CN115508800A - Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph - Google Patents
Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph Download PDFInfo
- Publication number
- CN115508800A CN115508800A CN202210998164.2A CN202210998164A CN115508800A CN 115508800 A CN115508800 A CN 115508800A CN 202210998164 A CN202210998164 A CN 202210998164A CN 115508800 A CN115508800 A CN 115508800A
- Authority
- CN
- China
- Prior art keywords
- extended
- phenomenon
- frequency
- screening
- picture
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/882—Radar or analogous systems specially adapted for specific applications for altimeters
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Multimedia (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention relates to a method and a system for screening an ionospheric frequency high-graph extended F phenomenon radar graph, wherein the method comprises the following steps: preprocessing a frequency-height image output by a height measuring instrument, inputting the preprocessed frequency-height image into a pre-established and trained extended F phenomenon radar pattern recognition model, obtaining a recognition result of whether an extended F phenomenon exists or not, and obtaining a corresponding extended F type; the extended F type includes: non-extension F, frequency-type FSF, region-type RSF, hybrid MSF and strong region-type SSF; the extended F phenomenon radar pattern recognition model is a resnet34Net network, an improved resnet34Net network or a residual _ attribution _ Net network, and is obtained by training through a supervised learning method. The invention can automatically carry out machine judgment on the occurrence and the type of the expansion F phenomenon in the Hainan altimeter frequency height diagram, provides an identification result and has extremely high accuracy of the judgment result.
Description
Technical Field
The invention belongs to the field of space target detection, relates to an automatic radar graph distinguishing method, and particularly relates to a method and a system for screening an ionospheric frequency hyperbologram extended F phenomenon radar graph.
Background
The ionosphere altimeter is a remote sensing device for detecting the ionosphere space environment based on radar echo, and the generated image data is called a frequency-altitude diagram (frequency-altitude diagram), and the change of electron density along with altitude is reflected by a clear trace line. When the ionosphere F layer (about 130 km to 1000 km, most spacecraft flight areas) is in a certain height range, the ionosphere F layer is not in a stable layer shape, but has some plasma fine structures (density inhomogeneities, or irregular bodies) which cause diffuse reflection to incident electric waves, and the incident electric waves are not clear linear traces but are dispersed pieces.
The spread F is a specific dispersion pattern generated in a frequency height map by the fact that the natural phenomenon of ionospheric plasma irregularities affects the propagation of radio waves, and the different forms correspond to different physical laws. The expanded F type classification with high international acceptance at present is the ionization map interpretation and measurement manual revised in 1978 by the international radio science union, and is classified into Frequency type (Frequency Spread F, FSF for short), range Spread F (RSF for short), hybrid type (MSF for short), and Branch type (BSF for short) according to the graphic features in the Frequency-height map, and proposes that each station can use its own targeted classification in view of the great difference of the phenomenon features of the expanded F of each station around the world. These 4 types, respectively, are: (1) The frequency type is a disturbance structure with clear tracing lines of an F layer in a low frequency band and expansion in a high frequency band, and corresponds to the height of a peak value of the F layer; (2) The region type is a structure with clear tracing lines of the F layer in a high frequency band, an expansion in a low frequency band and corresponding to uneven plasma density near the bottom of the F layer; (3) The hybrid type has the characteristics of both frequency type and region type, and the mechanism is relatively complex; (4) The bifurcation is that the F layer has an extension near the peak frequency and an extension F bifurcation different from the F layer trace, corresponding to the plasma structure with horizontal distribution and different density, which may be associated with ion deposition in the high latitude area.
In practice, researchers at the national space science center of the academy of Chinese sciences (abbreviated as the "Chinese academy space center") in the research subject group of the ionosphere consider that the extension F detected by a height gauge of a Hainan Fuke station (19.5 ° N,109.1 ° E) located in a low latitude area of China hardly appears in a manifold form, but a pattern exists in the extension from low frequency to high frequency, has small height change (less than 100 km) along with the increase of frequency, exceeds the peak frequency of the ionosphere, has a duration time exceeding half an hour, and is mostly divided into regional or mixed extension F in the past. Such patterns are mostly found in spring and fall 20-22%, with a probability of occurrence of more than 20% per year. Researchers believe that the ionospheric plasma bubble corresponds to a naturally-occurring large-scale structure called an ionospheric plasma bubble, and propose that the ionospheric plasma bubble is a classification type at low latitude, namely a Strong Range Spread F (SSF for short), and the researchers published academic papers and approved by a large number of international researchers.
Due to the scientificity and complexity of the frequency height map, the judgment can be only carried out by human eyes internationally. The method has a great defect in scientific research that human subjective judgment is mixed, the standards of expanded F type judgment of different researchers are different, and even the same researcher changes the judgment standards of expanded F graphic characteristics which change along with the year, season, local time and the like in the working process of the same researcher.
