Ship identification method, server and storage medium based on multi-source remote sensing images

Abstract

The present application discloses a ship identification method based on multi-source remote sensing images, a server and a storage medium, which belongs to the field of target identification, and includes the following steps: constructing a data set, the data set includes a ship target data set, a sea background data set and a port background data set; training a number of network models for each data set, fusing the network models in a linear weighted manner to obtain a cascade classifier; inputting a certain remote sensing image data to be classified into the cascade classifier, and identifying the target ship in a parallel processing manner under the Map / Reduce framework; combining the structured elements of the target ship, extracting key target information from other multi-source remote sensing image data to jointly obtain the target structured elements, and jointly constructing a Bayesian network with the ship prior rule knowledge, and obtaining the attribute information of the target ship based on the fusion of the Bayesian network. This method has a wider recognition range and can adapt to various image conditions and target attributes.

Description

Technical Field

[0001] The present application belongs to the field of target recognition, and in particular relates to a ship recognition method, server and storage medium based on multi-source remote sensing images. Background Art

[0002] In remote sensing ship identification, target detection and recognition based on image processing technology is a common method. This method usually requires preprocessing of the input image, including planning the type, resolution, shooting angle, coverage, etc. of the input image, and clarifying what the target to be detected is and whether there are clear and stable imaging features in the image.

[0003] Existing technologies mainly include image feature extraction algorithms and feature matching algorithms. Image feature extraction algorithms are used to extract representative features from images, such as edges, corners, colors, etc. These features can be used to describe the shape, texture, color and other attributes of the target. Feature matching algorithms are used to match the extracted features with the features of known targets to determine whether the target exists.

[0004] However, there are some problems with existing technologies. First, due to the complexity and diversity of remote sensing images, images of different types, resolutions, shooting angles, and coverage may have a significant impact on target detection and recognition results. When the conditions of the input image or the target attributes change, it may lead to the inability to accurately extract and recognize the target.

[0005] Secondly, existing image feature extraction algorithms and feature matching algorithms may not be able to effectively process complex image data in some cases. For example, when the target is blurred, blocked or deformed in the image, the traditional feature extraction algorithm may not be able to accurately extract the target's features, resulting in inaccurate recognition results.

[0006] Therefore, there is an urgent need for a ship identification method that can solve the above problems. Summary of the invention

[0007] In order to address the deficiencies of the prior art, the present application provides a ship identification method based on multi-source remote sensing images. On the basis of the random association analysis method, a target detection and recognition method based on the Adaboost algorithm is introduced to fuse and associate multi-source and multi-scale remote sensing image information, and the unique ship attribute knowledge of each is integrated to obtain specific and clear ship target information.

[0008] The technical effects to be achieved by this application are achieved through the following solutions:

[0009] According to a first aspect of the present application, a method for ship identification based on multi-source remote sensing images is provided, comprising the following steps:

[0010] Step 1: construct a data set, which includes a ship target data set, a sea background data set, and a port background data set;

[0011] Step 2: train several network models for each data set, and fuse the network models in a linear weighted manner to obtain a cascade classifier;

[0012] Step 3: input a certain remote sensing image data to be classified into the cascade classifier, and identify the target ship in a parallel processing manner under the Map / Reduce framework;

[0013] Step 4: Combine the structured elements of the target ship, extract key target information from other multi-source remote sensing image data to obtain the target structured elements, and construct a Bayesian network with the ship prior rule knowledge, and obtain the attribute information of the target ship based on the Bayesian network fusion.

[0014] Preferably, in step 1, the remote sensing images containing ships, port backgrounds and sea backgrounds from various data sources are cropped to a specified size and then labeled, and the labeled remote sensing images are augmented to form a ship target data set, a sea background data set and a port background data set; 80% of each of the data sets is used for training the network model, and 20% is used for testing the network model.

[0015] Preferably, in step 2, different network models are trained for the ship target data set, the sea background data set, and the port background data set, respectively, to generate recognition models with different data distributions and different classification scenarios, and the network model adopts the Fast R-CNN network model.

[0016] Preferably, in step 2, several network models are trained for each data set, and the network models are fused in a linear weighted manner to obtain a cascade classifier. The specific method is:

[0017] Step 21: Initialize the weight distribution and network parameters of the training data in the dataset;

[0018] Step 22: Train the network model, calculate the network loss according to the network test results, obtain the weight coefficient of the current round of network model, and update the weight of the training data and the network parameters;

[0019] Step 23: Adopt a linear weighted voting combination mechanism, assign corresponding weight coefficients to different recognition models according to the weighting rule, and reduce the weighted recognition models to form a cascade classifier.

[0020] Preferably, in step 2, a Map / Reduce framework is used to train the plurality of network models in parallel, and the training method is:

[0021] Step 221: input the data set into a Map / Reduce framework, split the data set into a number of Map task samples, input the Map task samples into a network model for training, and obtain weight distribution through a gradient descent method;

[0022] Step 222: Input the weight distribution into the Reduce function for statistics, and update the weight coefficient of the entire network. Repeat step 221 until the optimal solution of the target loss function is reached to obtain the final network weight distribution.

[0023] Preferably, in step 3, the target ship is identified in a parallel processing manner under the Map / Reduce framework, specifically:

[0024] Step 31: Segment the remote sensing image data into a number of samples to be identified;

[0025] Step 32: Send the samples to be identified to multiple Map functions for processing, and the cascade classifier outputs the results;

[0026] Step 33: The output results are input into the Reduce function, and the non-maximum suppression algorithm is adopted to merge and obtain the final target ship.

[0027] Preferably, in step 31, the specific method of segmenting the remote sensing image data into a plurality of samples to be identified is:

[0028] Set the image capture frame according to the needs of the sample to be identified;

[0029] The image capture frame is slid along a specified route on the remote sensing image data to capture a sample to be identified; there is an overlapping area between adjacent samples to be identified.

[0030] Preferably, in step 4, the other multi-source remote sensing image data are time-calibrated and space-calibrated according to the remote sensing image data in step 3 to obtain geographic location information, time information and ship element attributes associated with the target ship;

[0031] The ship prior rule knowledge includes ship type, ship shape, hull structure, main dimensions, weight and performance parameters.

[0032] According to a second aspect of the present application, a server is provided, comprising: a memory and at least one processor;

[0033] The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the above-mentioned ship recognition method based on multi-source remote sensing images.

[0034] According to a third aspect of the present application, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed, the above-mentioned ship identification method based on multi-source remote sensing images is implemented.

[0035] According to an embodiment of the present application, the beneficial effect of adopting the ship identification method based on multi-source remote sensing images is that, by using multi-source remote sensing image data, combined with the feature information such as position, size, color, shape, texture, etc. reflected by the remote sensing image, a ship measurement information data set and a ship target sample training set are established, multi-source and multi-scale remote sensing image information is fused and associated, and each unique ship attribute knowledge is integrated to obtain specific and clear ship target information. Compared with the traditional target detection and recognition based on single image processing, this method combines the characteristics of ship targets in each remote sensing image, optimizes the recognition process, and can also describe the characteristics of the target results at different granularities and levels, and finally collects the recognition results of multiple ship targets into a ship target library, which has a wider recognition range and can adapt to various image conditions and target attributes. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present application or the existing technical solutions, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0037] Figure 1 This is a flow chart of a ship identification method based on multi-source remote sensing images in one embodiment of the present application;

[0038] Figure 2 This is a schematic diagram of the structure of a server in one embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0040] like Figure 1 As shown, a ship identification method based on multi-source remote sensing images in an embodiment of the present application includes the following steps:

[0041] Step 1: construct a data set, which includes a ship target data set, a sea background data set, and a port background data set;

[0042] In this step, when constructing the sample set, the data are mainly collected from Google Earth, JB satellite, China Center GF-2 satellite, and China Center GF-5 satellite. The labelme source annotation tool is used to select remote sensing images containing ships, port backgrounds, and sea backgrounds based on the above data sources, and each original remote sensing image is cropped to a specified size. On this basis, the ship is annotated, the location information of the target is annotated, and the annotation results are saved in the corresponding file as a universal data format file XML to facilitate reading, writing and parsing.

[0043] Among them, the position information refers to the pixel coordinates of the upper left corner and lower right corner of the rectangular box covered by the ship target in the remote sensing image. Samples containing ship targets are called positive samples. In order to ensure uniform sample distribution and improve the generalization ability of the recognition method, a series of negative samples (noise samples) are added to the sample set. At this time, remote sensing images of the same scene but not containing ship targets are selected, that is, empty target samples / background samples.

[0044] In order to avoid the limited number of samples, improve the robustness and universality of the model, and prevent overfitting of the detection and recognition model, the present invention needs to augment the labeled data set. Method 1 is to rotate and scale the original remote sensing image, affine transform, and adjust the image color, contrast, and saturation, which can effectively expand the data volume of the original remote sensing image set and solve the problem of insufficient data volume.

[0045] After the preliminary data augmentation processing, after the initial sample library is established based on the sample set constructed above, further augmentation is performed. Common methods include flipping, offsetting, scaling, etc. At this time, the target annotation position is augmented and mapped using affine transformation to form the final expanded sample set. 80% of the samples in the sample set are used for model training, and 20% of the samples are used for model testing. In order to improve the generalization ability of the model, the training and test sample sets can be divided into multiple random proportions for multiple training and prediction.

[0046] Step 2: train several network models for each data set, and fuse the network models in a linear weighted manner to obtain a cascade classifier;

[0047] In this step, different network models are trained for the ship target data set, the sea background data set, and the port background data set, respectively, to generate recognition models with different data distributions, certain performances, and different classifier scenarios. This differentiation capability can effectively improve the generalization ability of the classification device, and help improve recognition accuracy and stability.

[0048] The network model uses the Fast R-CNN network model and a supervised learning method to train and learn the appearance features of different ships in remote sensing images. The specific training method is as follows:

[0049] Step 21: Initialize the weight distribution and network parameters of the training data in the data set; if the number of samples is N, the initial sample weight is: 1 / N. The network parameters are generally initialized to random numbers with a value range of [0, 1] and a Gaussian distribution;

[0050] Step 22: Train the network model, calculate the network loss according to the network test results, obtain the weight coefficient of the current round of network model, and update the weight of the training data and the network parameters;

[0051] Step 23: Adopt the combination mechanism of linear weighted voting, assign corresponding weight coefficients to different recognition models according to the weighting rules, and reduce the weighted recognition models to form a cascade classifier. Although the scenes of each data set are different, they all contain the same recognition target ship, so there is a strong connection between the basic classification components. The combination mechanism of linear weighted voting is adopted, and different classifiers are assigned certain weight coefficients according to the weighting rules. Different types of classifiers are reduced to strong processors, and finally an effective cascade classifier is formed.

[0052] Since convolutional neural networks have high computational complexity, and processing large amounts of data requires long training time and large memory usage, in order to improve processing speed and training efficiency, this step uses Map / Reduce parallel processing to train the model.

[0053] The Map / Reduce framework includes two main functions: Map function and Reduce function. All data structures in the model can be represented as key-value pairs.<Key,Value> , where the Map function runs on each processing node and processes the data objects according to<K1,V1> is initialized and used as input parameters, and the intermediate data after processing is represented as<K2,V2> The programming model processes the intermediate data and merges all V2 values ​​with the same K2 value to generate<K2,V2list> Passed as input to the Reduce function, and after processing is completed,<K3,V3> The final result. The specific method is:

[0054] First, the sample is sent to the parallel computing framework to calculate the iterative results of the current sample network weight;

[0055] Then the obtained results are updated in the entire network, and the next iteration and update are continued until the optimal solution of the network weights is found. Specifically:

[0056] First, split the training sample set into several Map tasks in the Num. / N format, where Num is the number of samples and N is the number of Maps. Input the sample file.

[0057] Secondly, the input file is sent to the neural network for network training. The weight distribution of the current task is iterated through the gradient descent method. The result obtained by the gradient descent algorithm is recorded as<key,value> , where key represents the coordinate position corresponding to the weight, and value is the size of the weight. The key-value pair input to the Map function for the first time is a randomly obtained initialization key-value pair, and then the key-value pair calculated by N tasks is saved.<key,value> Key-value pairs;

[0058] Next, the key-value pairs obtained after one iteration are input to the Reduce function for statistical update. The Reduce function updates the input weights so that the weights of the entire network are updated;

[0059] Finally, continue to repeat the above weight calculation and update process. After multiple iterations, find the optimal solution to the target loss function and obtain the final network weight distribution, that is, the final network result.

[0060] Step 3: input a certain remote sensing image data to be classified into the cascade classifier, and identify the target ship in a parallel processing manner under the Map / Reduce framework;

[0061] In this step, a remote sensing image data collected from Google Earth, JB satellite, China Center GF-2 satellite, China Center GF-5 satellite, etc. is selected as the data source, and the cascade classifier is used for analysis and calculation. The parallel computing model based on Map / Reduce is used to accelerate the processing process to improve the speed of ship feature detection and recognition. Specifically:

[0062] In the recognition sample collection stage, the size of the remote sensing image in the actual prediction scene is relatively large. In order to meet the specified input size of the model, the idea of ​​sliding image blocks is introduced, and sliding windows in different areas are detected in turn to achieve real-time target detection speed and accuracy. Specifically:

[0063] Set the image capture frame according to the needs of the sample to be identified, and set the size, displacement distance, etc. of the image capture frame according to the needs;

[0064] The image capture frame is slid along a prescribed route on the remote sensing image data, for example, from left to right or from top to bottom, to capture samples to be identified according to a fixed size, and there are overlapping areas between adjacent samples to be identified.

[0065] In the recognition stage, the samples to be recognized are sent to multiple Map functions for processing. Different from the training process, the Map function in the recognition process mainly realizes the forward propagation calculation of the trained model results. Next, the final calculation result is input to the Reduce function for merging the recognition results. Also different from the training process, the Reduce function in the recognition process realizes the merging of the results of all recognition samples. The specific implementation adopts the non-maximum suppression algorithm, as follows:

[0066] Sort the output result frames of all ships from high to low according to the scores; and find the frame bmax with the highest score at present;

[0067] For the border bmax with the highest score at present, delete the border from the input list and add it to the output list;

[0068] Calculate the intersection over union (IoU) of the remaining borders bi and bmax one by one. If IoU(bi,bmax)>Nth, delete the border from the input list (where Nth is manually set, such as 0.8).

[0069] Repeat the above steps until the input list is empty, and the border screening is completed;

[0070] Step 4: Combine the structured elements of the target ship, extract key target information from other multi-source remote sensing image data to obtain the target structured elements, and construct a Bayesian network with the ship prior rule knowledge, and obtain the attribute information of the target ship based on the Bayesian network fusion.

[0071] This method belongs to the two-classification recognition scenario, that is, identifying the target ship and other backgrounds. After the location information of the target ship is obtained in step 3, more attribute information is obtained based on multi-source remote sensing image data. Before the fusion recognition process, time calibration and space calibration are performed, and then the target is associated.

[0072] Time and space calibration mainly unifies the time and space benchmarks of various multi-source remote sensing data. During the collection process of different remote sensing data, sensors adopt different benchmarks, so that the subsequent fusion work is accurate and error-free.

[0073] Time calibration mainly uses the time reference conversion to unify the time-related attributes of remote sensing data from various sources, such as the collection time and update time, into the same time standard. The time reference is the time scale. With the mastery of the laws of the earth's rotation and seasonal changes, the time accuracy is constantly improving, such as solar time, world time, atomic time, coordinated universal time, and the national time and frequency standard - Julian day.

[0074] Time base conversion realizes the mutual conversion of various time frequency bases and various clock sources. In addition, the series data collection time adopts the timestamp standard, such as GPS, Beidou satellite, etc. The time base conversion also supports the mutual conversion and validity verification between timestamp and time.

[0075] Spatial calibration unifies the spatial reference systems of remote sensing data from various sources, including geographic coordinate conversion and projection coordinate conversion. Geographic coordinate conversion is the process of converting coordinates (expressed in the form of geodetic coordinate system or spatial rectangular coordinate system) in one geographic coordinate system to another geographic coordinate system through a certain conversion formula. For example, the conversion between commonly used coordinate systems such as Beijing 54 coordinate system, Xi'an 80 coordinate system, WGS84 coordinate system, 2000 China geodetic coordinate system, etc.

[0076] Projection coordinate transformation projects the elements (coordinates, orientation, distance) on the ellipsoid onto the plane according to certain mathematical rules. Its essence is to establish a one-to-one functional relationship between the points on the earth's ellipsoid and the plane. The process of converting geographic coordinates to plane rectangular coordinates is called forward calculation, and the conversion of plane rectangular coordinates to geographic coordinates is called inverse calculation.

[0077] The projections supported by this method include: conformal transverse elliptical cylindrical projection, orthographic conformal cylindrical projection, transverse conformal cylindrical projection, conformal tangent conic projection, conformal tangent conic projection, conformal azimuthal projection, equal-area azimuthal projection, equidistant azimuthal projection, Web Mercator projection, etc.

[0078] In the fusion recognition process, the target ship is identified according to the characteristics of different scales, different postures and complex backgrounds of multi-source remote sensing images. After sorting and identifying a single individual target, the complementary characteristics of multi-source images are used to associate and fuse the ship targets from various data sources in terms of time (moment), space (position, orientation), attributes (size), etc., to build a Bayesian network model of the target feature set, and to infer and analyze valuable ship target background information from multiple angles or multiple levels, so as to comprehensively identify the target ship. Specifically:

[0079] There is a positional association between the target elements of multi-source remote sensing data, and the attributes of each source data element are multifaceted. Taking different remote sensing acquisition methods as an example, the principles of active remote sensing and passive remote sensing are different, and the acquisition results contain different attribute characteristics. Active remote sensing has strong penetration and has the advantages of all-weather and all-day, which is significant for ship target identification and time monitoring. Passive remote sensing is a sensor that receives and records electromagnetic waves emitted or reflected by the target itself from natural radiation sources such as the sun. Different objects and targets have different reflection capabilities for geomagnetic waves. The difference in reflection characteristics can be applied to the knowledge refinement and knowledge reasoning of ship targets.

[0080] The remote sensing image data source has different acquisition characteristics and data characteristics. Passive remote sensing is suitable for target recognition due to its reflective characteristics. This type of remote sensing image suitable for classification is selected. The ship target recognition result is obtained through the aforementioned parallel recognition. The recognition result is a structured element, including the recognition target and the corresponding spatiotemporal attributes, band spectral attributes, etc.

[0081] The multi-source remote sensing image data from other sources are combined with the obtained ship target structured element information to perform rough temporal and spatial correlation, and extract key target information from the data to form structured elements. The elements contain attributes such as spectral reflectance characteristic curves. The same object may show different spectral reflectance characteristics, while different objects may show the same spectral reflectance characteristics, that is, the "same object, different spectrum, same spectrum, different objects" phenomenon.

[0082] In the Bayesian network fusion recognition process, in addition to the above-mentioned ship recognition results, geographic location information, time information, and ship input attributes, the known ship prior knowledge is also combined to achieve knowledge fusion and further knowledge reasoning. Ship knowledge rules include: different ship types, ship shapes, hull structures, main dimensions, weights, performance parameters and other information. With the continuous enrichment of information, the prior knowledge can be further expanded.

[0083] Based on structured prior knowledge and structured element information, a Bayesian network is constructed to achieve the association between various information based on the network, and finally obtain complete and rich ship target and attribute information.

[0084] The Bayesian network consists of nodes and directed arcs, where the nodes are variables, that is, the identified multi-source target information, and the directed arcs represent the causal relationship between the variables. The construction process of the Bayesian network is:

[0085] First, the correlation between variables is calculated and some variables with weak correlation are eliminated to improve the efficiency and accuracy of structural learning.

[0086] Secondly, the K2 algorithm is used for structural learning to determine the causal relationship between nodes.

[0087] Then, the posterior probability distribution of each node is determined by parameter learning, that is, the degree of association between nodes is quantified using sample data. In the present invention, parameter learning is mainly performed by combining sample data and prior rule knowledge.

[0088] Finally, after the Bayesian network is constructed, the model validity is manually tested to verify the model accuracy, and the model results are used for subsequent correlation derivation.

[0089] like Figure 2 As shown, a server in an embodiment of the present application includes: a memory 201 and at least one processor 202;

[0090] The memory 201 stores computer programs, and the at least one processor 202 executes the computer programs stored in the memory 201 to implement the above-mentioned ship recognition method based on multi-source remote sensing images.

[0091] In one embodiment of the present application, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed, the above-mentioned ship identification method based on multi-source remote sensing images is implemented.

[0092] According to an embodiment of the present application, the beneficial effect of adopting the ship identification method based on multi-source remote sensing images is that, by using multi-source remote sensing image data, combined with the position, size, color, shape, texture and other feature information reflected by the remote sensing images, a ship measurement information data set and a ship target sample training set are established, multi-source and multi-scale remote sensing image information is fused and associated, and the unique ship attribute knowledge of each is integrated to obtain specific and clear ship target information. Compared with traditional target detection and recognition based on image processing, it has a wider recognition range and can adapt to various image conditions and target attributes.

[0093] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0094] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0095] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0096] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0097] For ease of description, spatially relative terms, such as "above", "above", "on the upper surface of", "above", etc., may be used herein to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" may include both "above" and "below". The device may also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatially relative descriptions used herein are interpreted accordingly.

[0098] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0099] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A ship recognition method based on multi-source remote sensing images, characterized in that: The steps include: Step 1: construct a data set, which includes a ship target data set, a sea background data set, and a port background data set; Step 2: train several network models for each data set, and fuse the network models in a linear weighted manner to obtain a cascade classifier; Step 3: input a certain remote sensing image data to be classified into the cascade classifier, and identify the target ship in a parallel processing manner under the Map / Reduce framework; Step 4: Combine the structured elements of the target ship, extract key target information from other multi-source remote sensing image data to obtain the target structured elements, and construct a Bayesian network with the ship prior rule knowledge, and obtain the attribute information of the target ship based on the Bayesian network fusion.

2. The ship identification method based on multi-source remote sensing images according to claim 1 is characterized in that: In step 1, the remote sensing images containing ships, port backgrounds and sea backgrounds from various data sources are cropped to specified sizes and annotated, and the annotated remote sensing images are augmented to form ship target data sets, sea background data sets and port background data sets; 80% of each of the data sets is used for training the network model, and 20% is used for testing the network model.

3. The ship identification method based on multi-source remote sensing images according to claim 1 is characterized in that: In step 2, different network models are trained for the ship target data set, the sea background data set, and the port background data set to generate recognition models with different data distributions and classification scenarios. The network model adopts the Fast R-CNN network model.

4. The ship identification method based on multi-source remote sensing images according to claim 3 is characterized in that: In step 2, several network models are trained for each data set, and the network models are fused in a linear weighted manner to obtain a cascade classifier. The specific method is: Step 21: Initialize the weight distribution and network parameters of the training data in the dataset; Step 22: Train the network model, calculate the network loss according to the network test results, obtain the weight coefficient of the current round of network model, and update the weight of the training data and the network parameters; Step 23: Adopt a linear weighted voting combination mechanism, assign corresponding weight coefficients to different recognition models according to the weighting rule, and reduce the weighted recognition models to form a cascade classifier.

5. The ship identification method based on multi-source remote sensing images according to claim 4 is characterized in that: In step 2, the Map / Reduce framework is used to train several network models in parallel, and the training method is: Step 221: input the data set into a Map / Reduce framework, split the data set into a number of Map task samples, input the Map task samples into a network model for training, and obtain weight distribution through a gradient descent method; Step 222: Input the weight distribution into the Reduce function for statistics, and update the weight coefficient of the entire network. Repeat step 221 until the optimal solution of the target loss function is reached to obtain the final network weight distribution.

6. The ship identification method based on multi-source remote sensing images according to claim 1 is characterized in that: In step 3, the target ship is identified in a parallel processing manner under the Map / Reduce framework, specifically: Step 31: Segment the remote sensing image data into a plurality of samples to be identified; Step 32: Send the samples to be identified to multiple Map functions for processing, and the cascade classifier outputs the results; Step 33: The output results are input into the Reduce function, and the non-maximum suppression algorithm is adopted to merge and obtain the final target ship.

7. The ship identification method based on multi-source remote sensing images according to claim 6 is characterized in that: In step 31, the specific method of segmenting the remote sensing image data into a number of samples to be identified is: Set the image capture frame according to the needs of the sample to be identified; The image capture frame is slid along a specified route on the remote sensing image data to capture a sample to be identified; there is an overlapping area between adjacent samples to be identified.

8. The ship identification method based on multi-source remote sensing images according to claim 1 is characterized in that: In step 4, the other multi-source remote sensing image data are time-calibrated and space-calibrated according to the remote sensing image data in step 3 to obtain geographic location information, time information and ship element attributes associated with the target ship; The ship prior rule knowledge includes ship type, ship shape, hull structure, main dimensions, weight and performance parameters.

9. A server, characterized in that: include: memory and at least one processor; The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the ship recognition method based on multi-source remote sensing images as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the ship recognition method based on multi-source remote sensing images according to any one of claims 1 to 8 is implemented.

Images