Waterborne vessel name joint identification method and system, electronic device and storage medium

Abstract

The application discloses a water ship name joint identification method and system, an electronic device and a storage medium, and the method comprises the following steps: sending a joint identification ship name task of an initiator with a ship name identification requirement and image data of a ship name to be identified to a server; randomly selecting participants participating in the identification task on the server, distributing the image data to the participants participating in the joint identification ship name, and calculating the geographical weight parameters of each participant; after each participant receives the distributed image data, each participant uses a pre-trained local model to identify the image data to obtain an identified ship name result; uploading the ship name results obtained by all the selected participants to the server; according to the geographical weight coefficient, performing weight distribution on all the ship name results on the server to obtain a weight coefficient of each ship name result; and selecting a ship name result corresponding to a maximum weight coefficient and sending the ship name result to the initiator to complete the joint identification ship name task.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of water vessel supervision, and particularly relates to a water vessel name joint identification method and system, an electronic device and a storage medium. BACKGROUND

[0002] Nowadays, machine learning has penetrated into various fields of people's life, and is gradually changing the business and social life style of human beings. The method of machine learning is to obtain a certain model by using existing data, and to predict the future by using the model. Although the ability and application scene of machine learning have been greatly developed, there are still many problems to be solved in the process of landing practice.

[0003] Combining machine learning with the maritime field has certain difficulties. Although different maritime departments have different ship name data sets, due to the restriction of information security, different maritime departments cannot share the ship name data set. This greatly reduces the accuracy and efficiency of ship name identification.

[0004] In recent years, the method of using federal learning to solve centralized training of data has emerged. Federal machine learning is a machine learning framework that can effectively help multiple parties to use data and machine learning training while meeting user privacy protection, data security and government regulations.

[0005] At present, federal learning provides a potential solution to the two problems of data silo caused by information security problems and the use of ship name data sets of different maritime departments to improve model accuracy, that is, each maritime department trains a model locally, then uploads the trained model parameters to the server for weighted aggregation, thereby training a more efficient model. Intuitively, this can effectively solve the above problems. However, studies have shown that federal learning can easily infer user data from gradients and model parameters.

[0006] Therefore, how to invent a method that does not leak training data and can efficiently identify ship names is a problem that needs to be solved in the technical field. SUMMARY

[0007] The main purpose of the present application is to provide a water vessel name joint identification method, system, electronic device and storage medium, which aims to provide a technical solution that does not leak training data and can efficiently identify ship names.

[0008] To achieve the above object, the application provides a water ship name joint identification method, which comprises the following steps: sending a joint identification ship name task of an initiator with a ship name identification requirement and image data of a ship name to be identified to a server; randomly selecting participants involved in the identification task on the server, downloading the image data to the participants involved in the joint identification ship name, and calculating a geographical weight parameter of each participant; after each participant receives the downloaded image data, using a local model pre-trained by each participant to identify the image data to obtain a ship name result; uploading the ship name result obtained by all selected participants to the server; according to the geographical weight coefficient, performing weight distribution on all ship name results on the server to obtain a weight coefficient of each ship name result; selecting a ship name result corresponding to a maximum weight coefficient and sending the ship name result to the initiator to complete the joint identification ship name task.

[0009] Further, the training method of the local model is as follows: combining a yolo model, a CRNN convolutional recurrent neural network and a CTC neural network-based time sequence classification network to obtain a local training model; initializing the local training model; each participant uses a local ship name data set to train the local training model to obtain a personalized local model.

[0010] Further, the formula of the pre-trained local model is as follows:

[0011]

[0012] wherein, represents the initialized model, represents a model after t rounds of training, f represents a local training function, d represents a local ship name data set of a participant, and p represents a pth participant.

[0013] The identification formula of the ship name result is as follows:

[0014]

[0015] wherein, N is the image data, represents a model after t rounds of training, s p is a ship name result.