With the development of the aerospace technology in China, particularly the further construction of the second phase of meridian engineering involving the space center of Chinese academy of sciences, more than ten digital altimeters which are distributed in China and work for 24 hours are added in 2023, and the resolution is about 5-15 minutes, even high-precision detection networks with the resolution of 1 minute are developed by engineers. Under the condition, the conventional method of manually interpreting the frequency elevation map is not favorable for monitoring the space environment in real time, and the problem that all stations are identified by adopting a unified judgment standard in 24 hours manually cannot be solved, so that the development of an artificial intelligent identification method for the ionospheric frequency elevation map expansion F phenomenon is very necessary from the application point of view.
From the scientific research perspective, deep information related to a physical mechanism is hidden in a complex graph of an ionospheric frequency elevation map, and key information is screened by judging types, so that the method has important significance for researchers to research and expand the physical principle behind the F phenomenon.
Disclosure of Invention
The invention aims to overcome the problem that the traditional method can only be judged by human eyes internationally due to the scientificity and complexity of a frequency height map, develops automatic intelligent identification software, can automatically judge the occurrence and type of an extended F phenomenon in the frequency height map of a Hainan altimeter by a machine, gives an identification result and achieves the following aims:
(1) The real-time and automation of a scientific detection instrument of a ground ionosphere monitoring network are realized;
(2) The interference of subjective factors of personnel is eliminated by the unified machine standard;
(3) The method is helpful for scientific research personnel to research deep ionospheric characteristics reflected in an ionospheric frequency elevation map.
In order to achieve the above purpose, the present invention is realized by the following technical scheme.
The invention provides a method for screening an ionospheric frequency elevation map extended F phenomenon radar graph, which comprises the following steps:
preprocessing a frequency-height image output by a height measuring instrument, inputting the preprocessed frequency-height image into a pre-established and trained extended F phenomenon radar pattern recognition model, obtaining a recognition result of whether an extended F phenomenon exists or not, and obtaining a corresponding extended F type; the extended F type includes: non-extension F, frequency-type FSF, region-type RSF, hybrid MSF and strong region-type SSF;
the extended F phenomenon radar pattern recognition model is a resnet34Net network, an improved resnet34Net network or a residual _ attention _ Net network, and is obtained by training through a supervised learning method.
As an improvement of the above technical solution, the pretreatment comprises:
unifying the coordinates, pixels and formats of the frequency and height image output by the altimeter according to the setting requirements;
cutting the picture, removing the surrounding information part and coordinate axis information, and reserving the main part of the middle-frequency high-image of the picture;
and stretching the picture to standardize pixels.
As one improvement of the above technical solution, the ResNet34Net network model adopts a classical ResNet34 model, in the last layer convolution sum operation, the arithmetic mean obtained by every 4 pixels is changed into an adaptive arithmetic mean, and in the aspect of the learning rate, 100 epochs reduce the learning rate to 1/5 every 20 epochs;
the improved resnet34Net network sets 200 epochs in the learning rate, and reduces the learning rate to 1/2 every 20 epochs; integrating the convolution of the step 2 and the maximized down-sampling, and replacing the convolution with 4 x 4 and step 4; the block ratio is adjusted from 3;
the residual _ attribution _ Net network is a residual attention model, and in terms of learning rate, the learning rate is reduced to 1/5 every 25 epochs.
As an improvement of the above technical solution, the method further comprises a training step of expanding the F phenomenon radar pattern recognition model; the method specifically comprises the following steps:
acquiring a frequency height map output by an ionosphere altimeter, and carrying out standardized preprocessing on the image;
marking and recording each preprocessed picture, classifying and sampling the pictures according to different expansion F type characteristics by using labels given to the pictures by the record, and performing classification and calibration; the classification method adopted here is based on non-extension F, frequency type FSF, regional type RSF, mixed type MSF and manifold type BSF with higher international acceptability, and is a pattern feature description classification based on the position of dispersion extension in a frequency height diagram relative to an ionosphere F layer trace, and other subdivision on the basis, such as the strong regional type SSF used in the method, is within the same principle of the classification method;
down-sampling or up-sampling the pictures of each category to enable the pictures to reach the set number respectively, and distributing a training set and a test set;
respectively establishing identification models based on a resnet34Net network, an improved resnet34Net network or a residual _ attribution _ Net network model, and extracting and identifying the features of the extended F type;
and training the recognition model by using a training set to obtain a trained extended F phenomenon radar pattern recognition model.
As one improvement of the technical scheme, the acquired frequency height diagram output by the ionosphere altimeter comprises the years of high, medium and low solar activity and comprises all seasons and local time; meanwhile, the frequency elevation map includes image data continuously acquired at set time intervals.
As one improvement of the above technical solution, when classifying the different extended F-type features, the features of the front and rear ionization maps are referred to in a manual discrimination manner, and continuity of an ionospheric phenomenon is considered, and then the determination and classification are performed;
if the pattern in the picture is fuzzy or has dispersion similar to the expansion F phenomenon or has individual blank patterns due to ionosphere absorption, whether the expansion F phenomenon really occurs needs to be judged, which specifically comprises the following steps:
if the graph of the picture is related to the characteristics of the place, classifying and sampling the picture;
if the picture has a complete blank pattern which is continuous for tens of hours or even longer, the picture is not classified and sampled.
