A multi-fidelity hyperparameter optimization method and system for maritime target recognition

Abstract

The application belongs to the technical field of artificial intelligence and marine visual recognition, and discloses a multi-fidelity hyperparameter optimization method and system for marine target recognition; wherein the multi-fidelity hyperparameter optimization method comprises: obtaining image data sets and hyperparameter configuration spaces corresponding to a marine target recognition model to be optimized; using a multi-fidelity hyperparameter optimization framework to optimize hyperparameters in the training process of the marine target recognition model to be optimized, and obtaining optimized hyperparameter configurations. The application specifically discloses a multi-fidelity hyperparameter optimization scheme combining trend guiding decision for a marine target recognition task, solves technical problems such as large consumption of computing resources, low evaluation efficiency and false elimination of potential configurations in the hyperparameter tuning process of existing marine visual recognition models, and improves the training efficiency and recognition accuracy of deep learning models in marine target recognition tasks.

Description

Technical Field

[0001] This invention belongs to the fields of artificial intelligence and marine visual recognition technology, and specifically relates to a multi-fidelity hyperparameter optimization method and system for marine target recognition. Background Technology

[0002] In key application areas such as marine surveillance and national defense security, especially in maritime target identification tasks, deep learning methods have been widely used to automatically identify targets such as ships, small boats, buoys, and tankers in remote sensing images, replacing manual monitoring and improving identification efficiency, accuracy, and consistency. To achieve optimal identification performance, it is usually necessary to finely adjust the hyperparameters of the deep learning model, such as learning rate, batch size, and weight decay. This parameter tuning process is computationally expensive and time-consuming, especially when processing high-resolution remote sensing images, complex neural network structures, or large-scale datasets.

[0003] Currently, hyperparameter optimization (HPO) technology automates model tuning by systematically searching for hyperparameter configurations, significantly reducing manual workload. HPO methods can be divided into single-fidelity and multi-fidelity methods. Traditional single-fidelity methods (such as random search and Bayesian optimization) typically evaluate configurations under the full training budget, which is difficult to implement in maritime target identification tasks due to high resource consumption and long model training time. Multi-fidelity methods, on the other hand, rapidly pre-evaluate hyperparameter configurations at lower fidelity, gradually eliminating poorly performing configurations, thereby saving computational resources and improving overall optimization efficiency. Representative multi-fidelity methods, such as the Successive Halving (SH) algorithm, allocate resources round by round, retaining well-performing configurations. Improved methods, such as Hyperband, ASHA, BOHB, and DEHB, further optimize the exploration-exploitation tradeoff and have shown good performance in some remote sensing vision tasks. However, these methods typically rely solely on evaluation metrics (such as validation accuracy) at the current fidelity level for configuration ranking and elimination, implicitly assuming that learning curves between configurations will not intersect. In actual maritime target recognition tasks, the performance of hyperparameter configurations in the early stages does not accurately reflect their final potential. Some hyperparameter configurations with potential for later advantages may be prematurely eliminated due to poor initial performance. This "false elimination" problem can result in suboptimal performance of the final selected maritime target recognition model, weakening the ability to identify key targets and adversely affecting subsequent maritime monitoring, situational awareness, and decision-making response processes. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-fidelity hyperparameter optimization method and system for maritime target recognition, to solve one or more of the aforementioned technical problems. Specifically, the technical solution of this invention is a multi-fidelity hyperparameter optimization scheme combining trend-guided decision-making for maritime target recognition tasks. It solves the technical problems of high computational resource consumption, low evaluation efficiency, and misclassification potential in existing marine visual recognition models during hyperparameter tuning, thereby improving the training efficiency and recognition accuracy of deep learning models in maritime target recognition tasks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a multi-fidelity hyperparameter optimization method for maritime target identification, comprising the following steps: Obtain the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; Based on the acquired image dataset and the hyperparameter configuration space, a multi-fidelity hyperparameter optimization framework is used to optimize the hyperparameters during the training process of the maritime target recognition model to be optimized, so as to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive the acquired hyperparameter configuration space and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy; The performance evaluation and early termination module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set according to a comprehensive evaluation index, and obtain the evaluation results. Among them, the module terminates the poorly performing hyperparameter configurations early by ranking them according to the comprehensive evaluation index. The comprehensive evaluation index combines the validation accuracy, the trend slope based on the observed learning curve, and the predicted value of the performance under the maximum budget based on the LC-PFN method. The fidelity budget allocation module is used to dynamically allocate computing resources to different hyperparameter configurations based on the evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized.

[0006] A further improvement of the technical solution of the present invention is that, in the step of obtaining the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized, the maritime target recognition model to be optimized is a deep learning model for maritime target recognition.

[0007] A further improvement of the technical solution of the present invention is that, in the step of obtaining the image dataset and hyperparameter configuration space corresponding to the marine target recognition model to be optimized, the image dataset is an RGB image or grayscale image collected during marine monitoring; the hyperparameter configuration space includes relevant hyperparameters of various preset types of marine target recognition models to be optimized.

[0008] A further improvement of the technical solution of the present invention is that, in the sampling and selection module, in the step of dynamically sampling or selecting hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy, the strategy for dynamically sampling or selecting hyperparameter configurations is one or more of random sampling, grid search, Bayesian optimization, genetic algorithm and particle swarm optimization algorithm.

