A big data processing method based on convolutional neural network

Abstract

The present invention realizes a big data processing method based on convolutional neural network, in which the function processing of highly dense trajectory data in the terminal area can be processed, and the target tracking framework combining the multi-target detection model of deep learning and the filtering method can track the movement of multiple targets in a long time series after the improvement of multiple modules; it proposes a series of methods for detail enhancement based on optical flow information training and image features, so that the information tracked by the final framework contains more details of the target; finally, the dynamic convolution big data information is tracked by the framework in multiple scenes, and a variety of evaluation methods are designed, which effectively proves the accuracy and efficiency of the tracking method, supports local data processing of directed graph and undirected graph big data, and is easy to add and delete the connection between the features of the convolution big data, that is, it is easy to maintain the connection between the features of the convolution big data.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and specifically relates to a big data processing method based on convolutional neural networks. Background Art

[0002] The Voyage Plan Centralized Processing System is a centralized data processing system that centrally accepts, processes, and publishes dynamic voyage plan messages submitted by shipping companies. Based on a complete and unified pre-voyage plan, the system verifies and reviews the format and content of the pilot plan before distributing it to users for approval. The system's primary goal is to improve the quality of voyage plan data through centralized message acceptance and review, meeting the data needs of users such as regulatory authorities, traffic management users, statistical and clearing agencies, and the military.

[0003] With the increasing variety of route parameter data, integrated data playback in ground support equipment has become a key user requirement. The processing and playback of navigation data and multi-channel audio and video data allows for the integrated display of all three types of data. Currently, ground-based playback equipment for navigation parameter recording systems is relatively independent, and the data playback process lacks robust processing capabilities. Furthermore, even when multiple data types are played back simultaneously, inter-data processing is often limited, lacking a comprehensive processing mechanism. Summary of the Invention

[0004] The purpose of the present invention is to provide a big data processing method based on convolutional neural networks, which can better handle the data playback process of navigation data playback, audio data playback, video and other big data, and can be applied to various data processing and playback within a local area network to solve the processing problems and comprehensive playback problems between various route parameter data.

[0005] The present invention seeks to protect a big data processing method based on a convolutional neural network, characterized by comprising:

[0006] Step 001: Preprocess the newly added batch big data to obtain time-aligned trajectory pairs, extract input features as the input of the convolutional neural network, and then define the convolution kernel function, activation function, pooling layer, fully connected layer, and output layer of the convolutional neural network. Match the current trajectory pair with the trajectory in the collision avoidance knowledge base to obtain the convolutional big data;

[0007] Step 002: Perform function processing on the convolution big data by using cosine similarity, complete convolution distribution function processing of the data to be processed, decompose the grid big data, and describe the decomposition of the grid big data into the convolution big data features and the connections between the convolution big data features;

[0008] Step 003: Using sequence relationship mapping to complete time series function processing on the convolution big data processed by the convolution distribution function, and using an Oracle table of a relational database to store the convolution big data features and the connections between the convolution big data features;

[0009] Step 004: Initialize parameters, including the feature identifier of the post-convolution big data to be processed, the data to be processed and its column members in the column family, the processing depth, and the queue. Dynamic processing of the post-convolution big data is completed through the learning model's post-convolution big data dynamic functionalization processing method and the light tracing post-convolution big data feature functionalization processing method.

[0010] Step 005: According to the identifier of the big data feature after the start of convolution to be processed, query the big data feature after the start of convolution to be processed, insert the big data feature after the start of convolution to be processed into the queue, and end the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and process the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue.

[0011] Furthermore, the step 002 performs function processing on the convolution big data by using cosine similarity, completes convolution distribution function processing of the data to be processed, decomposes the grid big data, and describes the decomposition of the grid big data into convolution big data features and connections between the convolution big data features, and further includes:

[0012] Combine multiple function views with time information to obtain the active ship groups in a period;

[0013] Obtain the connection between data in the convolution of different features, the occupancy of the two regions, and then reflect the passing status of the convolution of the navigation program;

[0014] Classify navigational procedures and show differences within the same category based on their efficiency in handling them between ships;

[0015] Generate trajectories based on the raw data and normalize the data to organize it in time series;

[0016] The connection between the features of the big data after convolution is reflected through the features of the big data after convolution, and the features of the big data after convolution are distinguished by the feature identifier of the big data after convolution;

[0017] Process the convolution characteristics of ships, generate spiral structure views and safety views to show the status of navigation procedure convolution utilization and ship intervals, and generate trajectories;

[0018] The identifier that can globally distinguish each post-convolution big data feature is defined as the post-convolution big data feature identifier, and the post-convolution big data feature identifier corresponds one-to-one to the post-convolution big data feature;

[0019] The similarity matrix between trajectories is constructed by fast dynamic time warping algorithm, and then the trajectories are classified by spectral clustering; the features of track data are extracted, and then the actual navigation procedures passed in the terminal area are reconstructed based on the irregular navigation trajectory data.

[0020] Furthermore, the step 003 uses sequence relationship mapping to complete time series function processing on the convolution big data processed by the convolution distribution function, and uses the relational database Oracle table to store the convolution big data features and the relationship between the convolution big data features. It also includes:

[0021] Representing delay data in convolutions based on time series relationships, a function of delay variations over time, showing relationships between regions to capture and compare the spread of delays;

[0022] The convolutional big data feature identifier is converted into a row keyword of the record; the contact feature, processing feature and other features are respectively converted into a contact column family, column family and other column families of the Oracle table; the data to be processed or the data domain is stored in the corresponding column member of the column family;

[0023] Define the functional form of the delay impact factor, and do not obtain the weight of the functional impact factor in different areas over time.

[0024] Furthermore, the step 004 initializes parameters including feature identifiers of the post-convolution big data to be processed, the data to be processed and its column members in the column family, processing depth, and queue, and completes dynamic processing of the post-convolution big data through dynamic functional processing of the post-convolution big data of the learning model and feature functional processing of the post-convolution big data of the light and shadow tracing method, and further includes:

[0025] The peak of the scalar field corresponding to the density map is marked as the target in the training set. Based on the training results, the model can identify low-density targets in grayscale images;

[0026] Based on the dense optical flow method, the optical flow of all pixels in the image is calculated, the optical flow algorithm is selected to obtain the optical flow field of a pair of images, and the motion field between the paired images is calculated;

[0027] Set parameters to generate dynamic Perlin noise motion. The initial parameters generate a density map animation that includes a small number of density clumps. These targets have high saliency and short grasp intervals.

[0028] Reset parameters, increase the size of a single density cluster and the number of density clusters contained in a single area, expand the capture range and the interval of data collection;

[0029] Expand the collected data set. The expanded data set is also used for model training. After obtaining a certain amount of data, repeat the above steps until enough data is collected.

[0030] Through the multi-target tracking model, the multi-target tracking task is decomposed into target detection and motion detection, image features are extracted, and ResNet is selected to replace VGG in the original model;

[0031] The algorithm predicts the possible states of a moving target and compares the predicted target state with the target's measured state at the next moment. Based on the difference between the predicted and actual target states, the algorithm corrects the matching relationship, ultimately confirming the target's positional movement over time and further describing the target's movement trends.

[0032] Furthermore, the step 005 queries the feature of the big data after the start of convolution to be processed according to the identifier of the feature of the big data after the start of convolution to be processed, inserts the feature of the big data after the start of convolution to be processed into a queue, terminates the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and processes the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue, further comprising:

[0033] When the depth of the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature is greater than the processing depth, the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature are added to the queue, the team's first post-convolution big data feature has been visited and the current processing depth of the team's first post-convolution big data feature is recorded, and the team's first post-convolution big data feature element is deleted;

[0034] Determine whether the current processing depth is greater than the initialized processing depth and the queue is empty. If so, end the process; otherwise, go to step 003.

[0035] The present invention implements a big data processing method based on convolutional neural networks, which can handle function processing of highly dense trajectory data in the terminal area. The target tracking framework combines a deep learning multi-target detection model with a filtering method. After multi-module improvements, it can track the movement of multiple targets in a longer time series. It proposes a series of methods for optical flow information training and detail enhancement based on image features, so that the information tracked by the final framework contains more target details. Finally, the framework is used to track dynamic convolution big data information in multiple scenarios, and a variety of evaluation methods are designed to effectively prove the accuracy and efficiency of the tracking method. It supports local data processing of directed and undirected graph big data, and it is easy to add and delete connections between convolution big data features, that is, it is easy to maintain connections between convolution big data features. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 This is a workflow diagram of a big data processing method based on convolutional neural networks involved in the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the accompanying drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0039] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, features defined as "first" or "second" may explicitly or implicitly include one or more of such features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0040] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances.

[0041] The present invention seeks to protect a big data processing method based on a convolutional neural network, characterized by comprising:

[0042] Step 001: Preprocess the newly added batch big data to obtain time-aligned trajectory pairs, extract input features as the input of the convolutional neural network, and then define the convolution kernel function, activation function, pooling layer, fully connected layer, and output layer of the convolutional neural network. Match the current trajectory pair with the trajectory in the collision avoidance knowledge base to obtain the convolutional big data;

[0043] Step 002: Perform function processing on the convolution big data by using cosine similarity, complete convolution distribution function processing of the data to be processed, decompose the grid big data, and describe the decomposition of the grid big data into the convolution big data features and the connections between the convolution big data features;

[0044] Step 003: Using sequence relationship mapping to complete time series function processing on the convolution big data processed by the convolution distribution function, and using an Oracle table of a relational database to store the convolution big data features and the connections between the convolution big data features;

[0045] Step 004: Initialize parameters, including the feature identifier of the post-convolution big data to be processed, the data to be processed and its column members in the column family, the processing depth, and the queue. Dynamic processing of the post-convolution big data is completed through the learning model's post-convolution big data dynamic functionalization processing method and the light tracing post-convolution big data feature functionalization processing method.

[0046] Step 005: According to the identifier of the big data feature after the start of convolution to be processed, query the big data feature after the start of convolution to be processed, insert the big data feature after the start of convolution to be processed into the queue, and end the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and process the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue.

[0047] Throughout this specification, reference to terms such as "one embodiment," "some embodiments," "illustrative embodiments," "examples," "specific examples," or "some examples" means that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, illustrative uses of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

[0048] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to the embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the claims and their equivalents.

[0049] Furthermore, the step 002 performs function processing on the convolution big data by using cosine similarity, completes convolution distribution function processing of the data to be processed, decomposes the grid big data, and describes the decomposition of the grid big data into convolution big data features and connections between the convolution big data features, and further includes:

[0050] Combine multiple function views with time information to obtain the active ship groups in a period;

[0051] Obtain the connection between data in the convolution of different features, the occupancy of the two regions, and then reflect the passing status of the convolution of the navigation program;

[0052] Classify navigational procedures and show differences within the same category based on their efficiency in handling them between ships;

[0053] Generate trajectories based on the raw data and normalize the data to organize it in time series;

[0054] The connection between the features of the big data after convolution is reflected through the features of the big data after convolution, and the features of the big data after convolution are distinguished by the feature identifier of the big data after convolution;

[0055] Process the convolution characteristics of ships, generate spiral structure views and safety views to show the status of navigation procedure convolution utilization and ship intervals, and generate trajectories;

[0056] The identifier that can globally distinguish each post-convolution big data feature is defined as the post-convolution big data feature identifier, and the post-convolution big data feature identifier corresponds one-to-one to the post-convolution big data feature;

[0057] The similarity matrix between trajectories is constructed by fast dynamic time warping algorithm, and then the trajectories are classified by spectral clustering; the features of track data are extracted, and then the actual navigation procedures passed in the terminal area are reconstructed based on the irregular navigation trajectory data.

[0058] Data processing and functionalization are performed based on cosine similarity. The data construction includes time, flight number, longitude, latitude, altitude, ground speed, and heading. These signals are collected in a time series, covering an area larger than the actual terminal area. It is necessary to determine which portion of the collected data belongs to the terminal area under study. Ship data on high-altitude routes can interfere with modeling. Secondly, different ships transmit signals at different frequencies. Over a period of time, data from some ships may be too limited to be modeled, and this data should be properly excluded. Finally, ship data collected over a period of time may be part of a complete voyage, and accurately classifying trajectories based on partial data is also a challenge. Furthermore, as mentioned above, airspace processing in the terminal area is more microscopic than route processing, so data screening cannot be overly extensive. However, to avoid unnecessary computational overhead, data simplification is necessary. To normalize the data, trajectories are rasterized into a grid using the Bresenham algorithm and resampled. A similarity matrix is ​​constructed by calculating the similarity between any two trajectories. Trajectory data is sorted by flight number using the following algorithm:

[0059] a X，Y =Dist(X i , Y j )+min[D(X i-1 , Y j ), D(X i , Y j-1 ), D(X i-1 , Y j-1 )]

[0060] Where a is an element in the distance matrix, X and Y represent two time series data, which correspond to the trajectories collected by different ships, i and j represent the lengths of the corresponding X and Y data respectively. The distances calculated here are based on Euclidean distances, and the calculation process uses dynamic programming technology. The greater the difference between the two sequences X and Y, the greater the difference between a and Y. X，Y The higher the value, the higher the probability of trajectories being detected. A similarity matrix of all trajectories is further constructed in the terminal region, referred to as the sim matrix. Each element in sim is the difference between the maximum value in the distance matrix calculated by the above formula and the corresponding element. sim is set as a kernel function (such as the RBF kernel function), and then the eigenvalues ​​are calculated using the diagonal matrix and the Laplace matrix. In the next step, the data is clustered again from a new matrix consisting of eigenvectors.

[0061] Based on the statistical chart's airspace information overview, the parallel coordinate axes have been functionalized to represent airspace altitude transits at different time periods. For safety monitoring based on the relative position vector view, the center circle is set to the minimum safe separation distance (5 kilometers) in the civil aviation system, and the area between the two circles changes proportionally based on the difference in the set values. If the user wishes to view the data at a larger convolution scale, the safe separation distance value can be increased. Conversely, decreasing the horizontal safe separation distance value increases the distance between identical elements in the view. In this view, the center corresponds to the location of the runway, and all vessel positions are converted to polar coordinates. A discrete color coding scheme is used for the convolution functionalization based on a spiral structure model. The color changes in the spiral structure indicate the average speed of vessels in different areas. Safety separation processing, by reducing the altitude difference between flights, identifies vessels that are too close together at the same altitude. Subsequently, the safe separation standard was reduced to 10 kilometers, a value commonly used as a universal shipping safety standard.

[0062] Furthermore, the step 003 uses sequence relationship mapping to complete time series function processing on the convolution big data processed by the convolution distribution function, and uses the relational database Oracle table to store the convolution big data features and the relationship between the convolution big data features. It also includes:

[0063] Representing delay data in convolutions based on time series relationships, a function of delay variations over time, showing relationships between regions to capture and compare the spread of delays;

[0064] The convolutional big data feature identifier is converted into a row keyword of the record; the contact feature, processing feature and other features are respectively converted into a contact column family, column family and other column families of the Oracle table; the data to be processed or the data domain is stored in the corresponding column member of the column family;

[0065] Define the functional form of the delay impact factor, and do not obtain the weight of the functional impact factor in different areas over time.

[0066] Key information in the data includes flight number, departure sea area, destination, estimated departure time, actual departure time, estimated arrival time, actual arrival time, and delay duration. A dynamic delay representation based on the Poisson distribution is generated using a fast Poisson disk sampling method, generating a Poisson distribution centered on the sea area of ​​interest. Each data point in the distribution represents a delay event. To ensure a consistent distribution shape for each region, a Poisson disk is pre-generated in the system as a template, ensuring that all data points in this work follow the same Poisson distribution.

[0067] The initial Poisson distribution is centered at the location of the sea area of ​​interest. When a delay level jump occurs, the distribution center shifts toward the source data causing the delay, and newly generated points are distributed around the updated distribution center. The previous step is repeated for all relevant sea areas; each iteration generates a new point based on the existing Poisson distribution shape, or a new point around the new distribution shape due to the change in delay level. If a delay jump occurs, the kernel is shifted by the same distance as the Poisson distribution interval, and the new kernel position is checked for overlap with existing kernels.

[0068] Generate a delay variation field, generating a dynamic Poisson distribution of delays based on a continuous time series of delay events. Convert a contour map based on four delay levels to display the delay status at time t. In the experiment, when converting a color heatmap to a contour map, the density generated by the data is used as a scalar value to calculate the contour lines. StreamMap is used to calculate the variation field between two consecutive delay heatmaps, and these fields are then displayed as arrow distributions. Arrows indicate the trend of delay variation, and the vector field is projected onto the contour plane. The density of the arrows used to indicate field variation varies according to the length of the boundary, ensuring that the direction of change is clearly displayed regardless of the length of the field line.

[0069] Furthermore, the step 004 initializes parameters including feature identifiers of the post-convolution big data to be processed, the data to be processed and its column members in the column family, processing depth, and queue, and completes dynamic processing of the post-convolution big data through dynamic functional processing of the post-convolution big data of the learning model and feature functional processing of the post-convolution big data of the light and shadow tracing method, and further includes:

[0070] The peak of the scalar field corresponding to the density map is marked as the target in the training set. Based on the training results, the model can identify low-density targets in grayscale images;

[0071] Based on the dense optical flow method, the optical flow of all pixels in the image is calculated, the optical flow algorithm is selected to obtain the optical flow field of a pair of images, and the motion field between the paired images is calculated;

[0072] Set parameters to generate dynamic Perlin noise motion. The initial parameters generate a density map animation that includes a small number of density clumps. These targets have high saliency and short grasp intervals.

[0073] Reset parameters, increase the size of a single density cluster and the number of density clusters contained in a single area, expand the capture range and the interval of data collection;

[0074] Expand the collected data set. The expanded data set is also used for model training. After obtaining a certain amount of data, repeat the above steps until enough data is collected.

[0075] Through the multi-target tracking model, the multi-target tracking task is decomposed into target detection and motion detection, image features are extracted, and ResNet is selected to replace VGG in the original model;

[0076] The algorithm predicts the possible states of a moving target and compares the predicted target state with the target's measured state at the next moment. Based on the difference between the predicted and actual target states, the algorithm corrects the matching relationship, ultimately confirming the target's positional movement over time and further describing the target's movement trends.

[0077] Create a density map animation and extract a dataset in the form of a density map from it. Cropping and resizing the images serves as the learning data. The multiple objects in the two final partitioned images are treated as two disjoint sets in a bipartite graph. Each object can be treated as a vertex in the graph, transforming the object matching problem into a vertex matching problem. The algorithm finally recursively converts all vertices in the set to saturation points to ensure that all vertices have matching relationships.

[0078] Furthermore, the step 005 queries the feature of the big data after the start of convolution to be processed according to the identifier of the feature of the big data after the start of convolution to be processed, inserts the feature of the big data after the start of convolution to be processed into a queue, terminates the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and processes the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue, further comprising:

[0079] When the depth of the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature is greater than the processing depth, the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature are added to the queue, the team's first post-convolution big data feature has been visited and the current processing depth of the team's first post-convolution big data feature is recorded, and the team's first post-convolution big data feature element is deleted;

[0080] Determine whether the current processing depth is greater than the initialized processing depth and the queue is empty. If so, end the process; otherwise, go to step 003.

[0081] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A big data processing method based on convolutional neural network, characterized in that: include: Step 001, preprocess the newly added batch big data to obtain time-aligned trajectory pairs, extract input features as the input of the convolutional neural network, and then define the convolution kernel function, activation function, pooling layer, fully connected layer, and output layer of the convolutional neural network, match the current trajectory pair with the trajectory in the collision avoidance knowledge base, and obtain the convolutional big data; Step 002, completing function processing on the convolution big data through cosine similarity, completing convolution distribution function processing of the data to be processed, decomposing the grid big data, and describing the decomposition of the grid big data as the convolution big data features and the connections between the convolution big data features; Step 003, using sequence relationship mapping to complete time series functional processing of the convolution big data processed by the convolution distribution function, and using the relational database Oracle table to store the convolution big data features and the connection storage data between the convolution big data features; Step 004, initializing parameters: including the feature identifier of the big data after convolution to be processed, the data to be processed and the column members in the column family, the processing depth and the queue, and completing the dynamic processing of the big data after convolution through the dynamic functional processing of the big data after convolution of the learning model and the feature functional processing method of the big data after convolution of the light and shadow tracing; Step 005, according to the identification of the big data feature after the start of convolution to be processed, query the big data feature after the start of convolution to be processed, insert the big data feature after the start of convolution to be processed into the queue, end the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and process the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue; The step 002, performing function processing on the convolution big data by cosine similarity, completing convolution distribution function processing of the data to be processed, decomposing the grid big data, and describing the decomposition of the grid big data as the convolution big data features and the connections between the convolution big data features, also includes: Combine multiple function views with time information to obtain the active ship groups in a period; Obtain the connection between data in the convolution of different features, the occupancy of the two regions, and then reflect the passing status of the convolution of the navigation program; Classify navigational procedures and show differences within the same category according to the efficiency with which they are handled, based on the spacing between ships; Generate trajectories based on raw data and normalize the data to organize them in time series; The convolutional big data features are used to reflect the relationship between the convolutional big data features, and the convolutional big data feature identifiers are used to distinguish different convolutional big data features; Process the convolution characteristics of the ship, generate spiral structure view and safety view to show the status of navigation procedure convolution utilization and ship interval, and generate trajectory; The identifier that can globally distinguish each post-convolution big data feature is defined as the post-convolution big data feature identifier, and the post-convolution big data feature identifier corresponds to the post-convolution big data feature one by one; The similarity matrix between trajectories is constructed by fast dynamic time warping algorithm, and then the trajectories are classified by spectral clustering; The features of the track data are extracted, and then the actual passing navigation procedures in the terminal area are reconstructed based on the irregular navigation track data.

2. A method for processing big data based on a convolutional neural network as claimed in claim 1, characterized in that: include: The step 003 uses sequence relationship mapping to complete time series functional processing of the convolution big data processed by the convolution distribution function, and uses the relational database Oracle table to store the convolution big data features and the connection storage data between the convolution big data features, and also includes: Representing the delay data in convolutions based on time series relationships, a function of the delay variation over time, showing the relationship between regions to capture and compare the spread of delays; The convolutional big data feature identifier is converted into a row keyword of the record; the contact feature, processing feature and other features are respectively converted into the contact column family, column family and other column families of the Oracle table; the data to be processed or the data domain is stored in the corresponding column member of the column family; Define the functional form of the delay impact factor so that the functional impact factor changes over time in different areas without obtaining weights.

3. The method for processing big data based on convolutional neural network according to claim 1, characterized in that: include: The step 004, initializing parameters: including the feature identifier of the big data after convolution to be processed, the data to be processed and the column members in the column family, the processing depth and the queue, completing the dynamic processing of the big data after convolution by the learning model and the feature functional processing method of the big data after convolution by light and shadow tracing, and also includes: The peak of the scalar field corresponding to the density map is marked as the target in the training set. Based on the training results, the model can identify low-density targets in grayscale images. Based on the dense optical flow method, the optical flow of all pixels in the image is calculated, the optical flow algorithm is selected to obtain the optical flow field of a pair of images, and the motion field between the paired images is calculated; Set the parameters to generate dynamic Perlin noise motion. The initial parameters generate a density map animation that includes a small number of density clumps. These targets have high saliency and short grasp intervals. Reset parameters, increase the size of a single density cluster and the number of density clusters contained in a single area, expand the capture range and the interval of data collection; Expand the collected data set. The expanded data set is also used for model training. After obtaining a certain amount of data, repeat the above steps until enough data is collected. Through the multi-target tracking model, the multi-target tracking task is decomposed into target detection and motion detection, image features are extracted, and ResNet is selected to replace VGG in the original model; Predict the possible state of the moving target, compare the predicted target state with the target measured state at the next moment, correct the matching relationship according to the difference between the predicted target state and the actual state, finally confirm the process of the target's position moving over time, and then further describe the target's movement trend.

4. The method for processing big data based on convolutional neural network according to claim 1, characterized in that: include: The step 005, according to the feature identifier of the big data after the start of convolution to be processed, queries the big data feature after the start of convolution to be processed, inserts the big data feature after the start of convolution to be processed into the queue, terminates the processing when the current processing depth is greater than the initialized processing depth and the queue is empty, and processes the column members in the column family of the Oracle table corresponding to the big data feature after convolution at the head of the queue, and also includes: When the depth of the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature is greater than the processing depth, add the adjacent post-convolution big data features that have not been visited by the team's first post-convolution big data feature to the queue, record that the team's first post-convolution big data feature has been visited and the current processing depth of the team's first post-convolution big data feature, and delete the team's first post-convolution big data feature element; Determine whether the current processing depth is greater than the initialized processing depth and the queue is empty. If so, end the process; otherwise, go to step 003.

Images