As an improvement of the above technical solution, in the up-sampling process, the picture samples are extended by adding one or more noises in the picture; the noise is Poisson noise, gaussian noise, salt and pepper noise, line salt and pepper noise or salt and pepper noise.
The invention also provides a system for screening the ionospheric frequency elevation map extended F phenomenon radar graph, which is used for identifying and screening the extended F type of the ionospheric frequency elevation map based on one of the above methods for screening the ionospheric frequency elevation map extended F phenomenon radar graph, and the system comprises: the system comprises a data import module, a data preparation module and an extension identification module; the data import module is used for importing a frequency height map which is output by the altimeter and needs to identify the type of the expanded F phenomenon;
the data preparation module is used for preprocessing the imported frequency height map and inputting the preprocessed frequency height map into the expansion identification module;
and the extended identification module is established based on a trained extended F phenomenon radar pattern identification model and is used for identifying and screening the type of the extended F phenomenon of the imported frequency height map.
As an improvement of the above technical solution, the system further comprises:
the data output module is used for displaying and exporting the screened results;
the display form of the screening results comprises: "file by" and "time by":
representing a picture and a recognition result corresponding to the picture according to each line of the file;
"time by time" is to show the results in terms of the time period over which the extended F phenomenon occurs;
in response to this, the mobile terminal is allowed to,
when the screening result is exported according to the file, each row comprises the path and the identification result of the picture;
when deriving the screening results "in time", each row contains an extended period of occurrence of F.
As an improvement of the above technical solution, the system further comprises:
the recognition result checking module is used for checking and correcting the displayed screening results and comprises: the picture name, the recognition result, the non-zoomed picture and the sequence number currently recorded in the 'by file' list are checked, and if the recognition result has a problem, the recognition result is corrected by reselecting a result.
The invention has the technical effects that:
from the view of test results and software functions of the invention, based on the ionosphere frequency height map expansion F phenomenon of Hainan altimeter data, the artificial intelligence identification software developed by the invention can automatically perform machine judgment on the occurrence and type of the expansion F phenomenon in the Hainan altimeter frequency height map, give an identification result, and the judgment result has extremely high accuracy and is close to the artificial judgment of professional researchers; the software application is convenient and fast, and the real-time and automation of the scientific detecting instrument of the ground ionosphere monitoring network can be realized. The software can greatly save manpower of scientific research and operation control personnel, can obviously improve the working efficiency of scientific instruments, eliminates interference of subjective factors of personnel through unified machine standards, is also helpful for the scientific research personnel to research deep ionospheric characteristics reflected in an ionospheric frequency elevation diagram, and has scientific and application prospects.
Compared with the prior art, the invention has the advantages that:
(1) The classification method is original and belongs to the international scientific research frontier;
(2) The latest image depth convolution model is applied to the use of ionosphere digital altimeter data for the first time;
(3) The automatic discriminant analysis of the physical phenomenon of ionosphere expansion F is realized for the first time.
Drawings
FIG. 1 is a flow chart of a method for screening an ionospheric frequency histogram extended F phenomenon radar pattern according to the present invention;
FIG. 2 is a functional block diagram of the system architecture of the present invention;
FIG. 3 is a diagram of the "by file" recognition result of the system of the present invention;
FIG. 4 is a graph of "by time" recognition results for the system of the present invention.
Detailed Description
The technical scheme provided by the invention is further illustrated by combining the following embodiments.
Example 1
Fig. 1 is a flowchart of a method for screening an ionospheric frequency elevation map extended F phenomenon radar pattern according to embodiment 1 of the present invention.
Example 2
The method for screening the ionospheric frequency elevation map extended F phenomenon radar graph according to embodiment 1 of the present invention identifies and screens the extended F type of the frequency elevation map, and the system includes: the system comprises a data import module, a data preparation module, a data output module, an identification result checking module and an expansion identification module;
the data import module is used for importing a frequency height map which is output by the altimeter and needs to identify the type of the F phenomenon;
the data preparation module is used for preprocessing the imported frequency height map and inputting the preprocessed frequency height map into the expansion identification module;
and the extension identification module is established based on the trained extension F phenomenon radar pattern identification model and is used for identifying and screening the extension F phenomenon types of the imported frequency height map.
The data output module is used for displaying and exporting the screened results;
the display form of the screening results comprises: "file by" and "time by":
representing a picture and a recognition result corresponding to the picture according to each line of the file;
"time by time" is to show the results in terms of the time period over which the extended F phenomenon occurs;
in response to this, the mobile terminal is allowed to,
when the screening result is exported according to the file, each row comprises the path and the identification result of the picture;
when deriving the screening results "in time", each row contains an extended F epoch.
The identification result checking module is used for checking and correcting the displayed screening results and comprises the following steps: the picture name, the recognition result, the picture without zooming and the sequence number currently recorded in the 'per file' list are checked, and if the recognition result has a problem, the recognition result is corrected by reselecting a result.
Example 3
Fig. 2 is a block diagram of the modules of the automatic intelligent recognition software developed according to the method or system of the present invention in embodiment 3.
