CN113592341A - Measurement loss function, sector complexity evaluation method and system - Google Patents
Measurement loss function, sector complexity evaluation method and system Download PDFInfo
- Publication number
- CN113592341A CN113592341A CN202110912714.XA CN202110912714A CN113592341A CN 113592341 A CN113592341 A CN 113592341A CN 202110912714 A CN202110912714 A CN 202110912714A CN 113592341 A CN113592341 A CN 113592341A
- Authority
- CN
- China
- Prior art keywords
- sector
- data
- loss function
- network model
- air traffic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 37
- 238000005259 measurement Methods 0.000 title claims abstract description 27
- 230000006870 function Effects 0.000 claims abstract description 65
- 238000000034 method Methods 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 35
- 239000013598 vector Substances 0.000 claims abstract description 21
- 239000003795 chemical substances by application Substances 0.000 claims description 29
- 238000012163 sequencing technique Methods 0.000 claims description 17
- 238000012360 testing method Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 7
- 238000005520 cutting process Methods 0.000 claims description 6
- 238000003780 insertion Methods 0.000 claims description 5
- 230000037431 insertion Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 238000007635 classification algorithm Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000013507 mapping Methods 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 8
- 238000013135 deep learning Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 238000013145 classification model Methods 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000002194 synthesizing effect Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004590 computer program Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
- G06Q10/06313—Resource planning in a project environment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/40—Business processes related to the transportation industry
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Economics (AREA)
- Theoretical Computer Science (AREA)
- Strategic Management (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Operations Research (AREA)
- Game Theory and Decision Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Primary Health Care (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- Biodiversity & Conservation Biology (AREA)
- Educational Administration (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of airspace situation evaluation of air traffic control, and particularly relates to a method and a system for evaluating a measurement loss function and sector complexity, wherein the method for evaluating the sector complexity comprises the following steps: carrying out sector operation complexity grade marking on sector dynamic traffic data of the empty domain sector; constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking; constructing a measurement loss function; training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and sector operation complexity evaluation is carried out according to the trained network model, so that nonlinear mapping from original image air learning to semantic feature embedding space by using a depth measurement learning technology is realized, embedding vectors of semantic similar samples in the embedding space are closer, and samples with different semantics are separated from each other, so that more accurate sector operation complexity evaluation is realized.
Description
Technical Field
The invention belongs to the technical field of airspace situation assessment of air traffic control, and particularly relates to a method and a system for assessing a measurement loss function and sector complexity.
Background
In an Air Traffic Management System (ATMS), an airspace is divided into a plurality of sectors as basic control units. An air traffic controller (ATCos) monitors flight travel tracks in each sector in real time and sends control instructions to pilots so as to avoid congestion, conflict, flight delay and the like. With the rapid development of domestic air transportation industry in recent years, air traffic conditions become increasingly complex, and controllers are often in a high-density and overload working state due to limited airspace resources and greatly increased air traffic volume. Therefore, how to scientifically and accurately evaluate the sector operation complexity is one of the widely studied problems. In order to evaluate the complexity of air traffic, researchers combine machine learning techniques to classify the complexity of air traffic by synthesizing a plurality of closely related factors to generate a more comprehensive metric index. In 2021, thank to scholar et al provided an image representation of a sector operation scene for the first time, and further provided a sector operation complexity evaluation method based on a Deep Convolutional Neural Network (DCNN) in combination with the advantages of a deep learning technique in the aspect of feature extraction, so that the defect of manually making a feature set in a machine learning-based method is overcome.
It can be seen from the existing research that, for the problem of evaluating the operation complexity of the sector, most of the research is to analyze the operation state information of the sector, construct a feature set related to the operation complexity of the sector, and finally use some classification methods to classify the complexity of the sector. The method based on machine learning is to manually construct a feature set according to some priori knowledge, and the method based on deep learning adopts a deep convolutional neural network to automatically extract relevant deep features. However, for most of classification methods based on similarity idea, it is far from sufficient to ignore the similarity information of data samples, and train the sector complexity classification model by only considering constructing a more suitable related feature set. This is also one of the reasons why the accuracy of the current sector complexity evaluation method is not high.
In the prior art, only the construction of more comprehensive sector operation complexity features is considered, and although the deep learning algorithm can automatically extract more complicated and abstract deep features from image data, the improvement on the classification accuracy of a subsequent classification model is limited.