[0016] Further, the identification of the image data using the pre-trained local model comprises the following steps: acquiring a target character region image of the image data; extracting features of the target character region image to obtain a feature sequence; and transcribing the feature sequence to obtain a ship name result of the image data.

[0017] Further, the selection of the participants involved in the identification task on the server comprises the following steps:

[0018] m = max(C × n, 1)

[0019] Where m is the number of participants in the identification, C represents the proportion of participants in the identification, and n is the total number of participants in the identification and non-participants.

[0020] Furthermore, the geographical weighting coefficient for each participant is calculated as follows:

[0021] On the server, the Euclidean distance between the participant and the sender is calculated based on their geographical coordinates to obtain the geographical weight coefficient for each participant.

[0022] The calculation method is as follows:

[0023]

[0024] Where, d i Let d′ represent the Euclidean distance between the i-th participant and the sender. i This represents the Euclidean distance and geographical weighting coefficient.

[0025] Furthermore, the method for weighting all ship name results on the server includes: weighting the ship name recognition results on the server using a weighted average based on the sample size of each participant to obtain a sample weight coefficient for each participant; classifying the recognition results on the server; and obtaining a weight coefficient on the server by combining the sample weight coefficient and the geographical weight coefficient based on the classified recognition results; the calculation process for the sample weight coefficient is as follows:

[0026]

[0027] Where, n′ i Let n represent the sample weight coefficient of the i-th participant. i This represents the number of participants in the sample, where m represents the number of participants involved in the identification.

[0028] The identification result s is classified on the server to obtain S = {S1, s2, ..., s}. k}, where k represents the number of categories, S i The index value recorded in the middle is the index value of the same recognition result s;

[0029] The calculation process for the weighting coefficients is as follows:

[0030]

[0031] Where, n′ j Let d′ be the sample weight coefficient. j This is the geographical weighting coefficient.

[0032] The second aspect of the present application provides a water ship name joint identification system, comprising: an image transmission module, configured to send a joint identification ship name task of an initiator with a ship name identification requirement and image data of a ship name to be identified to a server; an image issuing module, configured to randomly select participants participating in the identification task on the server, issue the image data to the participants participating in the joint identification ship name, and calculate a geographical weight parameter of each participant; an image identification module, configured to, after each participant receives the issued image data, use a local model pre-trained by each participant to identify the image data to obtain an identified ship name result; a result uploading module, configured to upload the ship name result obtained by all selected participants to the server; a weight distribution module, configured to, according to the geographical weight coefficient, distribute weights to all ship name results on the server to obtain a weight coefficient of each ship name result; and a result selection module, configured to select a ship name result corresponding to a maximum weight coefficient and send the ship name result to the initiator to complete the joint identification ship name task.

[0033] The third aspect of the present application provides an electronic device, comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein when the processor executes the computer program, the water ship name joint identification method described in any one of the above aspects is implemented.

[0034] The fourth aspect of the present application provides a computer readable storage medium, having a computer program stored thereon, wherein when the computer program is executed by a processor, the water ship name joint identification method described in any one of the above aspects is implemented.

[0035] The present application provides a water ship name joint identification method, system, electronic device and storage medium, which has the beneficial effect that when there is a ship name identification requirement, the requester only needs to transmit the identification task and image data of a ship name to be identified to a server, and can obtain the identification result returned by the server. In the entire identification process, each participant performs identification on a local model, and no other data exchange is required, and the local data of the participants and the initiator does not need to be uploaded. Therefore, the problem of maritime data island and training data privacy is solved, and the data and model are available but invisible. The joint inference method is used to complete the task of jointly identifying a ship name by multiple maritime departments with data, thereby solving the problem of maritime data island caused by information security problems. In addition, the local model of the participant is pre-trained, and does not need to be trained when the water ship name joint identification scheme is executed, thereby improving the efficiency of identifying the ship name. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only constitute some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0037] Figure 1 Flow chart of the water ship name joint recognition method of the embodiment of the present application;

[0038] Figure 2 Implementation schematic diagram of the water ship name joint recognition method of the embodiment of the present application;

[0039] Figure 3 Framework diagram of the water ship name joint recognition system of the embodiment of the present application;

[0040] Figure 4 Schematic structural block diagram of the electronic device of the embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to make the purposes, features and advantages of the present application more obvious and easy to understand, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0042] Please refer to Figure 1 A water ship name joint recognition method, comprising:

[0043] S1, sending a joint recognition ship name task of an initiator with a ship name recognition requirement and image data of a ship name to be recognized to a server;

[0044] S2, randomly selecting participants participating in the recognition task on the server, downloading the image data to the participants participating in the joint recognition of the ship name, and calculating a geographical weight parameter of each participant;

[0045] S3, after each participant receives the downloaded image data, using a local model pre-trained by each participant to recognize the image data to obtain a recognized ship name result;

[0046] S4, uploading the ship name results obtained by all selected participants to the server;

[0047] S5, according to the geographical weight coefficient, performing weight distribution on all ship name results on the server to obtain a weight coefficient of each ship name result;

[0048] S6, the ship name result corresponding to the maximum weight coefficient is selected and sent to the initiator, and the task of jointly identifying the ship name is completed.

[0049] In the embodiment, the initiator and the participants are all communication devices in the maritime department. In the step S1, the initiator does not have image data samples of this category in the ship name data set, so the initiator sends the image to be identified in the identification system to the aggregation server, and uses the local training set of the maritime department in other places to achieve the purpose of joint identification.

[0050] In the step S2, each participant has a corresponding geographical weight parameter, which reflects the influence degree of the distance between the maritime departments on the identification results of each party.

[0051] The water ship name joint identification method provided in the embodiment has the overall process as shown in Figure 2 When a maritime department has a ship name identification requirement, the requester only needs to transmit the identification task and the image data of the ship name to be identified to the aggregation server, and can obtain the identification result returned by the server. In the whole identification process, each participant performs identification on the local model thereof. Except for the image data of the ship name to be identified and the identification result, no other data exchange is needed, and the local data of the participants and the initiator does not need to be uploaded. Therefore, the problems of maritime data island and training data privacy are solved, the data and the model are available but invisible, the joint inference is used to complete the task of jointly identifying a ship name by multiple maritime departments having data, and the problem of maritime data island caused by information security is solved. In addition, the local model of the participant is pre-trained, and does not need to be trained when the water ship name joint identification scheme is executed, so that the efficiency of identifying the ship name is improved.

[0052] In addition, the water ship name joint identification method provided in the embodiment uses the multi-model joint inference instead of the conventional machine learning single-model inference, and allocates corresponding weights to each participant according to the geographical position of the maritime department, so that the accuracy of ship name identification is improved.

[0053] In one embodiment, the training method of the local model is as follows:

[0054] The yolo model, the CRNN convolutional recurrent neural network and the CTC neural network-based time sequence classification network are combined to obtain the local training model.

[0055] The local training model is initialized.

[0056] Each participant trains the local training model using the local ship name data set thereof to obtain a personalized local model.

[0057] In this embodiment, the YOLO (you only look once) model is used to obtain the target text region image of the ship name image to be identified;

[0058] By employing a CRNN (Convolutional Recurrent Neural Network), the image features of the target text region in the image of the ship name to be identified can be extracted.

[0059] The CTC (Connectionist Temporal Classification) neural network-based temporal classification computation network can transcribe the feature sequences output by the CRNN recurrent layer.

[0060] In this embodiment, YOLO, CRNN, and CTC are combined to initialize the local training model. By training the model using a local ship name dataset, you can obtain a personalized local model for each individual. in, This includes the model structure and model parameters.

[0061] In this embodiment, the training process of the participant's local model is completed during the participant's idle time, rather than when the sender initiates the joint identification model task. This can reduce the time spent performing the joint identification of ship names and improve the efficiency of joint identification of ship names.

[0062] In one embodiment, the formula for the pre-trained local model is:

[0063]

[0064] in, This represents the initialized model. Let f represent the model after training for t rounds, f represent the local training function, d represent the local ship name dataset of the participants, and p represent the p-th participant.

[0065] The formula for identifying ship names is:

[0066]

[0067] Where N is the image data, s represents the model after training t rounds. p The result is the ship's name.

[0068] In this embodiment, taking 256×256 image data and the ship name "FLNG" as an example, with 5 participants, after ship name recognition, 5 ship name results s are obtained. ps1="FLNG", s2="FLNG", s3="FING", s4="FLNG", s5="FING"

[0069] In this embodiment, the participants input the image data N to be identified into the pre-trained local model and output the recognized ship name result s p It should be noted that the output recognition result s p is a ship name string.

[0070] In one embodiment, using a pre-trained local model to identify image data includes: obtaining a target text area image of the image data; extracting features of the target text area image to obtain a feature sequence; transcribing the feature sequence to obtain a ship name result of identifying the image data.

[0071] In this embodiment, yolo, CRNN, and CTC mentioned in the above embodiments are used respectively to achieve the recognition of the ship name result.

[0072] In one embodiment, the participants participating in the current identification task are randomly selected on the server, and the selection calculation process is:

[0073] m=max(Cxn, 1)

[0074] Where m is the number of participants participating in the identification, C represents the proportion of participants participating in the identification, and n is the total number of participants participating in the identification and participants not participating in the identification.

[0075] In this embodiment, still taking 5 participants as an example, the selection calculation process is:

[0076] 5=max(0.05x100, 1)

[0077] In one embodiment, the calculation method of the geographical weight coefficient of each participant is:

[0078] The Euclidean distance between the participants and the sender is calculated according to the geographical coordinates between the participants and the sender on the server to obtain the geographical weight coefficient of each participant;

[0079] The calculation method is:

[0080]

[0081] Where di represents the Euclidean distance between the i-th participant and the sender, and d' i represents the Euclidean distance and the geographical weight coefficient. i i

[0082] ​​In this embodiment, based on the selection calculation process of the above embodiment, the geographic weight coefficients D between the 5 participants and the initiator are calculated as {0.1, 0.15, 0.3, 0.25, 0.2}.

[0083] In one embodiment, the method of weight distribution of all ship name results on the server includes:

[0084] On the server, the ship name recognition results are weighted and distributed according to the sample quantity of each participant by using the weighted average method to obtain the sample weight coefficient of each participant;

[0085] On the server, the recognition results are classified;

[0086] On the server, the sample weight coefficient and the geographic weight coefficient are obtained according to the classified recognition results;

[0087] The calculation process of the sample weight coefficient is:

[0088]

[0089] wherein n' i represents the sample weight coefficient of the i th participant, n i represents the sample quantity of the participant, and m represents the number of participants participating in the recognition;

[0090] On the server, the recognition results s are classified to obtain S = {S1, S2,..., Sk}, wherein k represents the number of categories, and S k records the index value of the same recognition result s; i

[0091] The calculation process of the weight coefficient is:

[0092]

[0093] wherein n' j is the sample weight coefficient, and d' j is the geographic weight coefficient.

[0094] In this embodiment, taking the sample weight coefficients N calculated in the above embodiment as an example, after the recognition results are classified on the server, S = {(1, 2, 4), (3, 5)} is obtained, and S i records the index value of the same recognition result s;

[0095] ​In the calculation of the weight coefficient W, still using the data of the above-mentioned embodiment as an example, the number of participants is 5, and the weight coefficient is {0.1, 0.15, 0.3, 0.25, 0.2}, and the calculated weight coefficient W is {0.12, 0.045}, 0.12>0.045, the server returns "FLNG" to the initiator, and the joint identification task is completed.

[0096] Please refer to Figure 3 The embodiment of the present application also provides a water ship name joint identification system, which comprises an image transmission module 1, an image issuing module 2, an image identification module 3, a result uploading module 4, a weight distribution module 5 and a result selection module 6; the image transmission module 1 is used for sending the joint identification ship name task of an initiator who has a ship name identification demand and image data of a ship name to be identified to a server; the image issuing module 2 is used for randomly selecting participants who participate in the identification task on the server, issuing the image data to the participants who participate in the joint identification ship name, and calculating the geographical weight parameters of each participant; the image identification module 3 is used for identifying the image data using a local model pre-trained by each participant after each participant receives the issued image data, to obtain a ship name result of identification; the result uploading module 4 is used for uploading the ship name results obtained by all selected participants to the server; the weight distribution module 5 is used for distributing the weight of all ship name results according to the geographical weight coefficient, to obtain the weight coefficient of each ship name result; and the result selection module 6 is used for selecting the ship name result corresponding to the maximum weight coefficient and sending it to the initiator, to complete the joint identification ship name task.

[0097] In one embodiment, the water ship name joint identification system further comprises a model combination module, a model initialization module and a model training module, the model combination module is used for combining a yolo model, a CRNN convolutional recurrent neural network and a CTC neural network-based time sequence classification network, to obtain a local training model; the model initialization module is used for initializing the local training model; and the model training module is used for training the local training model using a local ship name data set of each participant, to obtain a personalized local model of each participant.

[0098] In one embodiment, the formula of the pre-trained local model in the image identification module 3 is as follows:

[0099]

[0100] wherein, represents the initialized model, represents the model after t rounds of training, f represents a local training function, d represents a local ship name data set of a participant, and p represents the pth participant;

[0101] The identification formula of the ship name result is as follows:

[0102]

[0103] wherein N is the image data, denotes the model after t rounds of training, s p is the ship name result.

[0104] In one embodiment, the image recognition module 3 comprises a region image acquisition unit, a feature extraction unit and a transcription unit; the region image acquisition unit is configured to acquire a target character region image of the image data; the feature extraction unit is configured to extract features of the target character region image to obtain a feature sequence; and the transcription unit is configured to transcribe the feature sequence to obtain a ship name result for recognizing the image data.

[0105] In one embodiment, when the image distribution module 2 randomly selects participants participating in the current recognition task on the server, the calculation process is as follows:

[0106] m = max(C x n, 1)

[0107] wherein m is the number of participants participating in the recognition, C represents the proportion of participants participating in the recognition, and n is the total number of participants participating in the recognition and participants not participating in the recognition.

[0108] In one embodiment, the calculation method of the image distribution module 2 for calculating the geographical weight coefficient of each participant is as follows: the Euclidean distance between the participant and the sender is calculated according to the geographical coordinates of the two on the server to obtain the geographical weight coefficient of each participant.

[0109] The calculation formula is as follows:

[0110]

[0111] wherein d i represents the Euclidean distance between the i th participant and the sender, d' i represents the Euclidean distance and the geographical weight coefficient.

[0112] In one embodiment, the weight distribution module 5 comprises a sample weight coefficient calculation unit, a classification unit and a weight coefficient calculation unit.

[0113] The sample weight coefficient calculation unit is configured to distribute weights of the ship name recognition results according to the sample quantities of each participant in a weighted average manner on the server to obtain a sample weight coefficient of each participant.

[0114] The classification unit is configured to classify the recognition results on the server.

[0115] The weight coefficient calculation unit is configured to calculate sample weight coefficients and geographical weight coefficients according to the classified recognition results on the server.

[0116] The calculation process of the sample weight coefficients is as follows:

[0117]

[0118] wherein n' i represents the sample weight coefficient of the i th participant, n i represents the sample quantity of the participant, and m represents the quantity of the participants participating in the recognition.

[0119] The recognition results s are classified on the server to obtain S={S1, S2,..., Sk}, wherein k represents the quantity of the categories, and S k records the index value of the same recognition result s. i

[0120] The calculation process of the weight coefficients is as follows:

[0121]

[0122] wherein n' j is the sample weight coefficient, and d' j is the geographical weight coefficient.

[0123] The water ship name joint recognition system provided by the embodiment of the present application can obtain the recognition result returned by the server when a certain maritime department has a ship name recognition demand, and the demander only needs to transmit the recognition task and the image data of the to-be-recognized ship name to the aggregation server. In the whole recognition process, each participant performs recognition on the local model of each participant. Except for the image data of the to-be-recognized ship name and the recognition result, no other data exchange is needed, and the local data of the participants and the initiator does not need to be uploaded. Therefore, the maritime data island and the training data privacy problem are solved, the data and the model are available but invisible, the joint inference mode is adopted to complete the task of jointly recognizing a ship name by multiple maritime departments with data, thereby solving the maritime data island problem caused by the information security problem. In addition, the local model of the participant is pre-trained, and does not need to be trained when the water ship name joint recognition scheme is executed, thereby improving the efficiency of recognizing the ship name.

[0124] In addition, the water ship name joint recognition system provided by the embodiment of the present application replaces the conventional machine learning single model inference with the multi-model joint inference, and allocates the corresponding weight of each participant according to the geographical position of the maritime department, thereby improving the accuracy of the ship name recognition.

[0125] The embodiment of the present application provides an electronic device, please refer to Figure 4 ​The electronic device includes a memory 601, a processor 602, and a computer program stored on the memory 601 and executable on the processor 602, and the processor 602 implements the water ship name joint identification method described in the foregoing when executing the computer program.

[0126] Further, the electronic device further includes at least one input device 603 and at least one output device 604.

[0127] The memory 601, the processor 602, the input device 603, and the output device 604 are connected through a bus 605.

[0128] The input device 603 can be a camera, a touch panel, a physical key, or a mouse, etc. The output device 604 can be a display screen.

[0129] The memory 601 can be a high-speed random access memory (RAM) or a non-volatile memory such as a disk memory. The memory 601 is used to store a set of executable program codes, and the processor 602 is coupled with the memory 601.

[0130] Further, the embodiment of the present application further provides a computer readable storage medium, which can be arranged in the electronic device in the foregoing embodiments, and the computer readable storage medium can be the memory 601 in the foregoing. The computer readable storage medium stores a computer program, and the program is executed by the processor 602 to implement the water ship name joint identification method described in the foregoing embodiments.

[0131] Further, the computer readable storage medium can be a U disk, a mobile hard disk, a read-only memory (ROM), a RAM, a magnetic disk, or an optical disk, etc. various media that can store program codes.

[0132] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiment described above is only schematic, for example, the division of the modules is only a logical function division, and actual implementation can have another division manner, for example, a plurality of modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed each other can be indirect coupling or communication connection through some interfaces, apparatuses or modules, and can be electrical, mechanical or other forms.

[0133] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, i.e., may be located in one place, or may be distributed to multiple network modules. Some or all of the modules can be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0134] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module.

[0135] The integrated module, if realized in the form of a software functional module and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0136] It should be noted that, for the foregoing method embodiments, in order to facilitate description, they are all described as a combination of a series of actions, but those skilled in the art should know that the present application is not limited by the order of the described actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily essential to the present application.

[0137] In the above embodiments, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0138] The above is a description of the water ship name joint identification method, system, electronic device and storage medium provided by the present application. For those skilled in the art, according to the idea of the embodiments of the present application, there will be changes in specific implementation and application range. In summary, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A method for joint identification of ship names on waterways, characterized in that, include: The joint ship name identification task initiated by the ship name identification requirement and the image data of the ship name to be identified are sent to the server; the initiator and the participant are all communication devices of the maritime department, and the ship name dataset of the initiator does not contain image data samples of the ship name to be identified. Participants for this identification task are randomly selected on the server. The image data to be identified is then sent to the participants in this joint identification of ship names, and the geographical weight parameter for each participant is calculated. The selection and calculation process is as follows: m = max(C × n, 1) Where m is the number of participants in the identification process, C represents the proportion of participants in the identification process, and n is the total number of participants in and out of the identification process. The geographical weight parameter is calculated as follows: The Euclidean distance between the participant and the sender is calculated on the server based on their geographical coordinates to obtain the geographical weight coefficient for each participant. The calculation method is as follows: Where, d i Let d represent the Euclidean distance between the i-th participant and the sender. i ′ Indicates the geographical weighting coefficient; After each participant receives the image data, they use their pre-trained local model to identify the image data and obtain the identified ship names. The training method is as follows: the YOLO model, CRNN convolutional recurrent neural network and CTC neural network-based temporal classification network are combined to obtain the local training model; the local training model is initialized; each participant uses their own local ship name dataset to train the local training model and obtain their own personalized local model. Upload the ship name results obtained by all selected participants to the server; Based on the geographical weight coefficient, weights are assigned to all ship name results on the server to obtain the weight coefficient for each ship name result; The ship name result corresponding to the maximum weight coefficient is selected and sent to the initiator to complete the task of jointly identifying the ship name; Throughout the entire identification process, apart from the image data of the ship name to be identified and the identification results, there is no other data exchange, and there is no need to upload the local data of the participants or initiators.

2. The method for joint identification of ship names on water as described in claim 1, characterized in that, The formula for a pre-trained local model is: in, This represents the model after initialization. Let f represent the model after training for t rounds, f represent the local training function, d represent the local ship name dataset of the participants, and p represent the p-th participant. The formula for identifying the ship name result is: Where N is the image data, S represents the model after training for t rounds. p The result is the ship's name.

3. The method for joint identification of ship names on water as described in claim 1, characterized in that, Recognizing the image data using a pre-trained local model includes: Obtain the target text region image of the image data; The features of the target text region image are extracted to obtain a feature sequence; The feature sequence is transcribed to obtain the ship name results for identifying the image data.

4. A joint identification system for ship names on waterways, characterized in that, include: The image transmission module is used to send the joint ship name identification task of the initiator with the need for ship name identification and the image data of the ship name to be identified to the server; the initiator and the participants are all communication devices of the maritime department, and the ship name dataset of the initiator does not contain image data samples of the ship name to be identified. The image distribution module is used to randomly select participants for this identification task on the server, distribute the image data to be distributed to the participants participating in this joint identification of ship names, and calculate the geographical weight parameters for each participant; wherein the selection and calculation process is as follows: m = max(C × n, 1) Where m is the number of participants in the identification process, C represents the proportion of participants in the identification process, and n is the total number of participants in and out of the identification process. The geographical weight parameter is calculated as follows: The Euclidean distance between the participant and the sender is calculated on the server based on their geographical coordinates to obtain the geographical weight coefficient for each participant. The calculation method is as follows: Where, d i Let d represent the Euclidean distance between the i-th participant and the sender. i ′ Indicates the geographical weighting coefficient; The image recognition module is used to recognize the image data after each participant receives the image data and uses the local model pre-trained by each participant to obtain the recognized ship name result. The training method is as follows: the YOLO model, CRNN convolutional recurrent neural network and CTC neural network-based temporal classification network are combined to obtain the local training model; the local training model is initialized; each participant uses its own local ship name dataset to train the local training model to obtain its own personalized local model. The results upload module is used to upload the ship name results obtained by all selected participants to the server; The weight allocation module is used to allocate weights to all ship name results on the server according to the geographical weight coefficient, and obtain the weight coefficient of each ship name result. The result selection module is used to select the ship name result corresponding to the maximum weight coefficient and send it to the initiator to complete the task of jointly identifying the ship name.

5. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the method according to any one of claims 1 to 3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 3.

Images