Aviation electromagnetic space abnormal coordinate data classification and filling method

Abstract

The invention discloses a method for classifying and filling abnormal coordinate data of an aviation electromagnetic space, which solves the problems of unreliable processing of the abnormal coordinate data of the electromagnetic space and the like, and comprises the following steps of: searching and identifying an abnormal value of coordinate data by adopting an arc ellipse model fitting algorithm, and extracting abnormal coordinate data information; carrying out interpolation by adopting an FCS stochastic regression model, correcting a coordinate data abnormal value, and carrying out data backfilling; acquiring abnormal coordinate data information and an original aviation electromagnetic coordinate data set; constructing a target feature division and recognition model, and performing feature and target division on the abnormal coordinate data information and the original aviation electromagnetic coordinate data set to obtain an aviation electromagnetic coordinate data set; and constructing a linear regression model, fitting the aviation electromagnetic coordinate data after correcting the abnormal value of the coordinate data to obtain a corrected value of the abnormal value of the coordinate data, and replacing the abnormal value of the coordinate data with the corrected value of the abnormal value of the coordinate data.

Description

Technical Field

[0001] The invention relates to the technical field of airborne electromagnetic data processing, and in particular to a method for classifying and filling airborne electromagnetic space abnormal coordinate data. Background Art

[0002] The UAV airborne electromagnetic method is widely used in steep high-altitude and dangerous mountainous areas, and the processing of its data has gradually matured. Among them, after the UAV airborne electromagnetic raw data is collected, the coordinate format needs to be converted in Aowei. However, occasionally, the coordinate data caused by satellites, environment, climate and other factors will be abnormal at a certain point or in some places. That is, because airborne electromagnetic data involves air influences, compared with magnetotelluric methods, airborne electromagnetic data is more disturbed by factors such as temperature, satellites, altitude, fuselage vibration, etc. At present, the traditional method of processing abnormal data is to fill in the coordinate data manually, which is more subjective and cannot scientifically simulate and correct the abnormal data. The traditional single linear regression model is also difficult to make more accurate changes to abnormal data, because the single random regression model cannot perform more complex calculations.

[0003] Therefore, it is urgent to propose a classification and filling method for airborne electromagnetic space anomaly coordinate data that is simple in logic, accurate and reliable. Summary of the invention

[0004] In view of the above problems, the purpose of the present invention is to provide a method for classifying and filling aero-electromagnetic space abnormal coordinate data. The technical solution adopted by the present invention is as follows: A method for classifying and filling aero-electromagnetic space abnormal coordinate data, comprising the following steps: Step S1, obtaining the collected original airborne electromagnetic coordinate data set, using the lone ellipse model fitting algorithm to find and identify abnormal values ​​of the coordinate data; checking the data column corresponding to the abnormal value of the coordinate data, and extracting the abnormal coordinate data information; Step S2, obtaining the original airborne electromagnetic coordinate data set, interpolating using the FCS random regression model, correcting coordinate data outliers, and performing data backfilling; Step S21, obtaining abnormal coordinate data information and original airborne electromagnetic coordinate data set; Step S22, construct a target feature division and recognition model, and divide the features and targets of the abnormal coordinate data information and the original aero-electromagnetic coordinate data set to obtain the aero-electromagnetic coordinate data set; the target feature division and recognition model includes an input layer, a backbone network, a neck network and an output prediction layer connected in sequence; the input layer receives the abnormal coordinate data information and the original aero-electromagnetic coordinate data set; the backbone network combines the convolution layer and batch normalization technology to accelerate the training process and stabilize the model performance, and enhance the network's ability to express features and the robustness of the model. The neck network processes the data for enhanced features of the aero-electromagnetic coordinate data, and outputs image data using the output prediction layer; the neck network is provided with several parallel spatial information enhancement modules SIE; the spatial information enhancement module SIE includes a coordinate convolution layer CoordConv, a maximum delay layer Maxpool, a twentieth standard convolution module CBL, a ninth convolution layer and a Sigmoid activation function connected in sequence, and divides the features of the spatial coordinates corresponding to the abnormal coordinate data information; Step S23, constructing a linear regression model, and fitting the aero-electromagnetic coordinate data after correcting the abnormal value of the coordinate data, obtaining the correction value of the abnormal value of the coordinate data, and using the correction value of the abnormal value of the coordinate data to replace the abnormal value of the coordinate data.

[0005] Furthermore, the estimated proportion of outliers in the original aeroelectromagnetic coordinate data set corresponding to the solitary ellipse model fitting algorithm is preset to 0.2.

[0006] Furthermore, the backbone network includes a first standard convolution and cross-stage local network group, a second standard convolution and cross-stage local network group, a third standard convolution and cross-stage local network group, a seventh standard convolution module CBL and a spatial pyramid delay layer connected in sequence; the first standard convolution and cross-stage local network group includes a convolution neural layer fcous, a first standard convolution module CBL, a first cross-stage local network CSP1_1, a second standard convolution module CBL, a first cross-stage local network CSP1_2, a third standard convolution module CBL and a first cross-stage local network CSP1_3 connected in sequence; the second standard convolution and cross-stage local network group includes a fourth standard convolution module CB connected in sequence L, the second cross-stage local network CSP1_2, the fifth standard convolution module CBL and the second cross-stage local network CSP1_3; the third standard convolution and cross-stage local network group includes the sixth standard convolution module CBL and the third cross-stage local network CSP1_3 connected in sequence; the convolution neural layer fcous is connected to the input layer; the first cross-stage local network CSP1_3 is connected to the fourth standard convolution module CBL; the second cross-stage local network CSP1_3 is connected to the sixth standard convolution module CBL and the neck network; the third cross-stage local network CSP1_3 is connected to the seventh standard convolution module CBL and the neck network; the spatial pyramid delay layer is connected to the neck network.

[0007] Further, the neck network includes a first cross-stage local network CSP2_3, an eighth standard convolution module CBL, a first upsampling, a first connection channel Concat, a second cross-stage local network CSP2_3, a ninth standard convolution module CBL, a second upsampling, a first spatial information enhancement module SIE, a second connection channel Concat and a third cross-stage local network CSP2_3 connected in sequence, a tenth standard convolution module CBL connected to the third cross-stage local network CSP2_3, a second spatial information enhancement module SIE connected to the ninth standard convolution module CBL and the tenth standard convolution module CBL, a third connection channel Concat connected to the second spatial information enhancement module SIE, a fourth cross-stage local network CSP2_3 connected to the third connection channel Concat, and a fourth cross-stage local network CSP2_3 connected to the fourth cross-stage local network The present invention relates to an eleventh standard convolution module CBL connected to the first cross-stage local network CSP2_3, a third spatial information enhancement module SIE connected to the eleventh standard convolution module CBL and the eighth standard convolution module CBL, a fourth connection channel Concat connected to the third spatial information enhancement module SIE, and a fifth cross-stage local network CSP2_3 connected to the fourth connection channel Concat; the first cross-stage local network CSP2_3 is connected to the spatial pyramid delay layer; the first connection channel Concat is connected to the third cross-stage local network CSP1_3; the first spatial information enhancement module SIE is connected to the second cross-stage local network CSP1_3; the third cross-stage local network CSP2_3, the fourth cross-stage local network CSP2_3 and the fifth cross-stage local network CSP2_3 are respectively connected to the output prediction layer.

[0008] Furthermore, the first cross-stage local network CSP1_1, the first cross-stage local network CSP1_2, the first cross-stage local network CSP1_3, the second cross-stage local network CSP1_2, the second cross-stage local network CSP1_3, and the third cross-stage local network CSP1_3 have the same structure, and all include a fourteenth standard convolution module CBL, a first residual unit ResUnit, a fifth convolution layer, a fifth connection channel Concat, a second batch normalization layer, a second rectified linear unit function leakyrelu and a fifteenth standard convolution module CBL connected in sequence, and a sixth convolution layer connected between the input of the fourteenth standard convolution module CBL and the fifth connection channel Concat.