In order to overcome the defects of the existing analysis technology of the frequency-height diagram extended F phenomenon of the ground altimeter on the outer space (ionized layer) environment monitoring data, the invention provides frequency-height diagram extended F phenomenon artificial intelligent recognition software based on machine learning, a deep convolution network is used for monitoring and learning the graphs of the altimeter of the Hainan Fuke station (19.5 degrees N and 109.1 degrees E) which is calibrated manually, the characteristics of the natural phenomenon in the data graphs are extracted, automatic intelligent recognition software is developed, the occurrence and the type of the extended F phenomenon in the frequency-height diagram of the Hainan altimeter can be judged by machines, and a recognition result is given.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
(1) Image normalization preprocessing: the invention is based on the graphic characteristics in the ionospheric frequency height map, so the frequency height map as a sample needs standardization processing firstly to ensure that the picture size is consistent with the horizontal and vertical coordinates.
(2) Generating a sample library with manual calibration results: researchers have manually marked and recorded each picture of an altimeter of Hainan Fuke station (19.5 degrees N,109.1 degrees E) in 2002-2015 based on own experience, and the picture labels given to (1) by using the records are divided into a non-extended F (marked 0), a frequency type FSF (marked 1), a region type RSF (marked 2), a mixed type MSF (marked 3) and a strong region type SSF (marked 4), and are subjected to secondary proofreading so as to meet the definition of the characteristics of the standard extended F types. After calibration, a total of 46,5493 samples were obtained.
Extracting a training set and a test set: the obtained sample set has the problem of uneven sample distribution, the number of the labels 0 is 426,555, and the labels 1-4 are all thousands to 1 ten thousand, so 20,000 are selected as the final number of the images of each category, the label 0 is subjected to down-sampling, the labels 1-4 are subjected to up-sampling, 5 categories in total are obtained, 100,000 picture samples in total are randomly distributed into a training set and a testing set according to the proportion of 8.
In the up-sampling process, in view of the characteristics of the height indicator scientific picture, the original image is reserved, and the number of sample images in different classes is increased to 20,000 by adding Poisson noise, gaussian noise, salt and pepper noise, row salt and pepper noise, column salt and pepper noise and the like in the image.
(3) Feature extraction and recognition of the normalized extended F phenomenon: and after an effective sample data set is obtained, training by adopting a supervised learning method.
Machine learning adopts Python language, and various different types of classification networks such as VGG16, VGG19, resNet34, resNet50, efficiency NetV2, convNeXt, resNet34_ Attention and ResNet34_ Modified are adopted for processing.
After training parameters are obtained by machine learning of a training set (8 ten thousand samples), 3 models with the best accuracy are selected from results obtained by testing the testing set (2 ten thousand samples), namely ResNet34_20 \\ 5 (rescet 34 Net) (93.20%), resNet34-modified-20-2-200 (93.50%), residual _ identification _ Net _ old _25 \5 (last _ model _92 \ sgd_25 \5. Pkl) (93.53%). And packaging the software by using the trained weight parameters.
The invention has the beneficial effects that: by utilizing the ionosphere expansion F phenomenon types which are manually identified by scientific researchers based on self experiences in 2002-2015 diagrams and utilizing an artificial intelligence technology to develop automatic intelligent identification software, the occurrence and the types of the expansion F phenomenon in the Hainan altimeter frequency height diagram can be automatically judged by a machine, an identification result is given, the accuracy rate exceeds 93%, and the effect is close to that of manual judgment of the scientific researchers with abundant experiences. The method can realize real-time and scale automation of scientific detection instruments of the ground ionosphere monitoring network, has uniform judgment standards, and eliminates interference of subjective factors of personnel. The machine judgment standard extracted in the research is helpful for scientific researchers to research deep ionospheric characteristics reflected in an ionospheric frequency elevation diagram.
The invention realizes the real-time and scale automation of the height indicator monitoring extension F by monitoring and learning different types of extension F phenomena in the frequency height map data of a scientific instrument, namely an ionosphere height indicator and using the obtained weight parameters for artificial intelligence identification.
The invention adopts the following technical scheme:
(1) Image normalization pre-processing
The method comprises the following steps that a frequency-height map is derived from an original file of the altimeter by SAO-X software, coordinates are unified into 0-17MHz and 90-800km, and a standard PNG picture is 700 pixels wide and 600 pixels high;
cutting the picture, removing the surrounding information part and coordinate axis information, and reserving the main part of the middle-frequency and high-frequency picture in the picture, namely taking the pixels of each picture from the upper left corner (150, 61) to the lower right corner (645, 520);
the pixels are stretched to specification 448 x 448 pixels, which facilitates the processing of subsequent classification nets (the existing nets are dominated by the images processed 224 x 224).
(2) Generating a sample library with artificial calibration results, extracting a test set and a training set
(2-1) generating a sample library with manual calibration results:
researchers have manually marked and recorded each picture of altimeters at hokkiso station (19.5 ° N,109.1 ° E) in hainan, 2002-2015 based on their own experiences, and given picture labels of (1) by using the records, the pictures are classified into non-extended F (marked 0), frequency-type FSF (marked 1), region-type RSF (marked 2), hybrid-type MSF (marked 3), and strong region-type SSF (marked 4).