Therefore, there is a need to design a new metric loss function, a sector complexity evaluation method and a system based on the above problems.
Disclosure of Invention
The invention aims to provide a method and a system for evaluating a metric loss function and sector complexity.
In order to solve the above technical problem, the present invention provides a metric loss function, including:
selecting an agent from each category of air traffic images as an anchor point sample p;
pulling all positive samples in the batch to be within a boundary alpha-m, pushing all negative samples to be outside the boundary alpha, and keeping an interval m between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+；
wherein x is an insertion corresponding to an image; p is an agent; d(x,p)Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only when the categories of x and p are the same, y is 1, otherwise y is 0;
the ranking proxy anchor loss is a metric loss function.
In a second aspect, the present invention further provides a sector complexity evaluation method based on dynamic air traffic image and depth metric learning, including:
carrying out sector operation complexity grade marking on sector dynamic traffic data of the empty domain sector;
constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking;
constructing a measurement loss function;
training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and
and evaluating the operation complexity of the sector according to the trained network model.
Further, the method for performing sector operation complexity level labeling on sector dynamic traffic data of the space domain sector includes:
acquiring data from an original air traffic operation database of a target airspace sector, and forming a sector dynamic traffic data set of the target airspace sector at different time intervals according to preset time granularity;
and dividing the sector dynamic traffic data set according to a preset time period, and carrying out sector operation complexity grade marking on the sector dynamic traffic data corresponding to each time period.
Further, the method for constructing the multi-channel air traffic scene image data of the airspace sector and constructing the sector dynamic traffic image library according to the sector operation complexity level mark comprises the following steps:
converting the dynamic traffic scene information of the target airspace sector into an air traffic scene multi-channel image according to an MTSI data representation method and sector operation complexity grade marking;
generating a height historical track image channel, a speed historical track image channel and a conflict prediction track image channel of a corresponding time period according to the dynamic traffic data of the sectors of each time period, and constructing a multi-channel air traffic scene image;
and associating the multi-channel air traffic scene image generated in each time period with the corresponding sector operation complexity grade label, acquiring a sector dynamic traffic image database, and dividing the sector dynamic traffic image database into a training data set and a test data set.
Further, the method for constructing the metric loss function comprises the following steps:
selecting an agent from each category of air traffic images as an anchor point sample p;
pulling all positive samples in the batch to be within a boundary alpha-m, pushing all negative samples to be outside the boundary alpha, and keeping an interval m between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+；
wherein x is an insertion corresponding to an image; p is an agent; d(x,p)Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only when the categories of x and p are the same, y is 1, otherwise y is 0;
the total penalty for the rank proxy anchor penalty is:
wherein, P is the set of all agents; p+A positive proxy set for data in the batch; for each agent p, the embedded vectors X are divided into two groups,
is a set of positive embedded vectors for p,
is a negative set of embedded vectors of p, and
for theEach proxy anchor p, according to wxp＝exp(T(α-d(x,p))),
Weighting the negative samples;
wherein, T is a parameter for controlling the weighting degree of the negative sample;
and taking the loss of the sequencing agent anchor as a loss function, and guiding the iterative optimization of the network model in the training process of the network model.
Further, the method for training the network model according to the multi-channel air traffic scene image data and the metric loss function comprises the following steps:
preprocessing a training data set, standardizing pixel values of images in the training data set, and performing data enhancement on input images through horizontal overturning and random cutting in a training process;
inputting the preprocessed training data set into a network model for training, taking the sequencing proxy anchor loss as a loss function, and continuously iterating and optimizing a target loss function through an optimizer to train the network model to obtain an optimal network model.
Further, the method for evaluating the complexity of sector operation according to the trained network model comprises the following steps:
carrying out standardized processing and center cutting on the test set data;
inputting the preprocessed image in the test set data into an optimal network model to obtain an embedded vector with semantic similarity of the test data;
and taking the embedded vector with semantic similarity as the input of a subsequent 1-NN classification algorithm, classifying the sector operation complexity, and finishing the evaluation of the target airspace sector operation complexity.
In a third aspect, the present invention further provides a sector complexity evaluation system based on dynamic air traffic image and depth metric learning, including:
the marking module is used for marking the sector operation complexity level of the sector dynamic traffic data of the space domain sector;
the database module is used for constructing multi-channel air traffic scene image data of an airspace sector and constructing a sector dynamic traffic image database according to sector operation complexity grade marks;
the loss function module is used for constructing a measurement loss function;
the network model training module trains a network model according to the multi-channel air traffic scene image data and the measurement loss function; and
and the complexity evaluation module is used for evaluating the operation complexity of the sector according to the trained network model.