[0009] Furthermore, the first cross-stage local network CSP2_3, the second cross-stage local network CSP2_3, the third cross-stage local network CSP2_3, the fourth cross-stage local network CSP2_3 and the fifth cross-stage local network CSP2_3 have the same structure, and all include a sixteenth standard convolution module CBL, a seventeenth standard convolution module CBL, an eighteenth standard convolution module CBL, a second residual unit ResUnit, a seventh convolution layer, a sixth connection channel Concat, a third batch normalization layer, a third rectified linear unit function leakyrelu and a nineteenth standard convolution module CBL, which are connected in sequence, and an eighth convolution layer connected between the input of the sixteenth standard convolution module CBL and the sixth connection channel Concat.

[0010] Furthermore, the first residual unit ResUnit and the second residual unit ResUnit have the same structure, and both include a twelfth standard convolution module CBL, a thirteenth standard convolution module CBL and a residual connection layer add which are connected in sequence.

[0011] Further, the first standard convolution module CBL, the second standard convolution module CBL, the third standard convolution module CBL, the fourth standard convolution module CBL, the fifth standard convolution module CBL, the sixth standard convolution module CBL, the seventh standard convolution module CBL, the eighth standard convolution module CBL, the ninth standard convolution module CBL, the tenth standard convolution module CBL, the eleventh standard convolution module CBL, the twelfth standard convolution module CBL, the thirteenth standard convolution module CBL, the fourteenth standard convolution module CBL, the fifteenth standard convolution module CBL, the sixteenth standard convolution module CBL, the seventeenth standard convolution module CBL, the eighteenth standard convolution module CBL, the nineteenth standard convolution module CBL and the twentieth standard convolution module CBL have the same structure, and all include a fourth convolution layer, a first batch normalization layer and a first rectified linear unit function leakyrelu connected in sequence.

[0012] Furthermore, the linear regression model is expressed as:

[0013] in, Represents the regression function of the linear regression model; Indicates the aero-electromagnetic coordinate data after the abnormal values ​​of the coordinate data are corrected; represents weight; Represents the bias vector.

[0014] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses a lone ellipse model fitting algorithm to find and identify abnormal values ​​of coordinate data. Because abnormal aeronautical electromagnetic coordinate data has a certain correlation with similar aeronautical electromagnetic coordinate data, the lone ellipse model will refer to adjacent data, so the judgment of whether a certain point is an abnormal value is more accurate.

[0015] The present invention adopts the FCS random regression model to estimate the abnormal coordinate data, and performs random sampling in the predicted distribution of the model to correct the abnormal coordinate value to meet the correlation between the abnormal point and the surrounding coordinates. At the same time, it can effectively avoid the subjectivity of artificially filling in the abnormal coordinate data points, making the interpreted data more reliable.

[0016] The present invention uses a target feature division and recognition model to divide the features and targets of data with missing values ​​and data without missing values ​​(i.e., abnormal coordinate data information and original aero-electromagnetic coordinate data sets), and sets several spatial information enhancement modules SIE in the target feature division and recognition model, so that the feature division of the coordinates around the abnormal data in the aero-electromagnetic spatial coordinate data is more obvious, and the spatial position point where the abnormal point is located can be better reflected, making the division result more reliable. After the present invention adds the spatial information enhancement module SIE, it realizes the full fusion of channel information for the aero-electromagnetic spatial coordinate data, and maintains high feature representation capability.

[0017] The spatial information enhancement module SIE of the present invention is composed of a coordinate convolution layer CoordConv, a maximum delay layer Maxpool, a twentieth standard convolution module CBL, a ninth convolution layer and a Sigmoid activation function. Among them, the coordinate convolution layer CoordConv modifies the standard convolution operation by adding an additional input channel encoding the coordinates of each pixel, so that the convolution filter in the coordinate convolution layer CoordConv can accurately know the position in the input space, thereby significantly improving the ability of the network to learn spatial transformation.

[0018] (5) The present invention adopts three spatial information enhancement modules SIE to supplement the scene context information for the data of outliers when correcting the data. The output data may include the location, scale or environmental characteristics of the object to assist target detection and image segmentation tasks.

[0019] In summary, the present invention has the advantages of simple logic, accuracy and reliability, and has high practical value and promotion value in the field of aviation electromagnetic data processing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope of protection. For those skilled in the art, other related drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 It is a logic flow chart of the present invention.

[0022] Figure 2 It is a structural diagram of the input layer and backbone network of the target feature segmentation and recognition model in the present invention.

[0023] Figure 3 It is a schematic diagram of the structure of the neck network and output prediction layer of the target feature segmentation and recognition model in the present invention.

[0024] Figure 4 It is a schematic diagram of the structure of the standard convolution module CBL in the present invention.

[0025] Figure 5 It is a structural schematic diagram of the residual unit ResUnit in the present invention.

[0026] Figure 6 It is a schematic diagram of the structure of the cross-stage local network CSP1_X in the present invention.

[0027] Figure 7 It is a schematic diagram of the structure of the cross-stage local network CSP2_Y in the present invention.

[0028] Figure 8 It is a structural diagram of the spatial information enhancement module SIE in the present invention. DETAILED DESCRIPTION

[0029] In order to make the purpose, technical scheme and advantages of this application clearer, the present invention is further described below in conjunction with the accompanying drawings and embodiments, and the embodiments of the present invention include but are not limited to the following embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0030] In this embodiment, the term "and / or" is merely a term used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent three situations: A exists alone, A and B exist at the same time, and B exists alone.

[0031] The terms "first" and "second" in the description and claims of this embodiment are used to distinguish different objects rather than to describe a specific order of objects. For example, a first target object and a second target object are used to distinguish different target objects rather than to describe a specific order of target objects.

[0032] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0033] In the description of the embodiments of the present application, unless otherwise specified, the meaning of "multiple" refers to two or more than two. For example, multiple processing units refer to two or more processing units; multiple systems refer to two or more systems.

[0034] like Figures 1 to 8 As shown, this embodiment provides a classification and filling method for abnormal coordinate data in airborne electromagnetic space, which solves the problem of abnormal coordinate data in airborne electromagnetic space coordinate data caused by factors such as satellites, environment, and climate, and removes and fills abnormal coordinate data. Specifically, it includes the following steps: The first step is to obtain the collected original airborne electromagnetic coordinate data set, and use the lone ellipse model fitting algorithm to find and identify the coordinate data outliers; check the data columns corresponding to the coordinate data outliers, and extract the abnormal coordinate data information. In this embodiment, the estimated proportion of outliers in the original airborne electromagnetic coordinate data set corresponding to the lone ellipse model fitting algorithm is preset to 0.2. For airborne electromagnetic space coordinate data, each coordinate is closely related to the surrounding coordinates, and the accuracy is improved by increasing the concentration value.

[0035] The second step is to obtain the original airborne electromagnetic coordinate data set, use the FCS random regression model to interpolate, correct the coordinate data outliers, and backfill the data. The specific steps are as follows: Step S21, extracting data with missing values ​​and data without missing values ​​(ie, abnormal coordinate data information and original aeroelectromagnetic coordinate data set).

[0036] Step S22, constructing a target feature division and recognition model, and dividing features and targets of the abnormal coordinate data information and the original airborne electromagnetic coordinate data set to obtain an airborne electromagnetic coordinate data set.

[0037] In this embodiment, the target feature segmentation and recognition model includes an input layer, a backbone network, a neck network, and an output prediction layer connected in sequence. The input layer receives abnormal coordinate data information and an original aero-electromagnetic coordinate data set. In addition, the backbone network performs aero-electromagnetic coordinate data feature enhancement on the abnormal coordinate data information and the original aero-electromagnetic coordinate data set. The neck network of this embodiment performs data processing on the aero-electromagnetic coordinate data feature enhancement, and outputs image data using the output prediction layer.

[0038] In this embodiment, the backbone network includes a first standard convolution and cross-stage local network group, a second standard convolution and cross-stage local network group, a third standard convolution and cross-stage local network group, a seventh standard convolution module CBL and a spatial pyramid delay layer connected in sequence. Among them, the first standard convolution and cross-stage local network group includes a convolution neural layer fcous, a first standard convolution module CBL, a first cross-stage local network CSP1_1, a second standard convolution module CBL, a first cross-stage local network CSP1_2, a third standard convolution module CBL and a first cross-stage local network CSP1_3 connected in sequence. In addition, the second standard convolution and cross-stage local network group includes a fourth standard convolution module CBL, a second cross-stage local network CSP1_2, a fifth standard convolution module CBL and a second cross-stage local network CSP1_3 connected in sequence. Here, the third standard convolution and cross-stage local network group includes a sixth standard convolution module CBL and a third cross-stage local network CSP1_3 connected in sequence. In this embodiment, the second standard convolution and cross-stage local network group is the same as the second standard convolution module CBL, the first cross-stage local network CSP1_2, the third standard convolution module CBL and the first cross-stage local network CSP1_3 in the first standard convolution and cross-stage local network group, and the third standard convolution and cross-stage local network group is the same as the fifth standard convolution module CBL and the second cross-stage local network CSP1_3 in the second standard convolution and cross-stage local network group. The advantage is that it speeds up the training speed, prevents the gradient disappearance problem, maintains the stability of the model, and enhances the feature expression capability, providing better feature information for subsequent tasks.

[0039] Here, the second standard convolution and cross-stage local network group transmits the output abnormal coordinate data information and the original aeronautical electromagnetic coordinate data set to the third standard convolution and cross-stage local network group and the first spatial information enhancement module SIE. In addition, the third standard convolution and cross-stage local network group transmits the abnormal coordinate data information and the original aeronautical electromagnetic coordinate data set to the seventh standard convolution module CBL and the first connection channel Concat. In addition, the spatial pyramid delay layer of this embodiment processes the aeronautical electromagnetic spatial coordinate data information as follows: it fuses image features of different scales to capture detailed information of objects at different scales, thereby improving the model's ability to recognize and locate objects. Detailed processing steps: 1) Construct a pyramid structure to amplify low-scale image features and combine them with high-scale image features; 2) Integrate information of different scales into the same channel through upsampling or downsampling operations; 3) Provide rich multi-scale description information to enhance the expression ability of the model.

[0040] In this embodiment, the first cross-stage local network CSP1_1, the first cross-stage local network CSP1_2, the first cross-stage local network CSP1_3, the second cross-stage local network CSP1_2, the second cross-stage local network CSP1_3, and the third cross-stage local network CSP1_3 have the same structure and all include a fourteenth standard convolution module CBL, a first residual unit ResUnit, a fifth convolution layer, a fifth connection channel Concat, a second batch normalization layer, a second corrected linear unit function leakyrelu and a fifteenth standard convolution module CBL connected in sequence, and a sixth convolution layer connected between the input of the fourteenth standard convolution module CBL and the fifth connection channel Concat. This type of cross-stage local network CSP1_X is used to enhance the network's ability to express features and the robustness of the model. Among them, X takes the values ​​of 1, 2, and 3 respectively, namely the corresponding first cross-stage local network CSP1_1, the first cross-stage local network CSP1_2, the first cross-stage local network CSP1_3, the second cross-stage local network CSP1_2, the second cross-stage local network CSP1_3, and the third cross-stage local network CSP1_3.

[0041] In this embodiment, the neck network includes a first cross-stage local network CSP2_3, an eighth standard convolution module CBL, a first upsampling, a first connection channel Concat, a second cross-stage local network CSP2_3, a ninth standard convolution module CBL, a second upsampling, a first spatial information enhancement module SIE, a second connection channel Concat and a third cross-stage local network CSP2_3 connected in sequence, a tenth standard convolution module CBL connected to the third cross-stage local network CSP2_3, a second spatial information enhancement module SIE connected to the ninth standard convolution module CBL and the tenth standard convolution module CBL, a third connection channel Concat connected to the second spatial information enhancement module SIE, a fourth cross-stage local network CSP2_3 connected to the third connection channel Concat, an eleventh standard convolution module CBL connected to the fourth cross-stage local network CSP2_3, a third spatial information enhancement module SIE connected to the eleventh standard convolution module CBL and the eighth standard convolution module CBL, a fourth connection channel Concat connected to the third spatial information enhancement module SIE, and a fifth cross-stage local network CSP2_3 connected to the fourth connection channel Concat. Among them, here, the three spatial information enhancement modules SIE supplement the scene context information for the data of the outliers when correcting the data, and its output data may include the location, scale or environmental characteristics of the object to assist target detection and image segmentation tasks.

[0042] In this embodiment, the first spatial information enhancement module SIE, the second spatial information enhancement module SIE, and the third spatial information enhancement module SIE have the same structure, which include a coordinate convolution layer CoordConv, a maximum delay layer Maxpool, a twentieth standard convolution module CBL, a ninth convolution layer and a Sigmoid activation function connected in sequence, and divides the features of the spatial coordinates corresponding to the abnormal coordinate data information.

[0043] In this embodiment, the first cross-stage local network CSP2_3, the second cross-stage local network CSP2_3, the third cross-stage local network CSP2_3, the fourth cross-stage local network CSP2_3 and the fifth cross-stage local network CSP2_3 have the same structure, and all include the sixteenth standard convolution module CBL, the seventeenth standard convolution module CBL, the eighteenth standard convolution module CBL, the second residual unit ResUnit, the seventh convolution layer, the sixth connection channel Concat, the third batch normalization layer, the third corrected linear unit function leakyrelu and the nineteenth standard convolution module CBL, and the eighth convolution layer connected between the input of the sixteenth standard convolution module CBL and the sixth connection channel Concat. This type of cross-stage local network CSP2_Y is used to enhance the network's ability to express features and the robustness of the model, where Y takes a value of 3.

[0044] In this embodiment, the first residual unit ResUnit and the second residual unit ResUnit have the same structure, and both include a twelfth standard convolution module CBL, a thirteenth standard convolution module CBL, and a residual connection layer add connected in sequence. In addition, the first standard convolution module CBL, the second standard convolution module CBL, the third standard convolution module CBL, the fourth standard convolution module CBL, the fifth standard convolution module CBL, the sixth standard convolution module CBL, the seventh standard convolution module CBL, the eighth standard convolution module CBL, the ninth standard convolution module CBL, the tenth standard convolution module CBL, the eleventh standard convolution module CBL, the twelfth standard convolution module CBL, the thirteenth standard convolution module CBL, the fourteenth standard convolution module CBL, the fifteenth standard convolution module CBL, the sixteenth standard convolution module CBL, the seventeenth standard convolution module CBL, the eighteenth standard convolution module CBL, the nineteenth standard convolution module CBL, and the twentieth standard convolution module CBL have the same structure, and both include a fourth convolutional layer, a first batch normalization layer, and a first rectified linear unit function leakyrelu connected in sequence.

[0045] Step S23: Construct a linear regression model, fit the airborne electromagnetic coordinate data after correcting the outliers of the coordinate data to obtain the correction values of the outliers of the coordinate data, and use the correction values of the outliers of the coordinate data to replace the outliers of the coordinate data.

[0046] The above embodiments are only the preferred embodiments of the present invention, and do not limit the protection scope of the present invention. Any changes made by using the design principle of the present invention and non-creative labor on this basis shall fall within the protection scope of the present invention.

Claims

1. A method for classifying and filling airborne electromagnetic space anomaly coordinate data, characterized in that: The following steps are involved: Step S1, obtaining the collected original airborne electromagnetic coordinate data set, and using the lone ellipse model fitting algorithm to find and identify abnormal values ​​of the coordinate data; Check the data columns corresponding to the abnormal values ​​of the coordinate data and extract the abnormal coordinate data information; Step S2, obtaining the original airborne electromagnetic coordinate data set, interpolating using the FCS random regression model, correcting coordinate data outliers, and performing data backfilling; Step S21, obtaining abnormal coordinate data information and original airborne electromagnetic coordinate data set; Step S22, constructing a target feature division and recognition model, and dividing the features and targets of the abnormal coordinate data information and the original aeroelectromagnetic coordinate data set to obtain the aeroelectromagnetic coordinate data set; the target feature division and recognition model includes an input layer, a backbone network, a neck network and an output prediction layer connected in sequence; the input layer receives the abnormal coordinate data information and the original aeroelectromagnetic coordinate data set; the backbone network performs coordinate data feature enhancement processing on the abnormal coordinate data information and the original aeroelectromagnetic coordinate data set; the neck network performs data processing on the enhanced features of the aeroelectromagnetic coordinate data, and outputs image data using the output prediction layer; a plurality of parallel spatial information enhancement modules SIE are arranged in the neck network; the spatial information enhancement module SIE includes a coordinate convolution layer CoordConv, a maximum delay layer Maxpool, a twentieth standard convolution module CBL, a ninth convolution layer and a Sigmoid activation function connected in sequence, and divides the features of the spatial coordinates corresponding to the abnormal coordinate data information; Step S23, constructing a linear regression model, and fitting the aero-electromagnetic coordinate data after correcting the abnormal value of the coordinate data, obtaining the correction value of the abnormal value of the coordinate data, and using the correction value of the abnormal value of the coordinate data to replace the abnormal value of the coordinate data.

2. The method for classifying and filling aero-electromagnetic space abnormal coordinate data according to claim 1, characterized in that: The estimated ratio of outliers in the original aeroelectromagnetic coordinate data set corresponding to the solitary ellipse model fitting algorithm is preset to 0.

2.

3. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 1 or 2, characterized in that: The backbone network includes a first standard convolution and cross-stage local network group, a second standard convolution and cross-stage local network group, a third standard convolution and cross-stage local network group, a seventh standard convolution module CBL and a spatial pyramid delay layer connected in sequence; the first standard convolution and cross-stage local network group includes a convolution neural layer fcous, a first standard convolution module CBL, a first cross-stage local network CSP1_1, a second standard convolution module CBL, a first cross-stage local network CSP1_2, a third standard convolution module CBL and a first cross-stage local network CSP1_3 connected in sequence; the second standard convolution and cross-stage local network group includes a fourth standard convolution module CBL, a first cross-stage local network CSP1_4, a second standard convolution module CBL, a first cross-stage local network CSP1_5, a third standard convolution module CBL and a first cross-stage local network CSP1_6 connected in sequence. two cross-stage local networks CSP1_2, a fifth standard convolution module CBL and a second cross-stage local network CSP1_3; the third standard convolution and cross-stage local network group includes a sixth standard convolution module CBL and a third cross-stage local network CSP1_3 connected in sequence; the convolution neural layer fcous is connected to the input layer; the first cross-stage local network CSP1_3 is connected to the fourth standard convolution module CBL; the second cross-stage local network CSP1_3 is connected to the sixth standard convolution module CBL and the neck network; the third cross-stage local network CSP1_3 is connected to the seventh standard convolution module CBL and the neck network; the spatial pyramid delay layer is connected to the neck network.

4. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 3, characterized in that: The neck network includes a first cross-stage local network CSP2_3, an eighth standard convolution module CBL, a first upsampling, a first connection channel Concat, a second cross-stage local network CSP2_3, a ninth standard convolution module CBL, a second upsampling, a first spatial information enhancement module SIE, a second connection channel Concat and a third cross-stage local network CSP2_3 connected in sequence, a tenth standard convolution module CBL connected to the third cross-stage local network CSP2_3, a second spatial information enhancement module SIE connected to the ninth standard convolution module CBL and the tenth standard convolution module CBL, a third connection channel Concat connected to the second spatial information enhancement module SIE, a fourth cross-stage local network CSP2_3 connected to the third connection channel Concat, and a fourth cross-stage local network CSP2_3 connected to the fourth cross-stage local network C SP2_3, an eleventh standard convolution module CBL connected to the eleventh standard convolution module CBL and the eighth standard convolution module CBL, a fourth connection channel Concat connected to the third spatial information enhancement module SIE, and a fifth cross-stage local network CSP2_3 connected to the fourth connection channel Concat; the first cross-stage local network CSP2_3 is connected to the spatial pyramid delay layer; the first connection channel Concat is connected to the third cross-stage local network CSP1_3; the first spatial information enhancement module SIE is connected to the second cross-stage local network CSP1_3; the third cross-stage local network CSP2_3, the fourth cross-stage local network CSP2_3 and the fifth cross-stage local network CSP2_3 are respectively connected to the output prediction layer.

5. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 4, characterized in that: The first cross-stage local network CSP1_1, the first cross-stage local network CSP1_2, the first cross-stage local network CSP1_3, the second cross-stage local network CSP1_2, the second cross-stage local network CSP1_3, and the third cross-stage local network CSP1_3 have the same structure, and all include a fourteenth standard convolution module CBL, a first residual unit ResUnit, a fifth convolution layer, a fifth connection channel Concat, a second batch normalization layer, a second rectified linear unit function leakyrelu and a fifteenth standard convolution module CBL connected in sequence, and a sixth convolution layer connected between the input of the fourteenth standard convolution module CBL and the fifth connection channel Concat.

6. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 5, characterized in that: The first cross-stage local network CSP2_3, the second cross-stage local network CSP2_3, the third cross-stage local network CSP2_3, the fourth cross-stage local network CSP2_3 and the fifth cross-stage local network CSP2_3 have the same structure, and all include a sixteenth standard convolution module CBL, a seventeenth standard convolution module CBL, an eighteenth standard convolution module CBL, a second residual unit ResUnit, a seventh convolution layer, a sixth connection channel Concat, a third batch normalization layer, a third rectified linear unit function leakyrelu and a nineteenth standard convolution module CBL, which are connected in sequence, and an eighth convolution layer connected between the input of the sixteenth standard convolution module CBL and the sixth connection channel Concat.

7. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 6, characterized in that: The first residual unit ResUnit and the second residual unit ResUnit have the same structure, and both include a twelfth standard convolution module CBL, a thirteenth standard convolution module CBL and a residual connection layer add which are connected in sequence.

8. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 7, characterized in that: The first standard convolution module CBL, the second standard convolution module CBL, the third standard convolution module CBL, the fourth standard convolution module CBL, the fifth standard convolution module CBL, the sixth standard convolution module CBL, the seventh standard convolution module CBL, the eighth standard convolution module CBL, the ninth standard convolution module CBL, the tenth standard convolution module CBL, the eleventh standard convolution module CBL, the twelfth standard convolution module CBL, the thirteenth standard convolution module CBL, the fourteenth standard convolution module CBL, the fifteenth standard convolution module CBL, the sixteenth standard convolution module CBL, the seventeenth standard convolution module CBL, the eighteenth standard convolution module CBL, the nineteenth standard convolution module CBL and the twentieth standard convolution module CBL have the same structure, and all include a fourth convolution layer, a first batch normalization layer and a first rectified linear unit function leakyrelu connected in sequence.

9. The method for classifying and filling aero-electromagnetic space anomaly coordinate data according to claim 1, characterized in that: The linear regression model is expressed as: ,in, Represents the regression function of the linear regression model; Indicates the aero-electromagnetic coordinate data after the abnormal values ​​of the coordinate data are corrected; represents weight; Represents the bias vector.

Images