Port berth information extraction method based on remote sensing image and AIS

Abstract

The invention discloses a port berth information extraction method based on a remote sensing image and an AIS (Automatic Identification System). The method comprises the following steps: S1, acquiring a historical remote sensing image of a port area and matched ship AIS data in the port area in the historical remote sensing image; performing image preprocessing on the remote sensing frame image to obtain a port type sample set and an optimized remote sensing image set; s2, performing grid division on a port region in the optimized remote sensing image set to obtain a region grid map, and obtaining an AIS data space-time diagram according to the ship AIS data; s3, constructing a port berth information extraction model, which comprises training a pre-established Vision Transform model according to the port type sample set to obtain a port type prediction model; training and constructing an ST-CNN model for identifying port berth information according to the AIS data space-time diagram, and obtaining a port berth information prediction model; and S4, on the basis of the constructed port berth information extraction model, port berth information extraction is performed on the remote sensing image of the port area after image preprocessing. The method solves the problem that the geographic precision of AIS data may be poor in some remote areas or areas with poor signals, so that the identification of the parking position is inaccurate. Secondly, the updating frequency of the AIS data is relatively low, so that the actual condition of the port cannot be reflected in time when the port layout is changed quickly; in addition, the AIS data lacks description capability for the type and layout information of port facilities, so that refined berth type identification is difficult to realize.

Description

Technical Field

[0001] The present invention relates to the field of smart maritime technology, and in particular to a method for extracting port berth information based on remote sensing images and AIS. Background Art

[0002] Berths are vital infrastructure in port operations and are important places for ships to dock, cargo to be loaded and unloaded, and personnel to board and disembark. With the rapid development of global maritime trade, the scale and complexity of ports are increasing. How to effectively extract the types of port berths and their spatial location information has become an important research direction for port management and maritime monitoring. Especially in different types of ports such as container ports, oil and gas ports, and bulk cargo ports, rapid identification of berth information plays a key role in improving port logistics efficiency.

[0003] At present, berth extraction technology mainly relies on AIS data for analysis. The AIS system can provide information such as the dynamic position, speed, and heading of the ship. Through these data, it can be inferred whether the ship is docked at the port berth and further determine the use of the berth. However, the AIS data itself mainly reflects the behavior of the ship and cannot directly provide static information about the port berth. In addition, the coverage of AIS data is limited during maritime monitoring, and there may be signal loss in some areas, which makes the berth extraction technology that relies on AIS data have certain limitations.

[0004] Although the existing berth extraction technology has solved the problem of dynamic monitoring of ships to a certain extent, it still has some shortcomings in practical applications. First, the geographical accuracy of AIS data may be poor in some remote or poorly signaled areas, resulting in inaccurate identification of berth locations. Secondly, the update frequency of AIS data is low, and when faced with rapidly changing port layouts (such as berth expansion, new facilities, etc.), it is unable to reflect the actual situation of the port in a timely manner. In addition, AIS data lacks the ability to describe the type and layout information of the port facilities themselves, making it difficult to achieve refined berth type identification. Summary of the invention

[0005] The present invention provides a method for extracting port berth information based on remote sensing images and AIS to overcome the above technical problems.

[0006] In order to achieve the above object, the technical solution of the present invention is:

[0007] A method for extracting port berth information based on remote sensing images and AIS specifically comprises the following steps:

[0008] S1: Obtain historical remote sensing images of the port area and AIS data of ships matching the port area in the historical remote sensing images;

[0009] And the historical remote sensing image is composed of a series of remote sensing frame images;

[0010] Performing image preprocessing on the remote sensing frame image;

[0011] The image preprocessing includes marking the port type of the port area in the remote sensing frame image to obtain a port type sample set;

[0012] Or annotate potential berth areas in the port area in the remote sensing frame image to obtain an optimized remote sensing image set;

[0013] S2: Gridding the port area in the optimized remote sensing image set to obtain a regional grid map, and based on the regional grid map, obtaining an AIS data spatiotemporal map according to the ship AIS data;

[0014] The AIS time-space diagram includes at least the ship's position, the ship's berthing time, the ship's port entry and exit time, the ship's speed and the ship's heading;

[0015] S3: Construct a port berth information extraction model;

[0016] And the port berth information extraction model includes

[0017] A pre-established Vision Transformer model is trained according to the port type sample set to obtain a port type prediction model for identifying the port type;

[0018] and the ST-CNN model for identifying port berth information constructed by training the spatiotemporal graph of AIS data, and the obtained port berth information prediction model;

[0019] And the port berth information at least includes the berth location and berth type;

[0020] S4: Based on the constructed port berth information extraction model, the port type and port berth information of the remote sensing image of the port area after image preprocessing is extracted.

[0021] Furthermore, the S2 specifically includes the following steps:

[0022] S21: gridding the port area in the optimized remote sensing image set to obtain a regional grid map with M rows and N columns;

[0023] And any unit grid in the regional grid map represents a fixed geographical area;

[0024] S22: According to the ship AIS data, the longitude and latitude of the port area position are converted into the grid position index (i, j) of the regional grid map;

[0025] And the expression of the grid position index (i,j) is

[0026]

[0027] Where: x min ,y min Respectively represent the minimum longitude and minimum latitude of the port area; x range ,y range Respectively represent the longitude range and latitude range of the port area; i represents the row number of the grid position index; j represents the column number of the grid position index;

[0028] S23: Setting the time step, determining the ship AIS data corresponding to a number of time periods, and discretizing the data to obtain the time step index t;

[0029] And the expression of time step index t is

[0030]

[0031] Where: t start Indicates the start time; t step Indicates the length of each time step; timestamp indicates the timestamp;

[0032] S24: According to the grid position index (i, j) and the time step index t, the three-dimensional matrix X used to obtain the AIS data spatiotemporal diagram is obtained, and its expression is:

[0033]

[0034] Where: X(i,j,t) represents the decision variable of whether the ship only appears at the position index (i,j) in the time step index t.

[0035] Furthermore, the pre-established Vision Transformer model is trained in S3 according to the port type sample set to obtain the port type prediction model. The training process is as follows:

[0036] S31: randomly dividing the port type sample set into a training set and a test set according to a preset ratio;

[0037] S32: inputting the training set into a pre-created Vision Transformer model for model training;

[0038] And based on the constructed model loss function, confirm whether the output of the trained Vision Transformer model converges;

[0039] If the output of the trained Vision Transformer model converges, execute step S33;

[0040] S33: Evaluate the model accuracy of the trained Vision Transformer model based on the test set;

[0041] The indicators for evaluating the accuracy of the model are the recall rate and accuracy of the model;

[0042] If it is confirmed that the model accuracy evaluation meets the preset model accuracy threshold, the trained VisionTransformer model is confirmed to be a port type prediction model for identifying remote sensing images of port areas;

[0043] If it is confirmed that the model accuracy evaluation does not meet the preset model accuracy threshold, the weight parameters of the trained Vision Transformer model are adaptively adjusted based on the back propagation method, and step S32 is repeated.

[0044] Furthermore, the expression of the model loss function constructed in S32 is:

[0045]

[0046] Where: C represents the port category; w i represents the weight of the i-th type of port; y i Indicates the real label of the port type, Indicates the model’s predicted port type label.

[0047] Furthermore, the ST-CNN model constructed in S3 for identifying port berth information includes an input layer, a spatial convolution layer, a temporal convolution layer, a fully connected layer, and an output layer;

[0048] The input layer is used to input the AIS data spatiotemporal graph into the spatial convolution layer and the temporal convolution layer respectively;

[0049] The spatial convolution layer is used to extract the distribution characteristics of AIS data in two-dimensional space according to the AIS data spatiotemporal graph, and obtain the spatial position characteristics of the spatial position relationship of different ships in the port area;

[0050] And transmitting the spatial position feature to the fully connected layer;

[0051] The temporal convolution layer is used to extract the time series characteristics of the AIS data according to the AIS data spatiotemporal graph, and obtain the movement trajectory and docking behavior characteristics of the ship at different times;

[0052] And transmit the moving trajectory and docking behavior characteristics at different times to the fully connected layer;

[0053] The fully connected layer is used to perform feature fusion on the output of the spatial convolution layer and the output of the temporal convolution layer to obtain a high-dimensional feature vector for representing the spatial position and temporal behavior of the ship, and obtain the port berth position and berth type in the potential berth area according to the high-dimensional feature vector;

[0054] The output layer is used to output the results of port berth locations and berth types.

[0055] Furthermore, the formula for obtaining the high-dimensional feature vector used to represent the spatial position and temporal behavior of the ship is:

[0056] h t =f(W t *X t +W s *X s +b)

[0057] Where: h t represents a high-dimensional feature vector; W t ,W s Respectively represent the convolution kernels of the temporal convolution layer and the spatial convolution layer; X t ,X s They represent temporal features and spatial features respectively; f represents the activation function; and b represents the bias term.

[0058] Beneficial effects: The present invention provides a port berth information extraction method based on remote sensing images and AIS. By gridding the port area in the optimized remote sensing image set to obtain a regional grid map, and obtaining an AIS data spatiotemporal map according to the ship AIS data, the ability of AIS data to describe the type and layout information of the port facilities themselves can be improved, and a port berth information extraction model can be constructed. By training a pre-established Vision Transformer model according to a port type sample set, a port type prediction model is obtained, which can automatically predict the port type, reducing the need for manual intervention. Compared with traditional manual labeling and classification methods, work efficiency can be greatly improved and labor costs can be reduced. By training an ST-CNN model for identifying port berth information according to the AIS data spatiotemporal map, the port berth information prediction model obtained can better capture the changes of ships in time and space, improve the accuracy of berth information recognition, and realize refined berth type recognition, and then based on the constructed port berth information extraction model, the remote sensing image of the port area after image preprocessing is used to extract port berth information. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] 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 for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0060] Figure 1 It is a flow chart of the method for extracting port berth information based on remote sensing images and AIS of the present invention;

[0061] Figure 2 A schematic diagram of potential berth areas marked in this embodiment;

[0062] Figure 3 Schematic diagram of port berth information extraction based on the constructed port berth information extraction model in this embodiment;

[0063] Figure 4 Schematic diagram of the structure of the ST-CNN model constructed in this embodiment for identifying port berth information. DETAILED DESCRIPTION

[0064] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0065] This embodiment provides a method for extracting port berth information based on remote sensing images and AIS. Figure 1 As shown, the specific steps include:

[0066] S1: Obtain historical remote sensing images of the port area and AIS data of ships matching the port area in the historical remote sensing images;

[0067] And the historical remote sensing image is composed of a series of remote sensing frame images;

[0068] Specifically, remote sensing images from the coast of Singapore, Thailand, and Vietnam in June were selected from the remote sensing image library, and their remote sensing frame images were cut into small blocks of 1024×1024 pixels in bmp format through the preset python code;

[0069] Performing image preprocessing on the remote sensing frame image;

[0070] The image preprocessing includes marking the port type of the port area in the remote sensing frame image to obtain a port type sample set;

[0071] Specifically, the port types are marked, including container ports, bulk cargo ports, oil and gas ports, fishing ports, and cruise ports, and sample sets are made. The sample set is randomly divided into a training set and a validation set in a ratio of 8:2; the name of the manual annotation software is QGIS. In this embodiment, a container port refers to a port that is mainly used to handle container cargo, has efficient loading and unloading equipment and a dedicated container yard, and is equipped with large container cranes, automated yard systems, container maintenance and storage facilities; a bulk cargo port refers to a port that is mainly used to handle bulk cargo, such as coal, ore, grain, etc., and is equipped with equipment for loading and unloading bulk cargo, and bulk cargo unloading facilities such as grabs, belt conveyors, storage silos, etc.; an oil and gas port refers to a port that is specifically used to handle liquid cargo, especially oil and natural gas, and has dedicated tanker berths and oil storage facilities, oil and gas storage tanks, oil unloading platforms, dedicated pipelines and pumping equipment; a fishing port refers to a port that is mainly used to handle fishery products, with fishing boat berths and refrigeration facilities; a cruise port provides services specifically for cruise ships, passenger facilities and dedicated cruise terminals.

[0072] like Figure 2 As shown, or manually marking the potential berth area of ​​the port area in the remote sensing frame image to form a vector range frame to obtain an optimized remote sensing image set;

[0073] S2: Gridding the port area in the optimized remote sensing image set to obtain a regional grid map, and based on the regional grid map, obtaining an AIS data spatiotemporal map according to the ship AIS data;

[0074] The AIS time-space diagram includes at least the ship's position, the ship's berthing time, the ship's port entry and exit time, the ship's speed and the ship's heading;

[0075] The specific steps include:

[0076] S21: According to the range of the remote sensing image, the port area in the optimized remote sensing image set is gridded to obtain a regional grid map with M rows×N columns;

[0077] And any unit grid in the regional grid map represents a fixed geographical area;

[0078] S22: According to the ship AIS data, the longitude and latitude of the port area position are converted into the grid position index (i, j) of the regional grid map;

[0079] And the expression of grid position index (9i,j) is

[0080]

[0081] Where: x min ,ymin Respectively represent the minimum longitude and minimum latitude of the port area; x range ,y range Respectively represent the longitude range and latitude range of the port area; i represents the row number of the grid position index; j represents the column number of the grid position index;

[0082] S23: Setting the time step, determining the ship AIS data corresponding to a number of time periods, and discretizing the data to obtain the time step index t;

[0083] And the expression of time step index t is

[0084]

[0085] Where: t start Indicates the start time; t step Indicates the length of each time step; timestamp indicates the timestamp;

[0086] S24: According to the grid position index (i, j) and the time step index t, the three-dimensional matrix X used to obtain the AIS data spatiotemporal diagram is obtained, that is, each record in the AIS data is mapped to a specific position in the three-dimensional matrix to represent it. Its expression is:

[0087]

[0088] Where: X(i,j,t) represents the decision variable of whether the ship only appears at the position index (i,j) in the time step index t, and if it exists, the value is 1; otherwise, it is 0;

[0089] In this embodiment, for missing or discontinuous ship AIS data, a linear interpolation method can be used to complete the data, and through space and time discretization and linear interpolation steps, a complete M×N×T space-time diagram can be obtained; wherein missing AIS data: refers to the period of time when the ship's AIS signal has not been successfully received or recorded. This may be caused by reasons such as the ship turning off the AIS equipment, signal obstruction, and equipment failure. Discontinuous AIS data: refers to the fact that although there is an AIS signal record, the recording time interval is too long. For example, the interval between two consecutive time points of a ship's AIS data is much longer than the normal time interval. This is usually caused by signal transmission delay or signal loss.

[0090] In summary, for the convenience of subsequent description, if there is no data at a timestamp interval of 15 minutes, it is defined as missing AIS data, and linear interpolation is needed to supplement the data. The specific steps are as follows:

[0091] Assume that this embodiment needs to complete a certain time t between time t1 and time t2 ithe ship position, then

[0092] For time t i the abscissa x of the position i :

[0093]

[0094] For time t i the ordinate y of the position i :

[0095]

[0096] where t1 and t2 are two adjacent time points, and t1 < t i < t2, x1, y1 and x2, y2 are the known position coordinates of these two time points respectively; x i , y i is the coordinate at time t deduced by linear interpolation; i time;

[0097] S3: Construct a port berth information extraction model;

[0098] And the port berth information extraction model includes

[0099] Train a pre-established Vision Transformer model according to the port type sample set to obtain a port type prediction model for identifying port types;

[0100] Specifically, training a pre-established Vision Transformer model according to the port type sample set to obtain a port type prediction model, and its training process is

[0101] S31: Randomly divide the port type sample set into a training set and a test set according to a preset ratio;

[0102] S32: Input the training set into the pre-created Vision Transformer model for model training;

[0103] And based on the constructed model loss function, confirm whether the output of the trained Vision Transformer model converges;

[0104] And the expression of the constructed model loss function is

[0105]

[0106] In the formula: C represents the port category; w i represents the weight of the i-th type of port; y i represents the true label of the port type, Indicates the model predicts the port type label;

[0107] If the output of the trained Vision Transformer model converges, execute step S33;

[0108] S33: Evaluate the model accuracy of the trained Vision Transformer model based on the test set;

[0109] The indicators for evaluating the accuracy of the model are the recall rate and accuracy of the model;

[0110] If it is confirmed that the model accuracy evaluation meets the preset model accuracy threshold, this embodiment uses the validation set to evaluate the accuracy of the trained model. If the recall rate is greater than 90% and the accuracy rate is greater than 90%, it is confirmed that the trained VisionTransformer model is a port type prediction model for identifying remote sensing images of port areas;

[0111] If it is confirmed that the model accuracy evaluation does not meet the preset model accuracy threshold, the weight parameters of the trained Vision Transformer model are adaptively adjusted based on the back propagation method, and step S32 is repeated;

[0112] The ST-CNN model for identifying port berth information is constructed based on the training of the AIS data spatiotemporal graph, and the port berth information prediction model is obtained; and the port berth information at least includes the berth location and berth type; the evaluation index in the training process of the constructed ST-CNN model adopts the cross entropy loss function, and other common classification problem loss functions can also be used, which will not be elaborated here;

[0113] In a specific embodiment, Figure 4 As shown, the ST-CNN model constructed for identifying port berth information includes an input layer, a spatial convolution layer, a temporal convolution layer, a fully connected layer, and an output layer;

[0114] The input layer is used to input the AIS data spatiotemporal graph into the spatial convolution layer and the temporal convolution layer respectively;

[0115] The spatial convolution layer is used to extract the distribution characteristics of AIS data in two-dimensional space according to the AIS data spatiotemporal graph, and obtain the spatial position characteristics of the spatial position relationship of different ships in the port area;

[0116] Specifically, the functions of the spatial convolution layer mainly include:

[0117] Convolution operation: The distribution of AIS data in a two-dimensional plane is scanned by the convolution kernel to capture the spatial position dependency and mutual relationship between ships; Activation function: The output of the convolution layer is passed through the ReLU activation function to introduce nonlinear changes and enhance the expression ability of the model; Pooling layer: Through the maximum pooling operation, the spatial size of the feature map is reduced, important spatial information is retained, and the amount of calculation is reduced;

[0118] The spatial position features are: that is, in the spatial dimension, the local features captured by ST-CNN are mainly related to the position, movement trajectory and surrounding environment of the ship. It should be noted that deep learning extraction is an abstract feature. Specifically, it includes the following features:

[0119] The geographical location of the ship: The latitude and longitude coordinates in the AIS data can be used to determine the specific location of the ship at each moment;

[0120] Relative position of ships: The model can learn the distribution characteristics of ships in space, such as whether they are close to other ships, ports or berths, etc. This information can reflect the spatial relationship between ships;

[0121] Ship motion status: The ship’s direction, speed change, acceleration and other motion information are captured through continuous position points in space. These are features extracted from local AIS data fragments through convolution operations.

[0122] And transmitting the spatial position feature to the fully connected layer;

[0123] The temporal convolution layer is used to extract the time series characteristics of the AIS data according to the AIS data spatiotemporal graph, and obtain the movement trajectory and docking behavior characteristics of the ship at different times;

[0124] Specifically, the main functions of the temporal convolutional layer include:

[0125] Temporal convolution operation: The one-dimensional convolution kernel slides on the time dimension to capture the time pattern of ship docking and departure. Different time steps can be flexibly set to adapt to AIS data of different lengths; Temporal pooling: Through the pooling operation on the time dimension, the length of the time series is reduced and key time points (such as the length of ship berthing, port entry and exit time, etc.) are retained; Long-term behavior capture: By stacking multiple temporal convolution layers, the model can capture the dynamic behavior of ships over a long period of time, such as long-term docking or frequent port entry and exit;

[0126] The time series characteristics of the AIS data are as follows: In the time dimension, ST-CNN captures the behavior patterns and changing trends of ships at different times. These time series change characteristics include:

[0127] Ship’s berthing time: AIS data can be used to detect the length of time a ship stays at a port or berth, and identify whether the ship is in a long-term berthing state;

[0128] Cyclic characteristics of ship activities: Some ships may have cyclical behavior patterns, such as scheduled port calls. The model can capture this cyclical feature and help identify the berthing behavior of ships more accurately.

[0129] Vessel trajectory change characteristics: By analyzing the AIS position points at different times, the vessel trajectory changes can be captured. This change can reflect the vessel's movement mode, such as arriving at a port, leaving a port, or cruising in a certain area.

[0130] And transmit the moving trajectory and docking behavior characteristics at different times to the fully connected layer;

[0131] The fully connected layer is used to perform feature fusion on the output of the spatial convolution layer and the output of the temporal convolution layer to obtain a high-dimensional feature vector for representing the spatial position and temporal behavior of the ship, and obtain the port berth position and berth type in the potential berth area according to the high-dimensional feature vector;

[0132] The formula for obtaining the high-dimensional feature vector used to represent the spatial position and temporal behavior of the ship is:

[0133] h t =f(W t *X t +W s *X s +b)

[0134] Where: h t represents a high-dimensional feature vector; W t ,W s Respectively represent the convolution kernels of the temporal convolution layer and the spatial convolution layer; X t ,X s They represent temporal features and spatial features respectively; f represents the activation function; b represents the bias term;

[0135] The output layer is used to output the results of port berth location and berth type. Figure 3 shown.

[0136] S4: Based on the constructed port berth information extraction model, the port type and port berth information of the remote sensing image of the port area after image preprocessing are extracted, and a berth information table is established according to the extracted port berth information, as shown in Table 1;

[0137] Table 1. Berth information table

[0138] Berth ID longitude latitude Port Type 001 -90.456000 30.123000 Container Port 002 -90.457000 30.124000 Bulk cargo port 003 -90.458000 30.125000 Oil and gas port ... ... ... ...

[0139] In this embodiment, the port area in the optimized remote sensing image set is gridded to obtain a regional grid map, and the AIS data spatiotemporal map is obtained according to the ship AIS data, so as to improve the ability of AIS data to describe the type and layout information of the port facilities themselves, and construct a port berth information extraction model. By training the pre-established Vision Transformer model according to the port type sample set, the port type prediction model obtained can automatically predict the port type, reducing the need for manual intervention. Compared with the traditional manual labeling and classification methods, it can greatly improve work efficiency and reduce labor costs; by training the ST-CNN model for identifying port berth information constructed according to the AIS data spatiotemporal map, the port berth information prediction model obtained can better capture the changes of ships in time and space, improve the accuracy of berth information recognition, so as to realize refined berth type recognition, and then based on the constructed port berth information extraction model, the port berth information is extracted from the remote sensing image of the port area after image preprocessing.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for extracting port berth information based on remote sensing images and AIS, characterized in that: The specific steps include: S1: Obtain historical remote sensing images of the port area and AIS data of ships matching the port area in the historical remote sensing images; And the historical remote sensing image is composed of a series of remote sensing frame images; Performing image preprocessing on the remote sensing frame image; The image preprocessing includes marking the port type of the port area in the remote sensing frame image to obtain a port type sample set; Or annotate potential berth areas in the port area in the remote sensing frame image to obtain an optimized remote sensing image set; S2: Gridding the port area in the optimized remote sensing image set to obtain a regional grid map, and based on the regional grid map, obtaining an AIS data spatiotemporal map according to the ship AIS data; The AIS time-space diagram includes at least the ship's position, the ship's berthing time, the ship's port entry and exit time, the ship's speed and the ship's heading; S3: Construct a port berth information extraction model; And the port berth information extraction model includes A pre-established Vision Transformer model is trained according to the port type sample set to obtain a port type prediction model for identifying the port type; and the ST-CNN model for identifying port berth information constructed by training the spatiotemporal graph of AIS data to obtain the port berth information prediction model; And the port berth information at least includes the berth location and berth type; S4: Based on the constructed port berth information extraction model, the port type and port berth information of the remote sensing image of the port area after image preprocessing is extracted.

2. The method for extracting port berth information based on remote sensing images and AIS according to claim 1 is characterized in that: The S2 specifically includes the following steps: S21: gridding the port area in the optimized remote sensing image set to obtain a regional grid map with M rows and N columns; And any unit grid in the regional grid map represents a fixed geographical area; S22: According to the ship AIS data, the longitude and latitude of the port area position are converted into the grid position index (i, j) of the regional grid map; And the expression of the grid position index (i,j) is Where: x min ,y min Respectively represent the minimum longitude and minimum latitude of the port area; x range ,y range Respectively represent the longitude range and latitude range of the port area; i represents the row number of the grid position index; j represents the column number of the grid position index; S23: Setting the time step, determining the ship AIS data corresponding to a number of time periods, and discretizing the data to obtain the time step index t; And the expression of time step index t is Where: t start Indicates the start time; t step Indicates the length of each time step; timestamp indicates the timestamp; S24: According to the grid position index (i, j) and the time step index t, the three-dimensional matrix X used to obtain the AIS data spatiotemporal diagram is obtained, and its expression is: Where: X(i,j,t) represents the decision variable of whether the ship only appears at the position index (i,j) in the time step index t.

3. The method for extracting port berth information based on remote sensing images and AIS according to claim 2 is characterized in that: In S3, the pre-established Vision Transformer model is trained according to the port type sample set to obtain the port type prediction model. The training process is as follows: S31: randomly dividing the port type sample set into a training set and a test set according to a preset ratio; S32: inputting the training set into a pre-created Vision Transformer model for model training; And based on the constructed model loss function, confirm whether the output of the trained Vision Transformer model converges; If the output of the trained Vision Transformer model converges, execute step S33; S33: Evaluate the model accuracy of the trained Vision Transformer model based on the test set; The indicators for evaluating the accuracy of the model are the recall rate and accuracy of the model; If it is confirmed that the model accuracy evaluation meets the preset model accuracy threshold, the trained Vision Transformer model is confirmed to be a port type prediction model for identifying remote sensing images of port areas; If it is confirmed that the model accuracy evaluation does not meet the preset model accuracy threshold, the weight parameters of the trained Vision Transformer model are adaptively adjusted based on the back propagation method, and step S32 is repeated.

4. The method for extracting port berth information based on remote sensing images and AIS according to claim 3 is characterized in that: The expression of the model loss function constructed in S32 is: Where: C represents the port category; w i represents the weight of the i-th type of port; y i Indicates the real label of the port type, Indicates the model’s predicted port type label.

5. The method for extracting port berth information based on remote sensing images and AIS according to claim 2 is characterized in that: The ST-CNN model constructed in S3 for identifying port berth information includes input layer, spatial convolution layer, temporal convolution layer, fully connected layer and output layer; The input layer is used to input the AIS data spatiotemporal graph into the spatial convolution layer and the temporal convolution layer respectively; The spatial convolution layer is used to extract the distribution characteristics of AIS data in two-dimensional space according to the AIS data spatiotemporal graph, and obtain the spatial position characteristics of the spatial position relationship of different ships in the port area; And transmitting the spatial position feature to the fully connected layer; The temporal convolution layer is used to extract the time series characteristics of the AIS data according to the AIS data spatiotemporal graph, and obtain the movement trajectory and docking behavior characteristics of the ship at different times; And transmit the moving trajectory and docking behavior characteristics at different times to the fully connected layer; The fully connected layer is used to perform feature fusion on the output of the spatial convolution layer and the output of the temporal convolution layer to obtain a high-dimensional feature vector for representing the spatial position and temporal behavior of the ship, and obtain the port berth position and berth type in the potential berth area according to the high-dimensional feature vector; The output layer is used to output the results of port berth locations and berth types.

6. The method for extracting port berth information based on remote sensing images and AIS according to claim 5 is characterized in that: The formula for obtaining the high-dimensional feature vector used to represent the spatial position and temporal behavior of the ship is h t =f(W t *X t +W s *X s +b) Where: h t represents a high-dimensional feature vector; W t ,W s Respectively represent the convolution kernels of the temporal convolution layer and the spatial convolution layer; X t ,X s Represent the temporal and spatial characteristics respectively; f represents the activation function; b represents the bias term.

Images