The method has the advantages that the method marks the sector operation complexity level of the sector dynamic traffic data of the empty domain sector; constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking; constructing a measurement loss function; training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and sector operation complexity evaluation is carried out according to the trained network model, so that nonlinear mapping from original image air learning to semantic feature embedding space by using a depth measurement learning technology is realized, embedding vectors of semantic similar samples in the embedding space are closer, and samples with different semantics are separated from each other, so that more accurate sector operation complexity evaluation is realized.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and drawings.
In order to make the aforementioned and other objects, features and advantages of the present invention comprehensible, preferred embodiments accompanied with figures are described in detail below.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a flow chart of a sector complexity evaluation method based on dynamic air traffic image and depth metric learning in accordance with the present invention;
FIG. 2 is a schematic diagram of a sorting agent penalty according to the present invention;
fig. 3 is a functional block diagram of a sector complexity evaluation system based on dynamic air traffic image and depth metric learning in accordance with the present invention.
Detailed Description
To make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings, and it is apparent that the described embodiments are some, but not all embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
This embodiment 1 provides a metric loss function, including: selecting an agent from each category of air traffic images as an anchor point sample p; pulling all positive samples (samples with the same category as the proxy anchor point p) in the batch to be within a boundary alpha-m, and pushing all negative samples (samples with different categories from the proxy anchor point p) to be outside the boundary alpha, so that an interval m is kept between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+；
wherein x is for the imageEmbedding; p is an agent; d(x,p)Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only when the categories of x and p are the same, y is 1, otherwise y is 0; the sequencing agent anchor loss is a measurement loss function; a new metric loss function for depth metric learning, namely sequencing agent anchor loss is constructed, and the depth metric learning technology is used for carrying out non-linear mapping from original images to semantic feature embedding space in the air, so that embedding vectors of semantically similar samples in the embedding space are closer, and samples with different semantemes are separated from each other, and more accurate sector operation complexity evaluation is realized.
Example 2
Fig. 1 is a flow chart of a sector complexity evaluation method based on dynamic air traffic image and depth metric learning according to the present invention.
As shown in fig. 1, on the basis of embodiment 1, this embodiment 2 further provides a sector complexity evaluation method based on dynamic air traffic image and depth metric learning, including: carrying out sector operation complexity grade marking on sector dynamic traffic data of the empty domain sector; constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking; constructing a measurement loss function; training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and sector operation complexity evaluation is carried out according to the trained network model, so that nonlinear mapping from original image air learning to semantic feature embedding space by using a depth measurement learning technology is realized, embedding vectors of semantic similar samples in the embedding space are closer, and samples with different semantics are separated from each other, so that more accurate sector operation complexity evaluation is realized.
In this embodiment, the metric loss function in embodiment 1 is used as the metric loss function.
In this embodiment, the method for performing sector operation complexity level labeling on sector dynamic traffic data of a null sector includes: acquiring data from an original air traffic operation database of a target airspace sector, and extracting sector dynamic traffic data from the original data according to a preset time-granularity time interval to form a sector dynamic traffic data set of the target airspace sector; dividing the sector dynamic traffic data set according to a preset time period, and carrying out sector operation complexity grade marking on the sector dynamic traffic data corresponding to each time period; the extracted sector dynamic traffic data mainly comprises longitude and latitude information of an aircraft in the sector and flight height, speed and course information of the aircraft; the sector dynamic traffic data set is divided according to preset time periods, sector operation complexity grade marking is carried out on the sector dynamic traffic data corresponding to each time period, specifically, the sector dynamic traffic data of each minute time period is selected as a sample, sector operation complexity grade marking is carried out by a professional air traffic controller, the sector complexity is marked into 5 different grades (1-5 grades), and the higher the number is, the higher the complexity of the target airspace sector is.
In this embodiment, the method for constructing multi-channel air traffic scene image data of an airspace sector and constructing a sector dynamic traffic image library according to sector operation complexity level labels includes: converting the dynamic traffic scene information of the target airspace sector into an air traffic scene multi-channel image according to an MTSI data representation method and sector operation complexity grade marking; according to the longitude and latitude of the dynamic traffic data of each time section, mapping a target airspace sector to a rectangular grid, filling the altitude operation information of the aircraft in the dynamic traffic data of each time section to a corresponding grid position as a pixel value to generate an altitude historical track image channel of the corresponding time section, and similarly, filling the speed operation information of the aircraft in the dynamic traffic data of each time section to a new rectangular grid to generate a speed historical track image channel of the corresponding time section. Predicting the track of the aircraft within 3 minutes according to the longitude and latitude, the speed and the course of the last track point of the aircraft within each time period, mapping the predicted track into a corresponding new rectangular grid, and generating a conflict predicted track image channel; synthesizing three-channel air traffic scene image data according to a height historical track image channel, a speed historical track image channel and a collision prediction track image channel of the aircraft, setting the initial size of an image to be 173 × 3, associating the image with corresponding sector operation complexity grade labels to obtain a sector dynamic traffic image database, and dividing the sector dynamic traffic image database into a training data set and a test data set according to the ratio of 7: 3.
Fig. 2 is a schematic diagram of a sorting agent penalty according to the present invention.
In this embodiment, the method for constructing the metric loss function includes: designing a new metric loss function for deep metric learning, namely, sequencing proxy anchor loss; in order to better measure the similarity between image data of multi-channel air traffic scenes, the embodiment designs a proxy-based ordered list loss function (RPAL) according to the advantages of the proxy anchor loss function (PAL) and the ordered list loss function (RLL);
as shown in fig. 2, each shape in the figure represents a category, a circle P at the center represents an agent anchor point of the category, and the purpose of sorting agent anchor losses is to select an agent from air traffic images of each category as an anchor point sample P by taking the agent anchor point P as the center;
pulling all positive samples (samples with the same category as the proxy anchor point p) in the batch to be within a boundary alpha-m, and pushing all negative samples (samples with different categories from the proxy anchor point p) to be outside the boundary alpha, so that an interval m is kept between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+formula (1);
wherein x is an insertion corresponding to an image; p is an agent; d(x,p)Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only if the categories of x and p are the same (y)x＝ypWhen y is 1, otherwise y is 0;
non-trivial sample pairs are mined through formula (1), i.e. data points with non-zero loss violate the constraints in the formula, to achieve fast convergence; according to the standard proxy distribution setting in the proxy anchor loss, distributing a proxy for each class, and referring to the form of a loss function of the ordered list, the total loss of the ordered proxy anchor loss is as follows:
wherein, P is the set of all agents; p + is the positive proxy set of data in the batch; for each agent p, the embedded vectors X are divided into two groups,
set of positive embedded vectors for p, Xp -Is a negative set of embedded vectors of p, and
for each proxy anchor point p, in order to better utilize the large number of non-trivial negative samples that exist, the weights of the negative samples set in the loss of the sorted list are referred to, according to the degree of violation of the constraint by each pair of negative samples, according to wxp＝exp(T(α-d(x,p))),
Weighting the negative samples;
wherein T is a parameter controlling the degree of weighting of negative samples, when T is 0, it treats all non-trivial counterexamples equally, if T is + ∞, it will become the most difficult counterexample to mine; and taking the loss of the sequencing agent anchor as a loss function, and guiding the iterative optimization of the network model in the training process of the network model.
In this embodiment, the method for training a network model according to multi-channel air traffic scene image data and a metric loss function includes: performing on a training data set using a torchvision extension tool in a Pytrch frameworkPreprocessing, namely performing standardization processing on pixel values of images in a training data set, performing data enhancement on input images through horizontal overturning and random cutting in the training process, and adjusting the sizes of the images to 224 × 3; adopting GoogleNet V2 pre-trained on the ImageNet data set and subjected to batch standardization as an embedded network model, modifying the size of the last full-connection layer according to the dimensionality of an embedded vector, and carrying out normalization processing on the final output by using L _2 standardization; inputting the preprocessed training data set into a GoogleNet V2 network model for 80-generation training, and setting the initial learning rate to be 10-4Using an AdamW optimizer, the weight decay rate is set to 10-4Sampling each batch of input images by using a random sampling strategy during training; in the training process, the sequencing agent anchor loss is used as a loss function, and an AdamW optimizer is used for continuously iteratively optimizing a target loss function so as to train the network model to obtain an optimal network model; for the selection of the proxy points in the sequencing proxy anchor loss, one proxy is designated for each class according to the setting in the proxy anchor loss, and the proxies are initialized by using normal distribution to ensure that the proxies are uniformly distributed on a unit hypersphere, and the values of several important hyper-parameters in the sequencing proxy anchor loss are set as alpha being 1.4, m being 0.4 and T being 20.
In this embodiment, the method for evaluating the complexity of sector operation according to the trained network model includes: carrying out standardized processing and center cutting on the test set data; inputting the preprocessed image in the test set data into an optimal network model to obtain an embedded vector with semantic similarity of the test data; and taking the embedded vector with semantic similarity as the input of a subsequent 1-NN classification algorithm, classifying the sector operation complexity, finishing the evaluation of the target airspace sector operation complexity, and realizing more accurate evaluation of the sector operation complexity.
Example 3
Fig. 3 is a functional block diagram of a sector complexity evaluation system based on dynamic air traffic image and depth metric learning in accordance with the present invention.
As shown in fig. 3, on the basis of embodiment 2, this embodiment 3 further provides a sector complexity evaluation system based on dynamic air traffic image and depth metric learning, including: the marking module is used for marking the sector operation complexity level of the sector dynamic traffic data of the space domain sector; the database module is used for constructing multi-channel air traffic scene image data of an airspace sector and constructing a sector dynamic traffic image database according to sector operation complexity grade marks; the loss function module is used for constructing a measurement loss function; the network model training module trains a network model according to the multi-channel air traffic scene image data and the measurement loss function; and the complexity evaluation module is used for evaluating the operation complexity of the sector according to the trained network model.
In this embodiment, the specific functional methods of the modules are described in detail in embodiment 2, and are not described again in this embodiment.
In summary, the sector operation complexity level marking is performed on the sector dynamic traffic data of the empty domain sector; constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking; constructing a measurement loss function; training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and sector operation complexity evaluation is carried out according to the trained network model, so that nonlinear mapping from original image air learning to semantic feature embedding space by using a depth measurement learning technology is realized, embedding vectors of semantic similar samples in the embedding space are closer, and samples with different semantics are separated from each other, so that more accurate sector operation complexity evaluation is realized.
In the embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative, and for example, the flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
In light of the foregoing description of the preferred embodiment of the present invention, many modifications and variations will be apparent to those skilled in the art without departing from the spirit and scope of the invention. The technical scope of the present invention is not limited to the content of the specification, and must be determined according to the scope of the claims.
Claims (8)
1. A metric loss function, comprising:
selecting an agent from each category of air traffic image data as an anchor point sample p;
pulling all positive samples in the batch to be within a boundary alpha-m, pushing all negative samples to be outside the boundary alpha, and keeping an interval m between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+；
wherein x is an insertion corresponding to an image; p is an agent; d(x,p) Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only when the categories of x and p are the same, y is 1, otherwise y is 0;
the ranking proxy anchor loss is a metric loss function.
2. A sector complexity evaluation method based on dynamic air traffic image and depth metric learning is characterized by comprising the following steps:
carrying out sector operation complexity grade marking on sector dynamic traffic data of the empty domain sector;
constructing multi-channel air traffic scene image data of an airspace sector, and constructing a sector dynamic traffic image library according to sector operation complexity level marking;
constructing a measurement loss function;
training a network model according to the multi-channel air traffic scene image data and the measurement loss function; and
and evaluating the operation complexity of the sector according to the trained network model.
3. The sector complexity evaluation method of claim 2,
the method for marking the sector operation complexity level of the sector dynamic traffic data of the empty domain sector comprises the following steps:
acquiring data from an original air traffic operation database of a target airspace sector, and forming a sector dynamic traffic data set of the target airspace sector at different time intervals according to preset time granularity;
and dividing the sector dynamic traffic data set according to a preset time period, and carrying out sector operation complexity grade marking on the sector dynamic traffic data corresponding to each time period.
4. The sector complexity evaluation method of claim 3,
the method for constructing the multi-channel air traffic scene image data of the airspace sector and constructing the sector dynamic traffic image library according to the sector operation complexity level mark comprises the following steps:
converting the dynamic traffic scene information of the target airspace sector into an air traffic scene multi-channel image according to an MTSI data representation method and sector operation complexity grade marking;
generating a height historical track image channel, a speed historical track image channel and a conflict prediction track image channel of a corresponding time period according to the dynamic traffic data of the sectors of each time period, and constructing a multi-channel air traffic scene image;
and associating the multi-channel air traffic scene image generated in each time period with the corresponding sector operation complexity grade label, acquiring a sector dynamic traffic image database, and dividing the sector dynamic traffic image database into a training data set and a test data set.
5. The sector complexity evaluation method of claim 4,
the method for constructing the metric loss function comprises the following steps:
selecting an agent from each category of air traffic images as an anchor point sample p;
pulling all positive samples in the batch to be within a boundary alpha-m, pushing all negative samples to be outside the boundary alpha, and keeping an interval m between the positive sample set and the negative sample set;
the formula for the loss of the sequencing proxy anchor is:
Lm(x,p)＝(1-y)[α-d(x,p)]++y[d(x,p)-(α-m)]+；
wherein x is an insertion corresponding to an image; p is an agent; d(x,p)Is the cosine distance of the embedded x from the proxy anchor point p; [. the]+As a function of the hinge; only when the categories of x and p are the same, y is 1, otherwise y is 0;
the total penalty for the rank proxy anchor penalty is:
wherein, P is the set of all agents; p+A positive proxy set for data in the batch; for each agent p, the embedded vectors X are divided into two groups,
is a set of positive embedded vectors for p,
is a negative set of embedded vectors of p, and
for each proxy anchor p, according to wxp＝exp(T(α-d(x,p))),
Weighting the negative samples;
wherein, T is a parameter for controlling the weighting degree of the negative sample;
and taking the loss of the sequencing agent anchor as a loss function, and guiding the iterative optimization of the network model in the training process of the network model.
6. The sector complexity evaluation method of claim 5,
the method for training the network model according to the multi-channel air traffic scene image data and the metric loss function comprises the following steps:
preprocessing a training data set, standardizing pixel values of images in the training data set, and performing data enhancement on input images through horizontal overturning and random cutting in a training process;
inputting the preprocessed training data set into a network model for training, taking the sequencing proxy anchor loss as a loss function, and continuously iterating and optimizing a target loss function through an optimizer to train the network model to obtain an optimal network model.
7. The sector complexity evaluation method of claim 6,
the method for evaluating the operation complexity of the sector according to the trained network model comprises the following steps:
carrying out standardized processing and center cutting on the test set data;
inputting the preprocessed image in the test set data into an optimal network model to obtain an embedded vector with semantic similarity of the test data;
and taking the embedded vector with semantic similarity as the input of a subsequent 1-NN classification algorithm, classifying the sector operation complexity, and finishing the evaluation of the target airspace sector operation complexity.
8. A sector complexity evaluation system based on dynamic air traffic image and depth metric learning, comprising:
the marking module is used for marking the sector operation complexity level of the sector dynamic traffic data of the space domain sector;
the database module is used for constructing multi-channel air traffic scene image data of an airspace sector and constructing a sector dynamic traffic image database according to sector operation complexity grade marks;
the loss function module is used for constructing a measurement loss function;
the network model training module trains a network model according to the multi-channel air traffic scene image data and the measurement loss function; and
and the complexity evaluation module is used for evaluating the operation complexity of the sector according to the trained network model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110912714.XA CN113592341B (en) | 2021-08-10 | 2021-08-10 | Method and system for evaluating metric loss function and sector complexity |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110912714.XA CN113592341B (en) | 2021-08-10 | 2021-08-10 | Method and system for evaluating metric loss function and sector complexity |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113592341A true CN113592341A (en) | 2021-11-02 |
| CN113592341B CN113592341B (en) | 2024-07-02 |
Family
ID=78256657
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110912714.XA Active CN113592341B (en) | 2021-08-10 | 2021-08-10 | Method and system for evaluating metric loss function and sector complexity |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113592341B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935700A (en) * | 2023-09-18 | 2023-10-24 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855778A (en) * | 2012-09-10 | 2013-01-02 | 南京航空航天大学 | Space-domain sector classification method based on complexity assessment |
| CN103226899A (en) * | 2013-03-19 | 2013-07-31 | 北京工业大学 | Method for dynamically dividing sector based on airspace traffic characteristics |
| CN109886352A (en) * | 2019-03-04 | 2019-06-14 | 北京航空航天大学 | A kind of unsupervised appraisal procedure of airspace complexity |
| CN111626366A (en) * | 2020-05-28 | 2020-09-04 | 南京航空航天大学 | Operation characteristic-based area sector scene similarity identification method |
| FR3099625A1 (en) * | 2019-07-31 | 2021-02-05 | Thales | SYSTEM AND METHOD FOR THE IMPROVED DETERMINATION OF AIR SECTOR COMPLEXITY |
| CN112465199A (en) * | 2020-11-18 | 2021-03-09 | 南京航空航天大学 | Airspace situation evaluation system |
-
2021
- 2021-08-10 CN CN202110912714.XA patent/CN113592341B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855778A (en) * | 2012-09-10 | 2013-01-02 | 南京航空航天大学 | Space-domain sector classification method based on complexity assessment |
| CN103226899A (en) * | 2013-03-19 | 2013-07-31 | 北京工业大学 | Method for dynamically dividing sector based on airspace traffic characteristics |
| CN109886352A (en) * | 2019-03-04 | 2019-06-14 | 北京航空航天大学 | A kind of unsupervised appraisal procedure of airspace complexity |
| FR3099625A1 (en) * | 2019-07-31 | 2021-02-05 | Thales | SYSTEM AND METHOD FOR THE IMPROVED DETERMINATION OF AIR SECTOR COMPLEXITY |
| CN111626366A (en) * | 2020-05-28 | 2020-09-04 | 南京航空航天大学 | Operation characteristic-based area sector scene similarity identification method |
| CN112465199A (en) * | 2020-11-18 | 2021-03-09 | 南京航空航天大学 | Airspace situation evaluation system |
Non-Patent Citations (1)
| Title |
|---|
| 高伟;马岚;王涛波;: "基于调速法的交叉航路空域复杂度研究", 中国民航飞行学院学报, no. 06, 15 November 2016 (2016-11-15) * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935700A (en) * | 2023-09-18 | 2023-10-24 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
| CN116935700B (en) * | 2023-09-18 | 2023-12-05 | 四川大学 | Sector traffic situation prediction method based on multi-source features |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113592341B (en) | 2024-07-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mustafa et al. | Comparing support vector machines with logistic regression for calibrating cellular automata land use change models | |
| JP2021504789A (en) | ESG-based corporate evaluation execution device and its operation method | |
| CN107833213A (en) | A kind of Weakly supervised object detecting method based on pseudo- true value adaptive method | |
| Li | Application of finite mixture of logistic regression for heterogeneous merging behavior analysis | |
| CN112990480B (en) | Method, device, electronic equipment and storage medium for constructing model | |
| CN114722805B (en) | Little sample emotion classification method based on size instructor knowledge distillation | |
| CN110070131A (en) | A kind of Active Learning Method of data-oriented driving modeling | |
| CN108960175A (en) | A kind of licence plate recognition method based on deep learning | |
| CN112990222A (en) | Image boundary knowledge migration-based guided semantic segmentation method | |
| CN111160959A (en) | User click conversion estimation method and device | |
| CN114372532B (en) | Method, device, equipment, medium and product for determining label labeling quality | |
| CN110310012B (en) | Data analysis method, device, equipment and computer readable storage medium | |
| KR102457455B1 (en) | Device and Method for Artwork Price Prediction Using Artificial intelligence | |
| CN113592341A (en) | Measurement loss function, sector complexity evaluation method and system | |
| US20220156519A1 (en) | Methods and systems for efficient batch active learning of a deep neural network | |
| CN116719748B (en) | Scene generation method, device and medium of ship system | |
| CN105654126A (en) | Computing equipment, kernel matrix evaluation method and multi-kernel learning method | |
| CN111832435A (en) | Beauty prediction method and device based on migration and weak supervision and storage medium | |
| CN115936003A (en) | Software function point duplicate checking method, device, equipment and medium based on neural network | |
| US20230252282A1 (en) | Method, server, and system for deep metric learning per hierarchical steps of multi-labels and few-shot inference using the same | |
| CN115422349A (en) | Hierarchical text classification method based on pre-training generation model | |
| CN114510873A (en) | Petroleum logging prediction method and device based on big data | |
| CN113837220A (en) | Robot target identification method, system and equipment based on online continuous learning | |
| KR20220098423A (en) | An apparatus for predicting stock price fluctuation using object detection model | |
| Thompson | Data mining methods and the rise of big data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |