
Further, the regional local map is a 9 × 9 local image centered on the X-ray source.
Further, the strength prediction network comprises a dense all-connected layer, a relu activation layer, a dense all-connected layer and a sigmoid activation which are sequentially arranged.
Further, the positioning network adopts a multi-stage loss function supervision mode, and each stage refines the positioning result of the previous stage.
Further, the wide-field training image sample for the observation of the astronomical telescope comprises 80% of the training sample and 20% of the test sample.
The invention has the beneficial effects that:
the method can eliminate noise from the observation image and extract data information of the source, supports the extension source, and has the source resolution capability of similar distances; meanwhile, the positioning accuracy and the recall rate on the test data can reach more than 99%, the mean square error of the intensity prediction and the true value is about 0.006, the astronomical observation can be served with high accuracy, and the problem that useful information is difficult to effectively extract due to the fact that X-ray source observation data are excessively numerous and complicated in the big data background is solved.
Drawings
FIG. 1 is a diagram of the overall network architecture of the present invention
FIG. 2 is a schematic diagram of a positioning network
FIG. 3 is a schematic diagram of an intensity prediction network
FIG. 4 location probability heatmap of positioning network output
FIG. 59X 9 partial image
Detailed Description
Example 1
An X-ray source detection method based on a human skeleton key point detection algorithm comprises the following steps:
s10, constructing an X-ray source detection network based on the convolutional neural network; the X-ray source detection network comprises a positioning network, a local maximum extraction algorithm, a key point region feature extraction algorithm and an intensity prediction network which are sequentially arranged;
in this embodiment, the positioning network is a ResNet residual network, and the ResNet residual network includes 4 ResBlock modules (as shown in fig. 2) arranged in sequence; the positioning network adopts a multi-stage loss function supervision mode, and each stage refines the positioning result of the previous stage.
The strength prediction network comprises a dense all-connected layer, a relu activation layer, a dense all-connected layer and a sigmoid activation which are sequentially arranged (as shown in figure 3).
In this embodiment, the key point region feature extraction algorithm is a human skeleton key point detection algorithm.
S20, training the X-ray source detection network by using the wide-field observation telescope observation image training sample; the training process comprises the following steps:
s21, inputting the wide-view-field observation image training sample of the astronomical telescope into a positioning network, and outputting a position probability heat map (as shown in figure 4) by the positioning network;
s22, inputting the position probability heat map into a local maximum extraction algorithm to obtain X-ray source coordinates;
s23, inputting the X-ray source coordinates into a key point region feature extraction algorithm to obtain a region local graph;
in this embodiment, the regional local map is a 9 × 9 local image (see fig. 5) centered on the X-ray source.
S24, inputting the regional local graph into an intensity prediction network to obtain a traffic intensity prediction value;
in this embodiment, the predicted value in the intensity prediction process is the ratio of the intensity of the extension plane to the intensity of the center, thereby stabilizing the training process.
S25, when the flow intensity prediction precision meets the requirement, finishing the training of the X-ray source detection network;
in this embodiment, the training samples of the observation image of the wide-field telescope include 80% of the training samples and 20% of the test samples.
In the embodiment, 10000 wide-view-field observation images of the astronomical telescope are selected as training samples, wherein 8000 training samples and 2000 testing samples are included. And respectively training a positioning network and an intensity prediction network, wherein the data set of the intensity prediction network is a local graph corresponding to all the sources in the sample. When the two networks have stable test precision and then the training is completed, the positioning precision requirement in the example is as follows: the recall rate and the accuracy rate both reach 99 percent, and the flow intensity prediction precision requirement is as follows: the mean square error of the intensity prediction and the true value is 0.006.
And S30, inputting the observation image of the wide-view-field astronomical telescope to be identified into the trained X-ray source detection network to obtain the flow intensity predicted value of the observation image of the wide-view-field astronomical telescope to be identified.
The satellite is deployed on a telescope data processing computer, and takes an Einstein probe satellite as an example, and the satellite carries a wide-view-field sky-seeking telescope and a follow-up observation telescope with higher precision.
In the embodiment, the observation image of the wide-view-field astronomical telescope is used as input, the position and flow information of an X-ray source in an observation result is identified, the result is fed back to a subsequent X-ray source classification algorithm (not the content of the invention) to distinguish the source of the target type of interest designed by a scientific researcher in advance, and the subsequent observation telescope is used for high-precision tracking observation.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the technical scope of the present invention, so that any minor modifications, equivalent changes and modifications made to the above embodiment according to the technical spirit of the present invention are within the technical scope of the present invention.
Claims (7)
1. An X-ray source detection method based on a human skeleton key point detection algorithm is characterized by comprising the following steps:
s10, constructing an X-ray source detection network based on the convolutional neural network; the X-ray source detection network comprises a positioning network, a local maximum extraction algorithm, a key point region feature extraction algorithm and an intensity prediction network which are sequentially arranged;
s20, training the X-ray source detection network by using a wide-field observation telescope observation image training sample; the training process comprises the following steps:
s21, inputting the wide-view-field observation image training sample of the astronomical telescope into the positioning network, and outputting a position probability heat map by the positioning network;
s22, inputting the position probability heat map into a local maximum extraction algorithm to obtain X-ray source coordinates;
s23, inputting the X-ray source coordinates into a key point region feature extraction algorithm to obtain a region local graph;
s24, inputting the regional local graph into the intensity prediction network to obtain a traffic intensity prediction value;
s25, when the flow intensity prediction precision meets the requirement, the training of the X-ray source detection network is completed;
and S30, inputting the observation image of the wide-view-field astronomical telescope to be recognized into the trained X-ray source detection network, and obtaining the flow intensity predicted value of the observation image of the wide-view-field astronomical telescope to be recognized.
2. The X-ray source detection method of claim 1, wherein the keypoint region feature extraction algorithm is a human bone keypoint detection algorithm.
3. The X-ray source detection method according to claim 1, wherein the positioning network is a ResNet residual network, and the ResNet residual network comprises 4 ResBlock modules arranged in sequence.
4. The X-ray source detection method according to claim 1, wherein the regional local map is a 9X 9 local image centered on the X-ray source.
5. The X-ray source detection method of claim 1, wherein the intensity prediction network comprises a dense fully-connected layer, a relu activation, a dense fully-connected layer and a sigmoid activation arranged in sequence.
6. The method according to claim 1, wherein the positioning network adopts a multi-stage loss function supervision mode, and each stage refines the positioning result of the previous stage.
7. The X-ray source detection method of claim 1, wherein the wide-field roving telescope observation image training samples comprise 80% of training samples and 20% of test samples.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010544700.2A CN111695529B (en) | 2020-06-15 | 2020-06-15 | X-ray source detection method based on human skeleton key point detection algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010544700.2A CN111695529B (en) | 2020-06-15 | 2020-06-15 | X-ray source detection method based on human skeleton key point detection algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN111695529A true CN111695529A (en) | 2020-09-22 |
CN111695529B CN111695529B (en) | 2023-04-25 |
Family
ID=72481253
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010544700.2A Active CN111695529B (en) | 2020-06-15 | 2020-06-15 | X-ray source detection method based on human skeleton key point detection algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111695529B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1766591A (en) * | 2004-09-21 | 2006-05-03 | 通用电气公司 | System and method for an adaptive morphology X-ray beam in an X-ray system |
CN101166996A (en) * | 2005-02-28 | 2008-04-23 | 先进燃料研究股份有限公司 | Apparatus and method for detection of radiation |
CN102881003A (en) * | 2012-08-30 | 2013-01-16 | 暨南大学 | Method for removing cosmic rays in charge-coupled device (CCD) astronomic image |
CN103971127A (en) * | 2014-05-16 | 2014-08-06 | 华中科技大学 | Forward-looking radar imaging sea-surface target key point detection and recognition method |
US20150371387A1 (en) * | 2014-06-19 | 2015-12-24 | Qualcomm Incorporated | Local adaptive histogram equalization |
FR3047829A1 (en) * | 2016-02-12 | 2017-08-18 | Cie Nat Du Rhone | METHOD FOR ESTIMATING THE POSITION OF THE SOLAR DISK IN A SKY IMAGE |
CN109064478A (en) * | 2018-07-17 | 2018-12-21 | 暨南大学 | A kind of astronomical image contour extraction method based on extreme learning machine |
CN109683208A (en) * | 2019-01-25 | 2019-04-26 | 北京空间飞行器总体设计部 | A kind of adaptation space X radiographic source Accuracy Analysis method |
-
2020
- 2020-06-15 CN CN202010544700.2A patent/CN111695529B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1766591A (en) * | 2004-09-21 | 2006-05-03 | 通用电气公司 | System and method for an adaptive morphology X-ray beam in an X-ray system |
CN101166996A (en) * | 2005-02-28 | 2008-04-23 | 先进燃料研究股份有限公司 | Apparatus and method for detection of radiation |
CN102881003A (en) * | 2012-08-30 | 2013-01-16 | 暨南大学 | Method for removing cosmic rays in charge-coupled device (CCD) astronomic image |
CN103971127A (en) * | 2014-05-16 | 2014-08-06 | 华中科技大学 | Forward-looking radar imaging sea-surface target key point detection and recognition method |
US20150371387A1 (en) * | 2014-06-19 | 2015-12-24 | Qualcomm Incorporated | Local adaptive histogram equalization |
FR3047829A1 (en) * | 2016-02-12 | 2017-08-18 | Cie Nat Du Rhone | METHOD FOR ESTIMATING THE POSITION OF THE SOLAR DISK IN A SKY IMAGE |
CN109064478A (en) * | 2018-07-17 | 2018-12-21 | 暨南大学 | A kind of astronomical image contour extraction method based on extreme learning machine |
CN109683208A (en) * | 2019-01-25 | 2019-04-26 | 北京空间飞行器总体设计部 | A kind of adaptation space X radiographic source Accuracy Analysis method |
Non-Patent Citations (3)
Title |
---|
ZHUOXI HUO等: "Point source detection performance of Hard X-ray Modulation Telescope imaging observation" * |
王强等: "基于X射线的盒装水饺异物自动检测与分类" * |
马志贤等: "基于SVM的X射线天文图像点源探测算法" * |
Also Published As
Publication number | Publication date |
---|---|
CN111695529B (en) | 2023-04-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111950453B (en) | Random shape text recognition method based on selective attention mechanism | |
CN110610166B (en) | Text region detection model training method and device, electronic equipment and storage medium | |
CN111401307B (en) | Satellite remote sensing image target association method and device based on depth measurement learning | |
US10049441B2 (en) | Iterative defect filtering process | |
CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
CN111861978A (en) | Bridge crack example segmentation method based on Faster R-CNN | |
JPH06162196A (en) | Method and apparatus for comparison of image-data stream | |
CN110570433A (en) | Image semantic segmentation model construction method and device based on generation countermeasure network | |
CN112084920B (en) | Method, device, electronic equipment and medium for extracting hotwords | |
CN111861880A (en) | Image super-fusion method based on regional information enhancement and block self-attention | |
CN111462069A (en) | Target object detection model training method and device, electronic equipment and storage medium | |
CN112308973A (en) | Three-dimensional ocean element field intelligent construction method and system based on machine learning | |
CN112597852A (en) | Cell sorting method, cell sorting device, electronic apparatus, and storage medium | |
CN111695529A (en) | X-ray source detection method based on human skeleton key point detection algorithm | |
Ma et al. | CNN-TransNet: A Hybrid CNN-Transformer Network with Differential Feature Enhancement for Cloud Detection | |
CN113344148A (en) | Marine ship target identification method based on deep learning | |
Shi et al. | A photometry pipeline for SDSS images based on convolutional neural networks | |
CN116823793A (en) | Device defect detection method, device, electronic device and readable storage medium | |
CN115272242B (en) | YOLOv 5-based optical remote sensing image target detection method | |
CN114595770B (en) | Long time sequence prediction method for ship track | |
CN113295142B (en) | Terrain scanning analysis method and device based on FARO scanner and point cloud | |
Jia et al. | AE-TransUNet+: An Enhanced Hybrid Transformer Network for Detection of Lunar South Small Craters in LRO NAC Images | |
CN111435537B (en) | Model training method and device and pose optimization method and device based on mosaic | |
Guo et al. | EDD-Net: An efficient defect detection network | |
CN111435086B (en) | Navigation method and device based on splicing map |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |