Radar intelligent model rapid upgrading platform based on incremental learning

Abstract

The invention provides a radar intelligent model rapid upgrading platform based on incremental learning, and the platform comprises the steps: directly collecting the real-time network data of a radar through a configuration method, and carrying out the analysis processing, and forming a standardized radar training sample; and for radar equipment, sample accumulation, incremental learning training of the model and pushing of the model are completed through periodic automatic scheduling, and intelligent rapid iteration upgrading of the radar is realized. Key technologies such as a real installation and incremental learning system data interaction bus, efficient model training and rapid model pushing are broken through, continuous iterative training of an intelligent algorithm model in the incremental learning system is supported, the intelligent algorithm performance is optimized, the algorithm after iterative optimization is seamlessly migrated to a real installation system, and the intelligent algorithm is rapidly pushed. Real-time automatic transplantation of an artificial intelligence algorithm in an embedded system is realized, so that the problem of long-term iteration of the artificial intelligence algorithm in real-time installation and use is solved.

Description

Technical Field

[0001] The present invention belongs to the field of software design, and in particular relates to a radar intelligent model rapid upgrade platform based on incremental learning. Background Art

[0002] Artificial intelligence algorithms face the problem of long-term updates and iterations during radar installation. Radar intelligent applications export data offline, train, and then perform installation updates. The entire cycle takes several months. This patent builds an intelligent incremental learning system through research on continuous iteration and rapid upgrade technologies for radar equipment. It is independent of the embedded platform for installation and receives installation data through the radar's DDS, multicast and other protocols. It continuously iterates and trains intelligent algorithm models, optimizes algorithm performance, and completes seamless migration of the iteratively optimized algorithm model to the installation system. Summary of the Invention

[0003] This invention proposes a rapid upgrade platform for radar intelligent models based on incremental learning. This platform directly collects and analyzes radar real-time network data through a configurable approach, generating standardized radar training samples. It then uses periodic automated scheduling to accumulate samples, train incremental learning models, and push models to radar equipment, enabling rapid iterative upgrades of radar intelligence. This breakthrough addresses key technologies such as the data exchange bus between the actual installation and incremental learning systems, efficient model training, and rapid model push. It supports the continuous iterative training of intelligent algorithm models within the incremental learning system, optimizes intelligent algorithm performance, and seamlessly migrates the iteratively optimized algorithms to the actual installation system, enabling real-time, automatic transplantation of artificial intelligence algorithms within embedded systems. This addresses the long-term iteration of artificial intelligence algorithms during actual installation.

[0004] The present invention provides a radar intelligent model rapid upgrade platform based on incremental learning, comprising:

[0005] The real-time data collection and governance module (1) completes incremental data collection through radar protocol access such as DDS and multicast, and then completes data feature matching and filtering of incremental data sets, selects compliant data samples for annotation, and generates incremental training samples for incremental learning of intelligent algorithm models.

[0006] Model training module (2): The model training module divides the newly added incremental training sample data in parallel, completes the rapid training of the algorithm by calling the intelligent algorithm script, generates the incremental learning model, and completes the storage of the model.

[0007] The model push module (3) pushes the trained model file to the radar embedded implementation system and completes the replacement of the intelligent model and the restart of the application. After the restart, the application reads the new model file to complete the rapid upgrade.

[0008] The method and steps for implementing the present invention include:

[0009] The real-time data acquisition module (1) provides a user-oriented data storage operation interface, supports users to configure parameters such as real-time data storage message parsing rules, data collection address settings, storage start and stop cycles, etc. After receiving the actual data, it parses and annotates to form standardized training samples. The model training module (2) automatically reads the training samples according to the set cycle and creates a model training task to complete the automatic training of the algorithm. During the process, the parameter convergence and change curve of the training process are displayed in real time. After the new model is formed, it is saved in the database and notified to the model push module (3). The model push module (3) obtains the latest model file from the database and pushes it to the actual radar, updates the model file and restarts the intelligent application to complete the entire process of incremental learning.

[0010] The beneficial effects of the present invention are

[0011] By building an AI incremental learning system independent of the actual installation system and leveraging an integrated data bus, the embedded installation system platform and incremental learning are managed in an integrated manner. Within the incremental learning platform, intelligent algorithm models are continuously iteratively trained to optimize their performance. Simultaneously, the optimized algorithms are seamlessly migrated to the installation system, addressing the long-term iteration requirements of AI algorithms during installation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 Module diagram of the incremental learning system for intelligent applications. DETAILED DESCRIPTION

[0013] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present invention.

[0014] like Figure 1 As shown, the present invention proposes a radar intelligent model rapid upgrade platform based on incremental learning, including:

[0015] The real-time data collection and governance module (1) collects incremental data through access, then completes data feature matching and filtering of incremental data sets, selects compliant data samples for annotation, and generates incremental training samples for incremental learning of intelligent algorithm models.

[0016] Model training module (2): The model training module divides the newly added incremental training sample data in parallel, completes the distributed accelerated training of the intelligent network model through multi-node parallel training, and generates the original incremental learning model. The model calculation amount is relatively complex.

[0017] The model push module (3) pushes the trained model file to the radar embedded system and restarts the intelligent application. After the restart, the application reads the new model file to complete the upgrade.

[0018] The implementation method and steps of this module are as follows:

[0019] The real-time data acquisition module (1) provides a user-oriented storage operation interface, supports users to configure parameters such as real-time data storage message parsing rules, data collection address settings, storage start and stop cycles, etc. After receiving the actual data, it parses and annotates it to form standardized training samples. The model training module (2) automatically reads the training samples according to the set cycle, and creates a model training task to complete the automatic training of the algorithm. After the new model is formed, it is saved in the database. The model push module (3) obtains the latest model file from the database and pushes it to the actual radar, updates the model file and restarts the intelligent application to complete the incremental learning of the entire process.

[0020] The present invention is not limited to the above specific embodiments, and various modifications and variations are possible. Any modification, equivalent replacement, improvement, etc. made to the above embodiments based on the technical essence of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A radar intelligent model rapid upgrade platform based on incremental learning, characterized by: include: Real-time data collection and management module, model training module and model push module; The real-time data collection and governance module is connected to the actual installation to complete the incremental data collection, complete the data feature matching and filtering of the incremental data set, select and annotate compliant data samples, and generate incremental training samples; The model training module divides the newly added incremental training sample data in parallel, completes the rapid training of the algorithm, generates the incremental learning model, and completes the storage of the model file; The model push module pushes the trained model file to the radar embedded implementation system and completes the replacement of the intelligent model and application restart. After the restart, the application reads the new model file to complete the rapid upgrade.

2. The radar intelligent model rapid upgrade platform based on incremental learning according to claim 1 is characterized by: The real-time data acquisition module provides a user-oriented data storage operation interface, supports users to configure real-time data storage message parsing rules, data collection address settings, storage start and stop cycles, and parses and annotates the actual data after receiving it to form standardized training samples; the model training module automatically reads the training samples according to the set cycle, and creates a model training task to complete the automatic training of the algorithm, and displays the parameter convergence and change curve of the training process in real time during the process. After forming a new model, it is saved in the database and notified to the model push module; the model push module (3) obtains the latest model file from the database and pushes it to the actual radar, updates the model file and restarts the intelligent application to complete the entire process of incremental learning.

Images