Intelligent algorithm model calling method and system of unmanned system and model management system

Abstract

The invention provides an intelligent algorithm model calling method and system of an unmanned system and a model management system, and relates to the technical field of unmanned cluster system management, and the method comprises the steps: obtaining a to-be-detected image based on the unmanned system; calling a target intelligent algorithm model group matched with the to-be-detected image in a pre-established unmanned cluster target intelligent algorithm model group; according to an intelligent algorithm model loading rule, loading a target intelligent algorithm model matched with the to-be-detected image from the target intelligent algorithm model group; and based on the target intelligent algorithm model matched with the to-be-detected image, detecting and identifying the to-be-detected image to obtain a detection result of the to-be-detected image. According to the method, the target intelligent algorithm model set matched with the to-be-detected image is obtained during calling, then the matched intelligent algorithm model is obtained, the to-be-detected image is detected according to the matched intelligent algorithm model, special requirements of different unmanned systems in an unmanned cluster are met, and differential deployment of the target intelligent algorithm model is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned cluster system management, and particularly to an intelligent algorithm model calling method, system and model management system for unmanned systems. Background Art

[0002] With the rapid development of artificial intelligence technology, unmanned cluster systems, such as drone swarms, unmanned vehicle swarms, etc., have been widely used in military, logistics, agriculture, environmental monitoring and other fields. Unmanned cluster systems can achieve large-scale and high-efficiency task execution through collaborative operations, thus greatly improving work efficiency and execution capabilities.

[0003] However, since each unmanned system in the existing unmanned cluster system has different hardware architectures and computing capabilities, the update and maintenance of models need to be adapted to different hardware versions. The existing technology is difficult to deploy different intelligent algorithm models for unmanned cluster systems, cannot meet the special needs of different unmanned systems in the unmanned cluster, and cannot achieve the differential deployment of target intelligent algorithm models for different unmanned systems. Summary of the Invention

[0004] To overcome the deficiencies of the above-mentioned prior art, the present invention provides an intelligent algorithm model calling method for unmanned systems, including:

[0005] Obtaining a to-be-detected image based on the unmanned system;

[0006] Calling a target intelligent algorithm model group that matches the to-be-detected image in a pre-established unmanned cluster target intelligent algorithm model group;

[0007] Loading a target intelligent algorithm model that matches the to-be-detected image from the target intelligent algorithm model group according to the intelligent algorithm model loading rule;

[0008] Detecting and recognizing the to-be-detected image based on the target intelligent algorithm model that matches the to-be-detected image to obtain a detection result of the to-be-detected image;

[0009] Wherein, the unmanned cluster target intelligent algorithm model group is established by training a to-be-trained data set obtained from the to-be-trained images generated based on a plurality of model training configuration files to generate an initial intelligent algorithm model group, and then performing format conversion on the initial intelligent algorithm model group in combination with the hardware information of each unmanned system in the unmanned cluster.

[0010] Optionally, the establishment of the unmanned cluster target intelligent algorithm model group includes:

[0011] Generating a to-be-trained data set based on the obtained to-be-trained images;

[0012] Train the dataset to be trained according to multiple model training configuration files, generate initial intelligent algorithm models corresponding to each model training configuration file, and package each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group;

[0013] Based on the hardware information of each unmanned system in the unmanned cluster, obtain the initial intelligent algorithm models adapted to each unmanned system from the initial intelligent algorithm model group;

[0014] Convert each adapted initial intelligent algorithm model through format conversion to generate a target intelligent algorithm model, and package the target intelligent algorithm model with the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster.

[0015] Optionally, after obtaining the detection result of the image to be detected, the method further includes:

[0016] If the evaluation effect of the detection result of the image to be detected does not reach the preset threshold, obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system;

[0017] According to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, switch to other target intelligent algorithm models in the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the switched other target intelligent algorithm models until the detection result of the switched other target intelligent algorithm models reaches the preset threshold.

[0018] Optionally, generating the dataset to be trained based on the obtained images to be trained includes:

[0019] Generate a VOC dataset by data annotation of the images to be trained;

[0020] Generate a COCO dataset by format conversion of the VOC dataset;

[0021] Split the COCO dataset into a training set, a validation set, and a test set according to a preset ratio, and determine the training set as the dataset to be trained.

[0022] Optionally, after packaging each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group, the method further includes:

[0023] Store the initial intelligent algorithm model group in the model database, and query the model database to check whether there is a prior initial intelligent algorithm model group that matches the initial intelligent algorithm model group;

[0024] If it exists, combine the prior initial intelligent algorithm model group with the initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and store the updated initial intelligent algorithm model group in the model database.

[0025] Optionally, based on the hardware information of each unmanned system in the unmanned cluster, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group, including:

[0026] Generate an intelligent algorithm model deployment rule based on the hardware information of each unmanned system in the unmanned cluster and a preset deployment rule;

[0027] According to the intelligent algorithm model deployment rule, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group.

[0028] Optionally, after generating the target intelligent algorithm model by format conversion of each adapted initial intelligent algorithm model and packing the target intelligent algorithm model with the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster, including:

[0029] Convert the adapted initial intelligent algorithm model into an Open Neural Network Exchange (ONNX) model through model conversion;

[0030] Generate a TensorRT model by accelerating the ONNX model through a Graphics Processing Unit (GPU);

[0031] After the TensorRT model passes the model test, pack the adapted initial intelligent algorithm model, the ONNX model, the tested TensorRT model, and the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster.

[0032] Optionally, after generating the target intelligent algorithm model by format conversion of the adapted intelligent algorithm model and packing the target intelligent algorithm model with the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster, the method further includes:

[0033] Store the target intelligent algorithm model group in the on-board model database, and query the on-board model database to check if there is a prior target intelligent algorithm model group that matches the target intelligent algorithm model group;

[0034] If it exists, combine the prior target intelligent algorithm model group with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group, and store the updated target intelligent algorithm model group in the on-board model database.

[0035] Based on the same inventive concept, the present invention also provides an intelligent algorithm model call system for an unmanned system, including:

[0036] An image to be detected acquisition module, configured to acquire an image to be detected based on an unmanned system;

[0037] A target intelligent algorithm model group matching module, configured to call a target intelligent algorithm model group that matches the image to be detected in a pre-established unmanned cluster target intelligent algorithm model group;

[0038] A target intelligent algorithm model matching module, configured to load a target intelligent algorithm model that matches the image to be detected in the target intelligent algorithm model group;

[0039] A detection result acquisition module, configured to perform detection and recognition on the image to be detected based on the target intelligent algorithm model, and obtain a detection result of the image to be detected;

[0040] Wherein, the unmanned cluster target intelligent algorithm model group is trained on a to-be-trained data set generated from acquired images to be trained based on multiple model training configuration files to generate an initial intelligent algorithm model group, and combined with the hardware information of each unmanned system in the unmanned cluster, and established after format conversion of the initial intelligent algorithm model group.

[0041] Optionally, the system further includes an unmanned cluster target intelligent algorithm model group establishment module, configured to:

[0042] Generate a to-be-trained data set based on the acquired images to be trained;

[0043] Train the to-be-trained data set according to multiple model training configuration files to generate an initial intelligent algorithm model corresponding to each model training configuration file, and package each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group;

[0044] Based on the hardware information of each unmanned system in the unmanned cluster, obtain an initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group;

[0045] Generate a target intelligent algorithm model by format conversion of each adapted initial intelligent algorithm model, package the target intelligent algorithm model with the corresponding model training configuration file, and establish an unmanned cluster target intelligent algorithm model group.

[0046] Optionally, the system further includes a detection result correction module, configured to:

[0047] If the evaluation effect of the detection result of the image to be detected does not reach a preset threshold, obtain a status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system;

[0048] Based on the evaluation effect and the status monitoring result, and according to the intelligent algorithm model switching rule, switch to other target intelligent algorithm models in the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the other target intelligent algorithm model switched to, until the detection result of the other target intelligent algorithm model switched to reaches the preset threshold.

[0049] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0050] Generate a VOC dataset by data annotation of the image to be trained;

[0051] Generate a COCO dataset by format conversion of the VOC dataset;

[0052] Split the COCO dataset into a training set, a validation set and a test set according to a preset ratio, and determine the training set as the dataset to be trained.

[0053] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is also used for:

[0054] Store the initial intelligent algorithm model group in the model database, and query the model database to check if there is a prior initial intelligent algorithm model group that matches the initial intelligent algorithm model group;

[0055] If there is, combine the prior initial intelligent algorithm model group with the initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and store the updated initial intelligent algorithm model group in the model database.

[0056] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0057] Generate an intelligent algorithm model deployment rule based on the hardware information of each unmanned system in the unmanned cluster and the preset deployment rule;

[0058] Obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group according to the intelligent algorithm model deployment rule.

[0059] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0060] Generate an Open Neural Network Exchange (ONNX) model by model conversion of the adapted initial intelligent algorithm model;

[0061] Generate a TensorRT model in real time by accelerating the ONNX model with a Graphics Processing Unit (GPU);

[0062] After the TensorRT model passes the model test, the adapted initial intelligent algorithm model, ONNX model, the tested TensorRT model and the corresponding model training configuration file are packaged to establish the target intelligent algorithm model group of the unmanned cluster.

[0063] Optionally, the target intelligent algorithm model group establishment module for the unmanned cluster is further configured to:

[0064] Store the target intelligent algorithm model group in the on-board model database, and query the on-board model database to check if there is a prior target intelligent algorithm model group that matches the target intelligent algorithm model group;

[0065] If there is, combine the prior target intelligent algorithm model group with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group, and store the updated target intelligent algorithm model group in the on-board model database.

[0066] Based on the same inventive concept, the present invention also provides an intelligent algorithm model management system for an unmanned cluster, including: an intelligent algorithm model cloud visualization management interface, an intelligent algorithm model cloud background management service, and an unmanned system on-board software;

[0067] The intelligent algorithm model cloud visualization management interface is used to send data storage instructions, model training instructions, and model deployment instructions to the intelligent algorithm model cloud background management service to achieve visual operations for data management, model training, and model deployment; it is also used to send model switching instructions to the unmanned system on-board software to achieve visual operations for model switching;

[0068] The intelligent algorithm model cloud background management service is used to generate a training data set for the to-be-trained images received from the unmanned system on-board software; it is also used to receive the model training instructions sent by the intelligent algorithm model cloud visualization management interface, train the to-be-trained data set based on the model training configuration file, generate an initial intelligent algorithm model corresponding to each model training configuration file, package and combine each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group; it is also used to receive the data storage instructions sent by the intelligent algorithm model cloud visualization management interface and store the generated initial intelligent algorithm model group; it is also used to receive the model deployment instructions sent by the intelligent algorithm model cloud visualization management interface, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group based on the hardware information of each unmanned system in the unmanned cluster; generate a target intelligent algorithm model after format conversion of each adapted initial intelligent algorithm model, package the target intelligent algorithm model with the corresponding model training configuration file to form a target intelligent algorithm model group, and send the target intelligent algorithm model group to the unmanned system on-board software;

[0069] The airborne software of the unmanned system is used to collect images to be trained and send the images to be trained to the cloud background management service of the intelligent algorithm model; it is also used to send the target intelligent algorithm model group to the airborne model database of the unmanned system airborne software and wait for the unmanned system to call; it is also used to receive the model switching instruction sent by the intelligent algorithm model cloud visualization management interface, obtain the image to be detected based on the unmanned system; call the target intelligent algorithm model group matching the image to be detected from the pre-established unmanned cluster target intelligent algorithm model group in the airborne model database; according to the intelligent algorithm model loading rule, load the target intelligent algorithm model matching the image to be detected from the target intelligent algorithm model group; detect and identify the image to be detected based on the target intelligent algorithm model matching the image to be detected, and obtain the detection result of the image to be detected; it is also used to, when the evaluation result of the detection result of the image to be detected does not reach the preset threshold, obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system, and based on the evaluation effect and the status monitoring result, switch to other target intelligent algorithm models of the target intelligent algorithm model group according to the intelligent algorithm model switching rule, and re-detect and identify the image to be detected according to the switched other target intelligent algorithm model until the detection result of the switched other target intelligent algorithm model reaches the preset threshold.

[0070] Based on the same inventive concept, the present invention also provides an electronic device, including: at least one processor and a memory; the memory and the processor are connected by a bus;

[0071] The memory is used to store one or more programs;

[0072] When the one or more programs are executed by the at least one processor, the intelligent algorithm model calling method of an unmanned system as described above is implemented.

[0073] Based on the same inventive concept, the present invention also provides a readable storage medium, on which an execution program is stored, and when the execution program is executed, the intelligent algorithm model calling method of an unmanned system as described above is implemented.

[0074] Compared with the closest prior art, the beneficial effects of the present invention are as follows:

[0075] The present invention provides a method for calling an intelligent algorithm model of an unmanned system, including: obtaining an image to be detected based on the unmanned system; calling a target intelligent algorithm model group that matches the image to be detected in a pre-established unmanned cluster target intelligent algorithm model group; loading a target intelligent algorithm model that matches the image to be detected from the target intelligent algorithm model group according to the intelligent algorithm model loading rule; and performing detection and recognition on the image to be detected based on the target intelligent algorithm model that matches the image to be detected, so as to obtain a detection result of the image to be detected. The present invention discloses that when calling, a target intelligent algorithm model group that matches the image to be detected is obtained, and then a matching intelligent algorithm model is obtained, and the image to be detected is detected according to the matching intelligent algorithm model, so as to meet the special requirements of different unmanned systems in the unmanned cluster and realize the differential deployment of the target intelligent algorithm model.

[0076] The present invention also provides an intelligent algorithm model management system for unmanned clusters, including: an intelligent algorithm model cloud visualization management interface, an intelligent algorithm model cloud background management service, and unmanned system airborne software; The intelligent algorithm model cloud visualization management interface is used to send data storage instructions, model training instructions, and model deployment instructions to the intelligent algorithm model cloud background management service to achieve visual operations on data management, model training, and model deployment; It is also used to send model switching instructions to the unmanned system airborne software to achieve visual operations on model switching; The intelligent algorithm model cloud background management service is used to generate a training data set for the to-be-trained images received from the unmanned system airborne software; It is also used to receive the model training instructions sent by the intelligent algorithm model cloud visualization management interface, train the to-be-trained data set based on the model training configuration file, generate an initial intelligent algorithm model corresponding to each model training configuration file, and package and combine each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group; It is also used to receive the data storage instructions sent by the intelligent algorithm model cloud visualization management interface and store the generated initial intelligent algorithm model group; It is also used to receive the model deployment instructions sent by the intelligent algorithm model cloud visualization management interface, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group based on the hardware information of each unmanned system in the unmanned cluster; Generate a target intelligent algorithm model after format conversion for each adapted initial intelligent algorithm model, package the target intelligent algorithm model with the corresponding model training configuration file to form a target intelligent algorithm model group, and send the target intelligent algorithm model group to the unmanned system airborne software; The unmanned system airborne software is used to collect to-be-trained images and send the to-be-trained images to the intelligent algorithm model cloud background management service; It is also used to send the target intelligent algorithm model group to the airborne model database of the unmanned system airborne software and wait for the unmanned system to call; It is also used to receive the model switching instructions sent by the intelligent algorithm model cloud visualization management interface, obtain the to-be-detected images based on the unmanned system; Call the target intelligent algorithm model group matching the to-be-detected images from the pre-established unmanned cluster target intelligent algorithm model group in the airborne model database; According to the intelligent algorithm model loading rule, load the target intelligent algorithm model matching the to-be-detected images from the target intelligent algorithm model group; Based on the target intelligent algorithm model matching the to-be-detected images, detect and identify the to-be-detected images to obtain the detection result of the to-be-detected images;It is also used to obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system when the evaluation result of the detection result of the image to be detected does not reach the preset threshold. According to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, switch to other target intelligent algorithm models in the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the other target intelligent algorithm models switched to until the detection result of the other target intelligent algorithm models switched to reaches the preset threshold. Through the unified management platform, the present invention realizes the integration of the development, deployment and operation management of the intelligent algorithm model, reduces the time for developers to switch between different environments, and improves the development efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] Figure 1 It is a schematic flowchart of a method for calling an intelligent algorithm model of an unmanned system provided by the present invention;

[0078] Figure 2 It is a schematic flowchart of data set production, training and management provided by the present invention;

[0079] Figure 3 It is a schematic flowchart of model deployment and management provided by the present invention;

[0080] Figure 4 It is a schematic flowchart of an intelligent algorithm model calling mechanism of an unmanned system provided by the present invention;

[0081] Figure 5 It is a structural diagram of an intelligent algorithm model calling system of an unmanned system provided by the present invention;

[0082] Figure 6 It is a schematic diagram of the overall framework of an intelligent algorithm model management system for an unmanned cluster provided by the present invention;

[0083] Figure 7 It is a schematic flowchart of the overall process of an intelligent algorithm model management system for an unmanned cluster provided by the present invention;

[0084] Figure 8 It is a schematic diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0085] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the drawings are exemplary and are intended to explain the present invention and should not be construed as limiting the present invention.

[0086] Embodiment 1:

[0087] The present invention provides a method for invoking an intelligent algorithm model of an unmanned system. Specifically, Figure 1 is a schematic flowchart of the method for invoking the intelligent algorithm model of the unmanned system provided by the embodiment of the present invention. As shown in the figure, the method includes the following steps:

[0088] S1: Obtain an image to be detected based on the unmanned system;

[0089] S2: Invoke a target intelligent algorithm model group that matches the image to be detected from a pre-established unmanned cluster target intelligent algorithm model group;

[0090] S3: Load a target intelligent algorithm model that matches the image to be detected from the target intelligent algorithm model group according to the intelligent algorithm model loading rule;

[0091] S4: Detect and identify the image to be detected based on the target intelligent algorithm model that matches the image to be detected, and obtain the detection result of the image to be detected;

[0092] Among them, the unmanned cluster target intelligent algorithm model group is trained on a training dataset generated from the obtained images to be trained based on multiple model training configuration files to generate an initial intelligent algorithm model group, and combined with the hardware information of each unmanned system in the unmanned cluster, and established after format conversion of the initial intelligent algorithm model group.

[0093] The present invention discloses the concept of a target intelligent algorithm model group, that is, combining and packaging multiple models with the same dataset but different configuration files, discloses obtaining a target intelligent algorithm model group that matches the image to be detected during invocation, and then obtaining a matching intelligent algorithm model, and detecting the image to be detected according to the matching intelligent algorithm model, meeting the special requirements of different unmanned systems in the unmanned cluster, and realizing the differential deployment of the target intelligent algorithm model.

[0094] In this embodiment, it is necessary to first establish an unmanned cluster target intelligent algorithm model group. The construction process of the unmanned cluster target intelligent algorithm model group is as follows:

[0095] Generate a training dataset based on the obtained images to be trained;

[0096] Train the training dataset according to multiple model training configuration files to generate an initial intelligent algorithm model corresponding to each model training configuration file, and package each model training configuration file with the corresponding initial intelligent algorithm model respectively to generate an initial intelligent algorithm model group;

[0097] Based on the hardware information of each unmanned system in the unmanned cluster, obtain an initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group;

[0098] Generate a target intelligent algorithm model by converting the format of each adapted initial intelligent algorithm model, package the target intelligent algorithm model with the corresponding model training configuration file, and establish a target intelligent algorithm model group for the unmanned cluster.

[0099] In a specific implementation manner, generate a training dataset based on the acquired images to be trained, including: generating a VOC (PASCAL Visual Object Classes) dataset by data annotation of the images to be trained; generating a COCO (Microsoft Common Objects in Context) dataset by converting the format of the VOC dataset; splitting the COCO dataset into a training set, a validation set, and a test set according to a preset ratio, and determining the training set as the training dataset.

[0100] Make the acquired images to be trained into a VOC-format dataset through data annotation, and then convert it into a COCO-format dataset through format conversion for easy model training. Then, split the COCO dataset into a training set, a validation set, and a test set according to a preset ratio, and store them in the COCO database. Determine the training set as the training dataset.

[0101] Figure 2 It is a schematic diagram of the process for dataset production, training, and management provided by the present invention.

[0102] As Figure 2 shown, in some specific implementation manners, use the acquired original images and calibration XML (Extensible Markup Language) files to make a VOC-format dataset and save it in the VOC database; then convert the VOC-format dataset to generate VOC dataset 1, VOC dataset 2, VOC dataset 3,..., VOC dataset n. After merging each VOC dataset, further classify and split the data into a training set, a validation set, and a test set, and store them in the COCO database. Determine the training set as the training dataset.

[0103] After determining the training dataset, train the training dataset according to multiple model training configuration files to generate an initial intelligent algorithm model corresponding to each model training configuration file, and package each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group.

[0104] In some specific embodiments, the data set in the COCO database is imported into the model training unit, and different model training configuration files are also imported, such as configuration file 1, configuration file 2, configuration file 3, ……, configuration file n. The same training data set is trained according to different model training configuration files to generate initial intelligent algorithm models corresponding to each model training configuration file, that is, initial intelligent algorithm model 1, initial intelligent algorithm model 2, initial intelligent algorithm model 3, ……, initial intelligent algorithm model n are obtained. Configuration file 1, configuration file 2, configuration file 3, ……, configuration file n are respectively packaged with initial intelligent algorithm model 1, initial intelligent algorithm model 2, initial intelligent algorithm model 3, ……, initial intelligent algorithm model n to obtain an initial intelligent algorithm model group composed of configuration file 1 + initial intelligent algorithm model 1, configuration file 2 + initial intelligent algorithm model 2, configuration file 3 + initial intelligent algorithm model 3, ……, configuration file n + initial intelligent algorithm model n, and the initial intelligent algorithm model group is stored in the model database.

[0105] In some alternative embodiments, after generating the initial intelligent algorithm model group, it further includes: storing the initial intelligent algorithm model group in the model database, and querying the model database to check if there is a prior initial intelligent algorithm model group that matches the initial intelligent algorithm model group; if so, combining the prior initial intelligent algorithm model group with the initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and storing the updated initial intelligent algorithm model group in the model database.

[0106] After generating the initial intelligent algorithm model group, based on the hardware information of each unmanned system in the unmanned cluster, the initial intelligent algorithm model adapted to each unmanned system is obtained from the initial intelligent algorithm model group.

[0107] Based on the hardware information of each unmanned system in the unmanned cluster and the preset deployment rules, intelligent algorithm model deployment rules are generated; according to the intelligent algorithm model deployment rules, the initial intelligent algorithm model adapted to each unmanned system is obtained from the initial intelligent algorithm model group.

[0108] After the unmanned system obtains the adapted initial intelligent algorithm model, each adapted initial intelligent algorithm model is converted through format conversion to generate a target intelligent algorithm model, and the target intelligent algorithm model is packaged with the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster.

[0109] In the process of establishing the intelligent algorithm model group for unmanned clusters, when training the training dataset to be processed, the concept of the intelligent algorithm model group is introduced, that is, the same training dataset to be processed is trained according to different model training configuration files to obtain different intelligent algorithm models. The same training dataset to be processed is combined and packaged with different intelligent algorithm models to meet the special needs of different unmanned systems in the unmanned cluster, and the differential deployment of the intelligent algorithm models is realized.

[0110] Figure 3 This is a schematic diagram of the process of model deployment and management provided by the present invention.

[0111] As Figure 3 shown, one or more initial intelligent algorithm models after matching are subjected to format conversion. First, they are converted into ONNX (Open Neural Network Exchange) intermediate format models, which are suitable for unmanned systems without GPUs (Graphics Processing Units). Then, they are converted into TensorRT (Tensor Runtime) format models through GPU acceleration, which are suitable for unmanned systems with GPUs. Model packaging and storage means that the initial intelligent algorithm models, the models in various formats after model conversion, and the corresponding configuration files are packaged and grouped in the form of a model group and stored in the on-board model database.

[0112] In some specific embodiments, according to the custom model deployment rules and hardware information generation rules, a model request is sent to the initial target intelligent model group in the model database according to the generated rules, and the initial intelligent algorithm models suitable for the unmanned system are screened out. Then, the initial intelligent algorithm models are downloaded. For example, the adapted initial intelligent algorithm models are Initial Intelligent Algorithm Model 1, Initial Intelligent Algorithm Model 2, and Initial Intelligent Algorithm Model 4, and the corresponding model training configuration files are Configuration File 1, Configuration File 2, and Configuration File 4. Next, Initial Intelligent Algorithm Model 1, Initial Intelligent Algorithm Model 2, and Initial Intelligent Algorithm Model 4 are converted into ONNX and TensorRT formats to obtain ONNX Model 1, ONNX Model 2, ONNX Model 4, TensorRT Model 1, TensorRT Model 2, and TensorRT Model 4. Further, TensorRT Model 1, TensorRT Model 2, and TensorRT Model 4 are tested and feedback is obtained, resulting in the tested TensorRT Model 1, TensorRT Model 2, and TensorRT Model 4. Finally, Initial Intelligent Algorithm Model 1, ONNX Model 1, the tested TensorRT Model 1, and Configuration File 1 are determined as a group of target intelligent algorithm models, namely Target Intelligent Algorithm Model 1; similarly, the other models and configuration files are also determined as target intelligent algorithm models, namely Target Intelligent Algorithm Model 2 and Target Intelligent Algorithm Model 4. Each group of target intelligent algorithm models, that is, Initial Intelligent Algorithm Model 1, Initial Intelligent Algorithm Model 2, Initial Intelligent Algorithm Model 4, ONNX Model 1, ONNX Model 2, ONNX Model 4, the tested TensorRT Model 1, TensorRT Model 2, TensorRT Model 4, Configuration File 1, Configuration File 2, and Configuration File 4 are packaged to establish the target intelligent algorithm model group of the unmanned cluster and stored in the on-board model database.

[0113] In some alternative embodiments, after the target intelligent algorithm model is generated by format conversion of the adapted intelligent algorithm model and the target intelligent algorithm model is packaged with the corresponding model training configuration file to establish the target intelligent algorithm model group of the unmanned cluster, the method further includes: storing the target intelligent algorithm model group in the on-board model database and querying the on-board model database to check if there is a prior target intelligent algorithm model group that matches the target intelligent algorithm model group; if so, combining the prior target intelligent algorithm model group with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group and storing the updated target intelligent algorithm model group in the on-board model database.

[0114] The above details the establishment process of the target intelligent algorithm model group of the unmanned cluster. Next, the intelligent algorithm model invocation process of the unmanned system is introduced.

[0115] In an unmanned cluster, multiple unmanned systems may be deployed. If one of the unmanned systems calls an intelligent algorithm model, the specific process includes:

[0116] Obtain the image to be detected collected by the unmanned system, and call the target intelligent algorithm model group that matches the image to be detected in the pre-established unmanned cluster target intelligent algorithm model group; then, according to the intelligent algorithm model loading rule, load the target intelligent algorithm model that matches the image to be detected from the target intelligent algorithm model group; finally, detect and identify the image to be detected based on the target intelligent algorithm model that matches the image to be detected, and obtain the detection result of the image to be detected.

[0117] In an alternative embodiment, after obtaining the detection result of the image to be detected, the detection result of the image to be detected can also be evaluated to further correct the detection result of the image to be detected. Specifically, if the evaluation effect of the detection result of the image to be detected does not reach the preset threshold, obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system; according to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, switch to another target intelligent algorithm model in the target intelligent algorithm model group, and perform re-detection and identification on the image to be detected according to the switched target intelligent algorithm model until the detection result of the switched target intelligent algorithm model reaches the preset threshold. Specifically, the switching process includes: obtaining the detection result of the image to be detected again; if the evaluation effect of the re-obtained detection result reaches the preset threshold, output the detection result of the current image to be detected; if the evaluation effect of the re-obtained detection result does not reach the preset threshold, continue to switch to another target intelligent algorithm model in the target intelligent algorithm model group until the detection result reaches the preset threshold.

[0118] When detecting the image to be detected, combining the hardware status monitoring mechanism and the intelligent algorithm model switching rule during the operation of the unmanned system can automatically switch the intelligent algorithm model when the obtained detection result does not meet the expectation, enhancing the stability and reliability of the unmanned cluster system.

[0119] Figure 4 This is a schematic flowchart of the intelligent algorithm model calling mechanism for the unmanned system provided by the present invention.

[0120] As Figure 4As shown, in some specific embodiments, after the image to be detected is input, a matching target intelligent algorithm model group is loaded from the airborne model database, and then a matching target intelligent algorithm model is further loaded. For example, the above-mentioned target intelligent algorithm model group is loaded, and then a matching target intelligent algorithm model 2 is further loaded. The image to be detected is detected and recognized according to the target intelligent algorithm model 2 to obtain the detection result of the image to be detected. The image to be detected with the obtained detection result can also be stored in the original dataset, and the obtained image to be detected can further update the unmanned cluster target intelligent algorithm model to optimize the unmanned cluster target intelligent algorithm model. Further, the evaluation effect of the detection result of the image to be detected can be obtained. If the evaluation effect does not reach the preset threshold, according to the hardware status monitoring mechanism of the unmanned system, such as GPU usage monitoring and CPU (Central Processing Unit) usage monitoring, the status monitoring result of the unmanned system is obtained; according to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, the target intelligent algorithm model 1 can be switched, and the image to be detected is detected and recognized again according to the target intelligent algorithm model 1 to obtain the detection result of the image to be detected. Further evaluation, switching, and detection are performed according to the detection result to obtain the final detection result of the image to be detected.

[0121] It can be understood that those skilled in the art can also set the specific evaluation effect method and the specific preset threshold according to actual needs, which are not limited herein.

[0122] The intelligent algorithm model calling method for the unmanned system provided by the present invention simplifies the development and deployment processes of the intelligent algorithm model through a unified management platform, improving the deployment efficiency and reliability. The concept of "model group" is innovatively introduced, that is, multiple models based on the same dataset but with different configurations are combined and packaged to meet the special needs of different unmanned systems and achieve differential deployment of the models. In addition, during the implementation of the intelligent algorithm model calling of the unmanned system, combined with the hardware status monitoring mechanism, the custom model switching rule, and the actual effect of the model operation during the operation of the unmanned system, the model can be intelligently loaded from the "model group" and automatically switched during operation, significantly enhancing the stability and reliability of the system.

[0123] Embodiment 2:

[0124] Based on the same inventive concept, the present invention also provides an intelligent algorithm model calling system 500 for an unmanned system. The structure of the system is as Figure 5 shown. The system 500 includes:

[0125] An image to be detected obtaining module 501, configured to obtain an image to be detected based on the unmanned system;

[0126] The target intelligent algorithm model group matching module 502 is used to call the target intelligent algorithm model group that matches the to-be-detected image in the pre-established unmanned cluster target intelligent algorithm model group;

[0127] The target intelligent algorithm model matching module 503 is used to load the target intelligent algorithm model that matches the to-be-detected image in the target intelligent algorithm model group;

[0128] The detection result obtaining module 504 is used to perform detection and recognition on the to-be-detected image based on the target intelligent algorithm model, and obtain the detection result of the to-be-detected image;

[0129] Among them, the unmanned cluster target intelligent algorithm model group is trained on the to-be-trained data set generated from the obtained to-be-trained images based on multiple model training configuration files to generate an initial intelligent algorithm model group, and combined with the hardware information of each unmanned system in the unmanned cluster, the initial intelligent algorithm model group is established after format conversion.

[0130] Optionally, the system further includes an unmanned cluster target intelligent algorithm model group establishment module, which is used for:

[0131] Generate a to-be-trained data set based on the obtained to-be-trained images;

[0132] Train the to-be-trained data set according to multiple model training configuration files to generate an initial intelligent algorithm model corresponding to each model training configuration file, and package each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group;

[0133] Based on the hardware information of each unmanned system in the unmanned cluster, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group;

[0134] Generate a target intelligent algorithm model after format conversion for each adapted initial intelligent algorithm model, and package the target intelligent algorithm model with the corresponding model training configuration file to establish the unmanned cluster target intelligent algorithm model group.

[0135] Optionally, the system further includes a detection result correction module, which is used for:

[0136] If the evaluation effect of the detection result of the to-be-detected image does not reach the preset threshold, obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system;

[0137] According to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, switch to other target intelligent algorithm models in the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the other target intelligent algorithm model switched to, until the detection result of the other target intelligent algorithm model switched to reaches the preset threshold.

[0138] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0139] Generate a VOC dataset by data annotation of the image to be trained;

[0140] Generate a COCO dataset by format conversion of the VOC dataset;

[0141] Split the COCO dataset into a training set, a validation set, and a test set according to a preset ratio, and determine the training set as the dataset to be trained.

[0142] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is also used for:

[0143] Store the initial intelligent algorithm model group in the model database, and query the model database to check if there is a prior initial intelligent algorithm model group that matches the initial intelligent algorithm model group;

[0144] If there is, combine the prior initial intelligent algorithm model group with the initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and store the updated initial intelligent algorithm model group in the model database.

[0145] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0146] Generate an intelligent algorithm model deployment rule based on the hardware information of each unmanned system in the unmanned cluster and the preset deployment rule;

[0147] Obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group according to the intelligent algorithm model deployment rule.

[0148] Optionally, the unmanned cluster target intelligent algorithm model group establishment module is specifically used for:

[0149] Generate an Open Neural Network Exchange (ONNX) model by model conversion of the adapted initial intelligent algorithm model;

[0150] Generate a TensorRT model in real time by accelerating the ONNX model with a Graphics Processing Unit (GPU);

[0151] After the TensorRT model passes the model test, the adapted initial intelligent algorithm model, ONNX model, the tested TensorRT model and the corresponding model training configuration file are packaged to establish the target intelligent algorithm model group for the unmanned cluster.

[0152] Optionally, the target intelligent algorithm model group establishment module for the unmanned cluster is further configured to:

[0153] Store the target intelligent algorithm model group in the on-board model database, and query the on-board model database to check if there is a prior target intelligent algorithm model group that matches the target intelligent algorithm model group;

[0154] If there is, combine the prior target intelligent algorithm model group with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group, and store the updated target intelligent algorithm model group in the on-board model database.

[0155] Embodiment 3:

[0156] Based on the same inventive concept, the present invention also provides an intelligent algorithm model management system for an unmanned cluster, as Figure 6 shown, which is a schematic diagram of the overall framework of the intelligent algorithm model management system for an unmanned cluster provided by the present invention. As Figure 7 shown, which is a schematic diagram of the overall process of the intelligent algorithm model management system for an unmanned cluster provided by the present invention.

[0157] An intelligent algorithm model management system for an unmanned cluster, implemented based on the B / S (Browser / Server, front-end - server) architecture, consists of an intelligent algorithm model cloud visualization management interface, an intelligent algorithm model cloud background management service, and an unmanned system on-board software, and realizes core functions such as dataset management, intelligent algorithm model training, intelligent algorithm model deployment, intelligent algorithm model invocation, and integrated management.

[0158] The intelligent algorithm model cloud visualization management interface is built based on Vue.js (Vue JavaScript, a progressive JavaScript framework for building user interfaces), provides an intuitive operation interface for users, is a user-friendly interaction interface, and includes: a data management interface, a model training management interface, a model deployment management interface, and an on-board model management interface, supporting users' visual operations on the training, deployment, invocation, and management of datasets and intelligent algorithm models, enabling users to efficiently develop models, manage models, train models, deploy models, and monitor the running status of on-board models.

[0159] The intelligent algorithm model cloud background management service is implemented based on Node.js (a JavaScript runtime environment based on the Chrome V8 engine) and MongoDB (a database based on distributed file storage). The MongoDB database is used to store model data, training data, test data, and running data.

[0160] Axios (a JavaScript library for making HTTP requests in browsers and Node.js) is used for data exchange between the intelligent algorithm model cloud visualization management interface and the intelligent algorithm model cloud background management service. The management and execution process scheduling logic for tasks such as model training, model testing, and model deployment is written using Node.js.

[0161] The unmanned system airborne software is implemented based on MQTT (Message Queuing Telemetry Transport), HTTP (Hyper Text Transfer Protocol), and the MMDetection (an open-source object detection based on deep learning) framework. It supports real-time online model updates to ensure that unmanned systems in the unmanned cluster can obtain the latest version of the target intelligent algorithm model in a timely manner. The MMDetection framework provides a test verification and running environment for the model.

[0162] The intelligent algorithm model cloud visualization management interface provides a visual operation interface for the intelligent algorithm model management system for the unmanned cluster. Specifically, when the intelligent algorithm model management system manages data, it sends a data storage instruction to the intelligent algorithm model cloud background management service based on the data management interface; when the intelligent algorithm model management system trains a model, it sends a model training instruction to the intelligent algorithm model cloud background management service based on the model training management interface; when the intelligent algorithm model management system deploys a model, it sends a model deployment instruction to the intelligent algorithm model cloud background management service based on the model deployment management interface; when the intelligent algorithm model management system manages the airborne model, it sends a model switching instruction to the unmanned system airborne software based on the airborne model management interface.

[0163] The intelligent algorithm model cloud background management service includes a dataset management unit, a database, a model training unit, and a model deployment unit.

[0164] Among them, the dataset management unit is used to perform operations such as dataset production, dataset conversion, dataset storage, and management on the collected images. Among them, when producing the dataset, the collected images are made into a dataset in VOC (PASCAL Visual Object Classes) format through data annotation and stored in the VOC database; dataset conversion is to convert the VOC dataset into a dataset in COCO (Microsoft Common Objects in Context) format for convenient model training; dataset storage and management is to store the VOC-format dataset and the COCO-format dataset in the dataset database and perform unified management and invocation through the intelligent algorithm model cloud visualization management interface.

[0165] The database includes a dataset database and a model database. The VOC database and the COCO database are both dataset databases. The dataset database is used to store the VOC database with calibrated datasets and the COCO dataset of the dataset to be trained after format conversion; the model database is used to save the model training results.

[0166] The model training unit is used to perform operations such as model training, model packaging, and model version management. Model training is used to train various types of intelligent algorithm models such as deep learning models and reinforcement learning models, and perform multiple trainings of the dataset to be trained with multiple configuration files to achieve functions such as dataset configuration, algorithm configuration, and training configuration; when packaging the model, multiple models with different model training configuration files for the same dataset are combined and packaged into an intelligent algorithm model group for convenient model management and differential deployment; model version management is used to manage different versions of the model for convenient backtracking and comparison.

[0167] In an alternative embodiment, it is possible to query the model database to check if there is a prior initial intelligent algorithm model group that matches the current generated initial intelligent algorithm model group; if so, combine the prior initial intelligent algorithm model group with the currently generated initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and store the updated initial intelligent algorithm model group in the model database.

[0168] The model deployment unit is used to deploy the trained initial intelligent algorithm model to each unmanned system in the unmanned cluster, including operations such as model distribution, model conversion, and model packaging and storage. The model distribution function is to push each initial intelligent algorithm model in the initial intelligent algorithm model group to each unmanned system in the unmanned cluster. The specific execution process includes two parts: rule generation and model download. Among them, rule generation means generating rules according to the custom model deployment rules and hardware information during model deployment, which are used to match the initial intelligent algorithm models suitable for the unmanned system to achieve differential deployment. Model download means downloading one or more matched initial intelligent algorithm models and configuration files from the initial intelligent algorithm model group in the model database. The model conversion function is to convert the format of one or more matched initial intelligent algorithm models. First, it is converted into the ONNX intermediate format model, which is suitable for unmanned systems without GPUs, and then it is converted into the TensorRT format model through GPU acceleration, which is suitable for unmanned systems with GPUs. Model packaging and storage is to package and group the initial intelligent algorithm model, the models in various formats after model conversion, and the corresponding configuration files in the form of a model group and store them in the on-board model database.

[0169] In an alternative embodiment, it is possible to query the on-board model database to check if there is a prior target intelligent algorithm model group that matches the target intelligent algorithm model group; if it exists, then combine the prior target intelligent algorithm model group with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group and store the updated target intelligent algorithm model group in the on-board model database.

[0170] The unmanned system on-board software includes image acquisition, a model invocation mechanism, and an on-board model database, and is used to implement the invocation and management of the intelligent algorithm model. Implementing the invocation of the intelligent algorithm model ensures that the unmanned system can correctly load the appropriate target intelligent algorithm model and implement the switching and effect evaluation of the target intelligent algorithm model, including functions such as model loading, model switching, and effect evaluation. Among them, during model loading and model switching, a suitable intelligent algorithm model is loaded from the intelligent algorithm model group according to the model loading and switching rules and the output of the hardware status monitoring mechanism, and intelligent switching is performed according to the results feedback by the effect evaluation and the custom switching rules. The effect evaluation is the overall operation effect evaluation based on the hardware operation status and the model operation effect, and is used to provide a basis for the switching of the target intelligent algorithm model.

[0171] The integrated management function is responsible for coordinating the development, deployment, and operation management of the intelligent algorithm model, including model scheduling and model monitoring. Model invocation schedules model training and deployment tasks according to the cluster task requirements; model monitoring is responsible for monitoring the performance of the model in actual applications, collecting, and displaying the operation data.

[0172] In an intelligent algorithm model management system for unmanned clusters, the image acquisition device of the unmanned system acquires images to form an original image dataset. After manual calibration, it is uploaded to the VOC database through the VOC dataset visualization operation interface to form a VOC calibrated dataset. After format conversion, a COCO dataset is formed and stored in the COCO database. The model training unit receives the model training instruction from the model training cloud visualization management interface of the intelligent algorithm model, reads the dataset from the COCO database, trains the dataset, and stores the training result in the model database in the form of an initial intelligent algorithm model group; during model deployment, after receiving the model deployment instruction from the model training cloud visualization management interface of the intelligent algorithm model, the model deployment unit broadcasts the initial intelligent algorithm model to each unmanned system in the cluster. Each unmanned system in the cluster downloads the corresponding model from the model database according to the received information, and converts and deploys the model according to its own hardware information, and saves the deployed target intelligent algorithm model to the on-board model database. When the model is loaded and called, the unmanned system automatically loads and records the models in the model database in the form of a model group, detects and identifies the acquired images, and uploads the running status of the model to the model monitoring and management visualization operation interface.

[0173] In the design and implementation process of existing unmanned cluster systems, the development and operation management of intelligent algorithm models face many challenges: First, there is a lack of an integrated management platform. Most existing solutions focus on specific links such as model training or model deployment, lacking a unified management platform to coordinate the entire life cycle; Second, the difficulty of model deployment is high. Each unmanned system member in the unmanned cluster system may have different hardware architectures and computing capabilities, and model updates and maintenance need to be adapted to different hardware versions; Third, the challenge of model operation management. The unmanned cluster system will face various dynamically changing environments and task requirements during actual operation. Due to the lack of a real-time monitoring mechanism, the performance of the model during actual operation is difficult to monitor in real time, and it is difficult to discover and solve problems in a timely manner.

[0174] The intelligent algorithm model management system for unmanned clusters disclosed in the present invention constructs a unified management platform by integrating advanced intelligent algorithms and efficient automation tools, realizing the efficient development, automated deployment, efficient operation and flexible adjustment of intelligent algorithm models.

[0175] Improve development efficiency: Through a unified management platform, the integration of intelligent algorithm model development, deployment and operation management is realized, reducing the time for developers to switch between different environments and improving development efficiency.

[0176] Simplify the deployment process: The automated model deployment and update mechanism reduces the need for manual intervention, improving the efficiency and reliability of model deployment.

[0177] Achieve differential deployment: The concept of an intelligent algorithm model group is proposed. According to the rule settings and hardware information of each unmanned system, differential model deployment of the model group for the same data set is achieved, improving the adaptability of the intelligent algorithm model.

[0178] Intelligent model switching: Through the introduced hardware status monitoring mechanism and custom model switching rules, automatic switching of the intelligent algorithm model during operation is achieved.

[0179] Enhance real-time monitoring: By monitoring the running status of the model in real time, potential problems are discovered and processed in a timely manner, improving the stability and reliability of the system.

[0180] Embodiment 4:

[0181] Based on the same inventive concept, as Figure 8 shown, the present invention also provides an electronic device, which may be a computer device, a single-chip microcomputer device, an intelligent mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, the processor, and the transceiver component are connected by a bus; the memory can be used to store an execution program, and an exemplary execution program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, and this data can be called and / or modified when the instructions are executed.

[0182] The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the readable storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of an intelligent algorithm model calling method for an unmanned system in the above embodiment.

[0183] Embodiment 5:

[0184] Based on the same inventive concept, the present invention also provides a readable storage medium, specifically an electronic device-readable storage medium (Memory). The electronic device-readable storage medium is a memory device in the electronic device and is used to store programs and data. It can be understood that the readable storage medium here can include both the built-in storage medium in the electronic device and, of course, the extended storage medium supported by the electronic device. The storage medium provides a storage space, and this storage space stores the operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are stored in this storage space, and these instructions can be one or more executable programs (including program codes). It should be noted that the storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the storage medium can be loaded and executed by the processor to implement the steps of the intelligent algorithm model calling method for an unmanned system in the above-mentioned embodiments.

[0185] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program codes.

[0186] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0187] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory generate a manufactured article including an instruction device, and the instruction device implements the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0188] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one process or multiple processes and / or blocks Figure 1 one process or multiple processes and / or blocks Figure 1 or steps for implementing the functions specified in one block or multiple blocks.

[0189] 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 the scope of its protection. Although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that after reading the present invention, various changes, modifications or equivalent replacements can still be made to the specific implementation manners of the application. However, these changes, modifications or equivalent replacements are all within the scope of the protection of the claims pending for the application.

Claims

1. A method for calling an intelligent algorithm model of an unmanned system, characterized in that: include: Acquire the image to be detected based on the unmanned system; Calling a target intelligent algorithm model group that matches the image to be detected in a pre-established unmanned cluster target intelligent algorithm model group; According to the intelligent algorithm model loading rule, a target intelligent algorithm model matching the image to be detected is loaded from the target intelligent algorithm model group; Based on the target intelligent algorithm model matched with the image to be detected, the image to be detected is detected and identified to obtain the detection result of the image to be detected; Among them, the unmanned cluster target intelligent algorithm model group is established after format conversion of the initial intelligent algorithm model group based on a plurality of model training configuration files, training the training data set generated by the acquired training images, generating an initial intelligent algorithm model group, and combining the hardware information of each unmanned system in the unmanned cluster.

2. The method according to claim 1, characterized in that The establishment of the unmanned cluster target intelligent algorithm model group includes: Generate a training data set based on the acquired training images; The training data set to be trained is trained according to a plurality of model training configuration files, an initial intelligent algorithm model corresponding to each of the model training configuration files is generated, and each of the model training configuration files is packaged with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group; Based on the hardware information of each unmanned system in the unmanned cluster, an initial intelligent algorithm model adapted to each unmanned system is obtained from the initial intelligent algorithm model group; Each adapted initial intelligent algorithm model is converted into a target intelligent algorithm model through format conversion, and the target intelligent algorithm model is packaged with the corresponding model training configuration file to establish a target intelligent algorithm model group for the unmanned cluster.

3. The method according to claim 1, characterized in that After obtaining the detection result of the image to be detected, the method further includes: If the evaluation effect of the detection result of the image to be detected does not reach the preset threshold, obtaining the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system; According to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rules, switch to other target intelligent algorithm models in the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the other target intelligent algorithm models switched to, until the detection result of the other target intelligent algorithm model switched to reaches the preset threshold.

4. The method according to claim 2, characterized in that: The step of generating a training data set based on the acquired training images includes: Generate a VOC data set by annotating the image to be trained; Generate a COCO dataset by converting the VOC dataset into a COCO dataset; The COCO dataset is split into a training set, a validation set, and a test set according to a preset ratio, and the training set is determined as the dataset to be trained.

5. The method according to claim 4, characterized in that After packaging each model training configuration file with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group, the method further includes: The initial intelligent algorithm model group is stored in a model database, and the model database is queried to determine whether there is a previous initial intelligent algorithm model group that matches the initial intelligent algorithm model group; If so, the previous initial intelligent algorithm model group is combined with the initial intelligent algorithm model group to generate an updated initial intelligent algorithm model group, and the updated initial intelligent algorithm model group is stored in the model database.

6. The method according to claim 5, characterized in that The step of obtaining an initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group based on the hardware information of each unmanned system in the unmanned cluster includes: Generate intelligent algorithm model deployment rules based on the hardware information and preset deployment rules of each unmanned system in the unmanned cluster; According to the intelligent algorithm model deployment rules, an initial intelligent algorithm model adapted to each unmanned system is obtained from the initial intelligent algorithm model group.

7. The method according to claim 6, characterized in that The method converts each adapted initial intelligent algorithm model into a target intelligent algorithm model by format conversion, packages the target intelligent algorithm model with the corresponding model training configuration file, and establishes a target intelligent algorithm model group of the unmanned cluster, including: The adapted initial intelligent algorithm model is converted into an open neural network exchange ONNX model; The ONNX model is accelerated by the image processor GPU to generate a tensor real-time TensorRT model; After the TensorRT model is tested, the adapted initial intelligent algorithm model, ONNX model, the tested TensorRT model and the corresponding model training configuration file are packaged to establish the target intelligent algorithm model group of the unmanned cluster.

8. The method according to claim 7, characterized in that After converting the adapted intelligent algorithm model into a target intelligent algorithm model through format conversion, packaging the target intelligent algorithm model with a corresponding model training configuration file, and establishing a target intelligent algorithm model group of an unmanned cluster, the method further includes: The target intelligent algorithm model group is stored in an onboard model database, and the onboard model database is queried to determine whether there is a previous target intelligent algorithm model group that matches the target intelligent algorithm model group; If so, the prior target intelligent algorithm model group is combined with the target intelligent algorithm model group to generate an updated target intelligent algorithm model group, and the updated target intelligent algorithm model group is stored in the onboard model database.

9. An intelligent algorithm model calling system for unmanned systems, characterized in that: include: The module for obtaining the image to be detected is used to obtain the image to be detected based on the unmanned system; A target intelligent algorithm model group matching module is used to call a target intelligent algorithm model group that matches the image to be detected in a pre-established unmanned cluster target intelligent algorithm model group; A target intelligent algorithm model matching module, used for loading a target intelligent algorithm model matching the image to be detected into the target intelligent algorithm model group; A detection result acquisition module, used to detect and identify the image to be detected based on the target intelligent algorithm model to obtain the detection result of the image to be detected; Among them, the unmanned cluster target intelligent algorithm model group is established after format conversion of the initial intelligent algorithm model group based on a plurality of model training configuration files, training the training data set generated by the acquired training images, generating an initial intelligent algorithm model group, and combining the hardware information of each unmanned system in the unmanned cluster.

10. An intelligent algorithm model management system for unmanned clusters, characterized in that: include: Intelligent algorithm model cloud visualization management interface, intelligent algorithm model cloud background management service and unmanned system airborne software; The intelligent algorithm model cloud visual management interface is used to send data storage instructions, model training instructions, and model deployment instructions to the intelligent algorithm model cloud backend management service to achieve visual operations on data management, model training, and model deployment; it is also used to send model switching instructions to the drone-mounted software to achieve visual operations on model switching; The intelligent algorithm model cloud backend management service is used to generate a training data set for the to-be-trained images received from the unmanned system airborne software; It is also used to receive the model training instruction sent by the cloud-based visual management interface of the intelligent algorithm model, train the data set to be trained based on the model training configuration file, generate an initial intelligent algorithm model corresponding to each of the model training configuration files, and package and combine each of the model training configuration files with the corresponding initial intelligent algorithm model to generate an initial intelligent algorithm model group; It is also used to receive the data storage instruction sent by the cloud-based visualization management interface of the intelligent algorithm model, and store the generated initial intelligent algorithm model group; it is also used to receive the model deployment instruction sent by the cloud-based visualization management interface of the intelligent algorithm model, and based on the hardware information of each unmanned system in the unmanned cluster, obtain the initial intelligent algorithm model adapted to each unmanned system from the initial intelligent algorithm model group; generate a target intelligent algorithm model by format conversion for each adapted initial intelligent algorithm model, package the target intelligent algorithm model with the corresponding model training configuration file to form a target intelligent algorithm model group, and send the target intelligent algorithm model group to the onboard software of the unmanned system; The unmanned system airborne software is used to collect images to be trained and send the images to be trained to the cloud-based background management service of the intelligent algorithm model; it is also used to send the target intelligent algorithm model group to the airborne model database of the unmanned system airborne software, waiting for the unmanned system to call it; it is also used to receive the model switching instruction sent by the cloud-based visual management interface of the intelligent algorithm model, and obtain the image to be detected based on the unmanned system; call the target intelligent algorithm model group that matches the image to be detected from the pre-established unmanned cluster target intelligent algorithm model group in the airborne model database; load the target intelligent algorithm model group that matches the image to be detected from the target intelligent algorithm model group according to the intelligent algorithm model loading rule. algorithm model; based on the target intelligent algorithm model matching the image to be detected, the image to be detected is detected and identified to obtain the detection result of the image to be detected; it is also used to obtain the status monitoring result of the unmanned system according to the hardware status monitoring mechanism of the unmanned system if the evaluation result of the detection result of the image to be detected does not reach the preset threshold, and according to the evaluation effect and the status monitoring result, based on the intelligent algorithm model switching rule, switch to other target intelligent algorithm models of the target intelligent algorithm model group, and re-detect and identify the image to be detected according to the other target intelligent algorithm models switched to, until the detection result of the other target intelligent algorithm models switched to reaches the preset threshold.

11. An electronic device, characterized in that: include: at least one processor and memory; The memory and the processor are connected via a bus; The memory is used to store one or more programs; When the one or more programs are executed by the at least one processor, the intelligent algorithm model calling method of the unmanned system as described in any one of claims 1 to 8 is implemented.

12. A readable storage medium, characterized in that: An execution program is stored thereon, and when the execution program is executed, the intelligent algorithm model calling method of the unmanned system as described in any one of claims 1 to 8 is implemented.

Images