However, due to the nature of scientific research, during recording, researchers are used to consider the same type of complete event, namely, a space weather event which is continuous for hours at a time, and if the characteristics of dozens of and dozens of images in the period are ambiguous, the images are uniformly marked as the same type. However, this work requires the samples to have sharp features, so the manual secondary classification calibration is performed according to the standard extended F-type feature definition. The various sample sizes after recalibration are shown in the following table:
table 1 labels the number of samples of each type of extension F in the sample library
When samples are classified, based on scientific principles and experience, the following special cases and treatment methods exist:
the influence of season, year, solar activity week, etc. on the ionospheric parameter characteristics:
according to the statistics in the past, the incidence of different types of ionospheric expansion F changes with the place and the season, and the characteristics of the graph change due to the change of the foF2 with the time.
The data used for training and validation in this work was more than one solar week (11 years), and thus included the years of high, medium and low solar activity, and included data for one piece (24 consecutive hours) every 15 minutes/5 minutes for all seasons and places. Thus, all the varying factors and results are covered in the process of unified machine learning.
Sporadic E-layer (Es-layer for short) effects:
the Es layer is a natural phenomenon in the ionosphere around 100 km in height, is thin (about several hundred meters), has high electron density, and is displayed as a flat straight line around 100 km in a frequency height diagram of a height finder, sometimes accompanied by some similarly extended graphs.
Traces of the Es layer may obscure extended F patterns of the F layer at a height of about two to three hundred kilometers, or multiple echoes of the Es layer and the ground appear at a height of several hundred kilometers and overlap with traces of the F layer of the ionosphere, or cause dispersion of echo signals of the altimeter at a lower frequency band (about 1 to 3 Mhz), and these patterns interfere with identification of the extended F phenomenon.
For the phenomena, the characteristics of front and back ionization maps are referred as much as possible during manual judgment, the continuity of the ionosphere phenomenon is considered, judgment and classification are carried out, and therefore the reliability of training samples is improved.
Influence of instrument detection capability:
when the foF2 is very small (about 1-1.5MHz, mainly in the local morning), the pattern may be blurred or appear to be dispersed like the spread F due to factors such as the echo signal-to-noise ratio, even a separate blank pattern appears due to ionosphere absorption, and whether the spread F really appears is judged according to experience. Such patterns relate to features of the ionosphere when in different places, yet should incorporate machine-learned samples.
Some completely blank patterns, which are continuous for tens of hours or even longer, are related to instrument maintenance and the like, and are not included in machine learning samples.
Multiple echoes, noise and other effects:
radio waves reflected by an ionized layer are reflected by the ground and are reflected by the ionized layer again, and the radio waves are received by the altimeter antenna, so that multiple echo patterns can appear at integral multiple heights of the corresponding ionized layer height in a frequency height diagram.
In addition, noise patterns such as horizontal lines, vertical lines, color blocks, and the like often appear in echo patterns of a height finder using the radio wave radar principle.
When the samples of the phenomena appear, manual judgment is carried out according to experience during marking, the characteristics of the models can be mastered through machine learning, and finally the interferences are eliminated as much as possible.
(2-2) extracting a test set and a training set:
after obtaining the sample library shown in table 1, it can be seen that the number of each type of image data is different, and in order to avoid the uneven distribution of samples (the machine learning is not friendly to the class of small samples), 20,000 is selected as the final number of images of each class.
Samples of others (no extension) were downsampled: 426,555 samples, generating random numbers of 0 and 1, and selecting 20,000 png images corresponding to 1 as final samples of the category;
and (3) respectively upsampling the samples of the 4 types of the extended F, namely, keeping the original images in the images of all the types, and respectively adding Poisson noise, gaussian noise, salt and pepper noise, row salt and pepper noise, column salt and pepper noise and the like into the images according to the characteristics of the ionization maps to increase the number of the sample images of different types to 20,000.
Assigning a training set and a test set: and randomly dividing the 5 classes of 100,000 photos into a training set and a testing set according to the proportion of 8.
(3) The characteristic extraction and identification of the normalized extended F phenomenon are carried out, and an identification model is established
The recognition model is the technical core for guaranteeing the business indexes of the project, and recognition requirements are realized through artificial marking data provided by scientific research personnel.
The key point of machine supervision learning in the work is to extract a key area reflecting the characteristic of the expansion F phenomenon from a picture marked artificially, judge the layout or trend of data points and summarize the rule. And training the obtained weight parameters for the test set and the automatic identification process of the new data in the future.
Considering that the analysis is based on images, machine learning is mainly based on a deep convolution neural model, and a ViT model is adopted according to the experimental result;
due to the characteristics and the location of the 5 extended F types (4 occurring in hainan fuke station), the basic area of their diffusion pattern distribution is determined, and therefore a suitable attention model is sought.
For the case of sample distribution and its imbalance, where the ratio of positive and negative samples exceeds 1.
(3-1) after the training parameters are obtained by machine learning in the training set (8 ten thousand samples), and the training parameters are tested in the testing set (2 ten thousand samples), the effects of the obtained models are shown in table 2:
TABLE 2 basic method and judgment accuracy of selected models
As shown in the table, the 3 models with the best accuracy are ResNet34_20_5_100 (rescet 34 Net) (93.20%), RESNet34-modified-20-2-200 (93.50%), residual _ identification _ Net _ old _25_5 (last _ model _92_sgd _25_5. Pkl) (93.53%), respectively. Compared with the result of human eye judgment by scientific researchers, the result of the automatic judgment extension F has an accuracy rate of over 93 percent.
The 3 models are realized in a specific way:
1)ResNet34_20_5_100(resnet34Net)
in the aspect of learning rate, 100 epochs reduce the learning rate to 1/5 every 20 epochs;
the model as a whole is a classical ResNet34 model, and in the final layer convolution sum operation, the arithmetic mean of every 4 pixels is changed into an adaptive arithmetic mean.
2)resNet34-modified-20-2-200
In the aspect of learning rate, 200 epochs are used, and the learning rate is reduced to 1/2 every 20 epochs;
the convolution of step 2 and the maximum downsampling are integrated and replaced by the convolution of 4 x 4 and step 4;
the block ratios are adjusted from 3.
3)residual_attention_Net_old_25_5
In the aspect of learning rate, the learning rate is reduced to 1/5 every 25 epochs;
belonging to the residual attention model, machine learning focuses on the changed part.
Due to the difference of the complexity of the graphs, taking the training result of the Attention residual error ResNet34_ Attention model as an example, the overall accuracy of the model in the test set is 93.525% (the recognition capability is basically consistent with the judgment of a human):
the accuracy of the judgment of no extension F is 96%, namely, whether the ionosphere has a disturbance phenomenon (extension F) can be accurately identified;
the accuracy of frequency type FSF judgment is 93%, the accuracy of region type RSF judgment is 99%, the two types are the most basic and most researched types of all station extension F in the world, and the model is accurate in judgment;
the accuracy rate of judgment on the strong region type SSF is 98%, the type of expanded F is a natural phenomenon which is particularly important in low-latitude areas of China, such as Hainan and the like, has strong correlation with equatorial plasma bubbles, can strongly interfere with the stability of electromagnetic wave signals, and is one of the most concerned science and application, and the model is very accurate in judgment;
the accuracy of the mixed type MSF determination is 82%, which contains a large number of figures with insignificant and ambiguous features of other types of extended F, so that the accuracy is reasonably low, and 82% accuracy is within an acceptable range.
(3-2) to verify the accuracy and validity of these 3 models in terms of automation, further verification was performed using the frequency and height pattern of the Hainan Fuke altimeter in 2013-2016.
With over 4 million 6 thousand samples each year 2013-2015, only a few previous upsamples and downsamples enter the training set and sample set of machine learning. All samples in the whole year are tested (mainly to check whether the phenomenon of F is mistakenly identified or not), the inference results of the 3 models on the sample data are compared with the GT-SF results, and the results show that the annual comparison difference in 2013 is about 2.5%, the annual comparison difference in 2014 is about 2.3%, the annual comparison difference in 2015 is about 5.2, and the consistency is very good.
The 2016 data was not identified and identified manually and was not involved in machine learning. The parameters trained by the 3 models are used for automatically judging 2016 year-round samples, and the difference of the 3 model reasoning result comparison is within 5%.
In conclusion, tests show that the machine learning 3 models have excellent accuracy and automation for automatically judging the ionospheric frequency height map expansion F phenomenon of Hainan altimeter data, can greatly improve the ionospheric space environment monitoring efficiency of the altimeter, and have practical application prospects.
Therefore, the invention utilizes the trained weight parameters of the above 3 models to package and manufacture intelligent recognition software, and can be directly used in real working scenes. The modular function of the software is shown in figure 2.
In use, the original PNG picture output by the altimeter can be directly imported, the import is divided into single file import and folder-data time range selection, and the picture file needing to be judged is directly read according to the date and time in the file name;
confirm the selected weight file (corresponding to 3 models), click the "identify" button, the system pops up the progress bar and starts running, and after testing, 100 pictures are selected for the average processing time: in the GPU mode, the model loading time is 3226ms, the model identification average time is 80.6ms, in the CPU mode, the model loading time is 795ms, and the model identification average time is 439.4ms;
-after the operation is finished, a statistical result, namely how many pictures each of the type extensions F occupies, is skipped;
the results show two "per file" and "per time" tabs in the bottom half of the software:
each line of "per file" represents a file and the recognition result, as shown in fig. 3;
"time by" is to show the results in terms of the time period over which the extended F phenomenon occurs, as shown in fig. 4.
-the software supports checking and correcting the recognition result, clicking the result directly opens a dialog box showing the name of the picture, the recognition result, the picture without scaling and the serial number currently recorded in the "by file" list, if the user feels that the recognition result is a problem, the recognition result can be corrected, selecting a new result from the recognition result list, clicking "correction", the software pops up a confirmation box, and after clicking yes, the result is updated;
the user can export the recognition result to a text file, click the "export" button on "press file" or "press time" tab, and after popping up the dialog box to select the saved location and file name, the system starts execution and prompts the dialog box to complete;
an export file of 'per file' tab, wherein each line comprises a path of the picture and an identification result;
the export file of "by time" tab, each row contains an occurrence period of the extension F.
As can be seen from the above detailed description of the present invention, the artificial intelligence recognition software developed by the present invention can automatically perform machine judgment on the occurrence and type of the extended F phenomenon in the hainan altimeter frequency height map, and provide a recognition result with extremely high accuracy.
Finally, it should be noted that the above embodiments are only used for illustrating the technical solutions of the present invention and are not limited. Although the present invention has been described in detail with reference to the embodiments, it will be understood by those skilled in the art that various changes may be made and equivalents may be substituted without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A method for screening ionospheric frequency elevation map extension F phenomenon radar patterns is characterized by comprising the following steps:
preprocessing a frequency-height image output by a height measuring instrument, inputting the preprocessed frequency-height image into a pre-established and trained extended F phenomenon radar pattern recognition model, obtaining a recognition result of whether an extended F phenomenon exists or not, and obtaining a corresponding extended F type; the extended F type includes: non-extension F, frequency-type FSF, region-type RSF, hybrid MSF and strong region-type SSF;
the extended F phenomenon radar pattern recognition model is a resnet34Net network, an improved resnet34Net network or a residual _ attention _ Net network, and is obtained by training through a supervised learning method.
2. The method of screening ionospheric frequency elevation map extended F phenomenon radar plots according to claim 1, wherein said preprocessing comprises:
unifying the coordinates, pixels and formats of the frequency height map images output by the altimeter according to the setting requirements;
cutting the picture, removing the surrounding information part and coordinate axis information, and reserving the main part of the medium-frequency high-image in the picture;
and stretching the picture to standardize pixels.
3. The method for screening ionospheric frequency histogram extension F phenomenon radar patterns according to claim 1, wherein the renet 34Net network model adopts a classical ResNet34 model, and in a final layer convolution sum operation, an arithmetic mean of every 4 pixels is changed into an adaptive arithmetic mean, and in terms of a learning rate, 100 epochs are used, and the learning rate is reduced to 1/5 every 20 epochs;
the improved resnet34Net network sets 200 epochs in terms of learning rate, and reduces the learning rate to 1/2 every 20 epochs; the convolution of step 2 and the maximum downsampling are integrated and replaced by the convolution of 4 x 4 and step 4; adjusting the block ratio from 3;
the residual _ attention _ Net network is a residual attention model, and the learning rate is reduced to 1/5 every 25 epochs in the learning rate.
4. The method for screening ionospheric frequency elevation map extended F-phenomenon radar patterns according to claim 1, further comprising a training step of an extended F-phenomenon radar pattern recognition model; the method specifically comprises the following steps:
acquiring a frequency height map output by an ionosphere altimeter, and carrying out standardized preprocessing on the image;
marking and recording each preprocessed picture, classifying and sampling the pictures according to different expansion F type characteristics by using labels given to the pictures by the record, and performing classification and calibration;
down-sampling or up-sampling the pictures of each category to enable the pictures to reach the set number respectively, and distributing a training set and a test set;
respectively establishing recognition models based on a resnet34Net network, an improved resnet34Net network or a residual _ attribution _ Net network model, and extracting and recognizing the features of the extended F type;
and training the recognition model by using a training set to obtain a trained extended F phenomenon radar pattern recognition model.
5. The method for screening ionospheric frequency height map extended by F phenomenon radar graphs as claimed in claim 4, wherein said acquired ionospheric frequency height map output by the ionospheric altimeter includes the years of high, medium and low solar activity, and includes all seasons and local times; meanwhile, the frequency elevation map includes image data continuously acquired at set time intervals.
6. The method for screening the ionospheric frequency elevation map extended F phenomenon radar pattern according to claim 4, wherein when classifying different extended F type features, the method of manual discrimination is adopted, and the features of the ionization maps before and after are referred to, and the continuity of the ionospheric frequency elevation map is considered, and then the judgment and classification are performed;
if the pattern in the picture is fuzzy or dispersion similar to the phenomenon of the expanded F appears or a separate blank pattern appears due to ionosphere absorption, whether the phenomenon of the expanded F appears really needs to be judged, which specifically comprises the following steps:
if the graph of the picture is related to the characteristics of the place, classifying and sampling the picture;
if the picture has a complete blank pattern which is continuous for tens of hours or even longer, the picture is not classified and sampled.
7. The method of claim 4, wherein in the up-sampling process, the image samples are extended by adding one or more noises to the images; the noise is Poisson noise, gaussian noise, salt and pepper noise, line salt and pepper noise or salt and pepper noise.
8. A system for screening ionospheric frequency elevation map extended F phenomenon radar graphs, which is used for identifying and screening extended F types of an ionospheric frequency elevation map based on the method for screening ionospheric frequency elevation map extended F phenomenon radar graphs as claimed in any one of claims 1 to 7, the system comprising: the system comprises a data import module, a data preparation module and an extension identification module; the data import module is used for importing a frequency height map which is output by the altimeter and needs to identify the type of the F phenomenon;
the data preparation module is used for preprocessing the imported frequency height map and inputting the preprocessed frequency height map into the expansion identification module;
and the extended identification module is established based on a trained extended F phenomenon radar pattern identification model and is used for identifying and screening the extended F phenomenon types of the imported frequency-height diagram.
9. The system for screening ionospheric frequency elevation map extended F phenomenon radar plots of claim 8, further comprising:
the data output module is used for displaying and exporting the screened results;
the display form of the screening results comprises: "file by" and "time by":
representing a picture and a recognition result corresponding to the picture according to each line of the file;
"time by time" is to show the results in terms of the time period over which the extended F phenomenon occurs;
in response to this, the mobile terminal is allowed to,
when the screening result is exported according to the file, each row comprises the path and the identification result of the picture;
when deriving the screening results "in time", each row contains an extended F epoch.
10. The system for screening ionospheric frequency elevation map extended F phenomenon radar plots of claim 9, further comprising:
the recognition result checking module is used for checking and correcting the displayed screening results and comprises: the picture name, the recognition result, the non-zoomed picture and the sequence number currently recorded in the 'by file' list are checked, and if the recognition result has a problem, the recognition result is corrected by reselecting a result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210998164.2A CN115508800A (en) | 2022-08-19 | 2022-08-19 | Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210998164.2A CN115508800A (en) | 2022-08-19 | 2022-08-19 | Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115508800A true CN115508800A (en) | 2022-12-23 |
Family
ID=84501635
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210998164.2A Pending CN115508800A (en) | 2022-08-19 | 2022-08-19 | Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115508800A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115525786A (en) * | 2022-10-11 | 2022-12-27 | 中国传媒大学 | Method for constructing ionospheric frequency high-graph classification sample library |
| CN116664789A (en) * | 2023-07-24 | 2023-08-29 | 齐鲁空天信息研究院 | Global ionosphere grid data rapid visualization method and system |
-
2022
- 2022-08-19 CN CN202210998164.2A patent/CN115508800A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115525786A (en) * | 2022-10-11 | 2022-12-27 | 中国传媒大学 | Method for constructing ionospheric frequency high-graph classification sample library |
| CN115525786B (en) * | 2022-10-11 | 2024-02-20 | 中国传媒大学 | Method for constructing ionospheric frequency high-graph classification sample library |
| CN116664789A (en) * | 2023-07-24 | 2023-08-29 | 齐鲁空天信息研究院 | Global ionosphere grid data rapid visualization method and system |
| CN116664789B (en) * | 2023-07-24 | 2023-10-24 | 齐鲁空天信息研究院 | Global ionosphere grid data rapid visualization method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115508800A (en) | Method and system for screening ionospheric frequency elevation map extension F phenomenon radar graph | |
| US5053778A (en) | Generation of topographic terrain models utilizing synthetic aperture radar and surface level data | |
| CN112068153B (en) | Crown clearance rate estimation method based on foundation laser radar point cloud | |
| CN102354328B (en) | Leaf area index (LAI) product inversion method and system for global earth surface | |
| CN107479065B (en) | Forest gap three-dimensional structure measuring method based on laser radar | |
| CN110263790A (en) | A kind of power plant's ammeter character locating and recognition methods based on convolutional neural networks | |
| CN109840497A (en) | A kind of pointer-type water meter reading detection method based on deep learning | |
| CN115223054A (en) | Remote sensing image change detection method based on partition clustering and convolution | |
| CN110826454B (en) | Remote sensing image change detection method and device | |
| CN117423013B (en) | Ore potential prediction method based on geological big data | |
| CN111651411A (en) | Complex terrain remote sensing soil moisture product downscaling method | |
| CN112766381B (en) | Attribute-guided SAR image generation method under limited sample | |
| CN114943902A (en) | Urban vegetation unmanned aerial vehicle remote sensing classification method based on multi-scale feature perception network | |
| CN117710508A (en) | Near-surface temperature inversion method and device for generating countermeasure network based on improved condition | |
| CN109359631A (en) | A kind of Sea Ice Types Classification in Remote Sensing Image method based on convolutional neural networks | |
| CN116206203B (en) | Oil spill detection method based on SAR and Dual-EndNet | |
| CN104915925A (en) | Dry valley extraction method by preparing high-precision mask file | |
| CN116883594A (en) | Ocean three-dimensional temperature remote sensing super-resolution reconstruction method integrating satellite and buoy observation | |
| CN111444824A (en) | Vegetation spatial distribution pattern investigation method and vegetation classification method based on unmanned aerial vehicle technology | |
| Wang | Forecasting and identifying the meteorological and hydrological conditions favoring the occurrence of severe hazes in Beijing and Shanghai using deep learning | |
| CN115128569A (en) | Radar reconnaissance full-pulse data visual sorting method, server and storage medium | |
| CN114255388A (en) | Artificial intelligent automatic detection method for embedding quality of seismic acquisition receiving device | |
| CN115525786B (en) | Method for constructing ionospheric frequency high-graph classification sample library | |
| CN102495405B (en) | Evaluation method of TM/ETM (thematic mapper/enhanced thematic mapper) and image-based atmospheric correction product quality | |
| CN113887567B (en) | Vegetable quality detection method, system, medium and equipment |
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 |