[0009] A further improvement to the technical solution of the present invention lies in the comprehensive evaluation index. The calculation expression is: ; In the formula, To verify accuracy, it is used to represent the model's recognition accuracy on a validation set of maritime target images; This is the slope value calculated in the learning curve analysis; The predicted value obtained using the LC-PFN learning curve prediction method at the maximum budget B; , and These are the corresponding adaptive coefficients; To ensure fidelity; This represents the number of data points needed for learning curve prediction.

[0010] A further improvement of the technical solution of the present invention is that, in the fidelity budget allocation module, in the step of dynamically allocating computing resources to different hyperparameter configurations based on the evaluation results, the Hyperband algorithm is used to explore the hyperparameter configuration space through multiple rounds of parallel search. In each round, a continuous halving algorithm is used to gradually reduce the budget, and in each round, more resources are allocated to the better-performing configuration.

[0011] In a second aspect, the present invention provides a multi-fidelity hyperparameter optimization system for maritime target identification, comprising: The data acquisition module is used to acquire the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; The optimization module is used to optimize the hyperparameters in the training process of the maritime target recognition model to be optimized based on the acquired image dataset and the hyperparameter configuration space, using a multi-fidelity hyperparameter optimization framework, to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive the acquired hyperparameter configuration space and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy; The performance evaluation and early termination module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set according to a comprehensive evaluation index, and obtain the evaluation results. Among them, the module terminates the poorly performing hyperparameter configurations early by ranking them according to the comprehensive evaluation index. The comprehensive evaluation index combines the validation accuracy, the trend slope based on the observed learning curve, and the predicted value of the performance under the maximum budget based on the LC-PFN method. The fidelity budget allocation module is used to dynamically allocate computing resources to different hyperparameter configurations based on the evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized.

[0012] In a third aspect, the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of the first aspects of the present invention.

[0013] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of the first aspects of the present invention.

[0014] In a fifth aspect, the present invention provides a computer program product comprising a computer program or instructions which, when executed by a processor, implement the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of the first aspects of the present invention.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention specifically provides a multi-fidelity hyperparameter optimization method for maritime target recognition tasks, which combines trend-guided decision-making and aims to improve the training efficiency and final recognition performance of maritime target recognition models under limited computing resources. By introducing a comprehensive evaluation index (TLC index), this invention can more accurately evaluate the potential of each configuration without completing full training. This effectively avoids the problem of some configurations with long-term potential being prematurely eliminated due to poor initial performance, thus achieving a significant effect in improving the rationality of hyperparameter ranking and the stability of optimization results. Further, this invention improves the accuracy of judging the merits of configurations during multi-fidelity hyperparameter optimization by constructing a comprehensive evaluation index (TLC index) based on learning curve trend analysis and learning curve prediction, thereby enhancing overall optimization performance. The proposed TLC index integrates model evaluation performance at the current fidelity, trend change information of the learning curve, and prediction results of future performance. It can more accurately predict the performance of hyperparameter configurations at high fidelity without full-fidelity training. Compared with traditional methods that rank based solely on verification accuracy, the TLC index can effectively avoid misjudging and prematurely eliminating some hyperparameter configurations that initially perform poorly but have strong potential in maritime target identification, solving the problems of one-sided evaluation criteria and poor ranking correlation in existing multi-fidelity HPO methods.

[0016] The method proposed in this invention can be seamlessly integrated into existing multi-fidelity hyperparameter optimization frameworks (such as Hyperband, ASHA, etc.), and is applicable to various marine target recognition models (such as convolutional neural networks, residual networks, etc.) and marine vision tasks (such as marine target edge recognition, small boat recognition, foreign object recognition, etc.), and has good versatility, scalability and engineering practical value.

[0017] This invention proposes a multi-fidelity hyperparameter optimization method for maritime target recognition, which introduces a TLC evaluation index that integrates learning curve trend analysis and learning curve prediction, and provides a specific calculation expression. This index, even before the model has undergone complete training, provides a more accurate and stable ranking basis for the multi-fidelity optimization process by comprehensively considering current validation accuracy, learning curve trends, and future performance predictions, effectively improving the accuracy of evaluation and decision-making. Especially in maritime target recognition scenarios, where model training is costly and sample distribution is complex, traditional algorithms are prone to mistakenly eliminating potentially high-quality configurations due to inaccurate early evaluations. The TLC index significantly reduces this risk of "false elimination," enhances the optimization process's ability to identify high-quality configurations, and thus improves the model's final accuracy and robustness in target recognition tasks. Furthermore, the TLC index has good versatility and can be seamlessly integrated into existing multi-fidelity optimization algorithms based on the SH framework (such as Hyperband and ASHA) without modifying the core structure, demonstrating high engineering feasibility and industrial application value.

[0018] In summary, the novel solution disclosed in this invention can effectively alleviate the problems of inaccurate sorting and unreasonable configuration selection in the non-full evaluation stage of existing multi-fidelity HPO methods, and is particularly suitable for marine visual recognition tasks that are sensitive to computational resources and have high performance requirements. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating a multi-fidelity hyperparameter optimization method for maritime target identification provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of the multi-fidelity hyperparameter optimization framework in an embodiment of the present invention; Figure 3 This is a schematic diagram of the TLC index structure in an embodiment of the present invention; Figure 4 This is a schematic diagram of the hyperparameter optimization results in an embodiment of the present invention; Figure 5 This is a schematic diagram of a multi-fidelity hyperparameter optimization system for maritime target identification provided in an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0022] Based on the technical solutions disclosed in the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0023] Please see Figure 1 and Figure 2The present invention provides a multi-fidelity hyperparameter optimization method for maritime target identification, comprising the following steps: Step 1: Obtain the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; Step 2: Based on the image dataset and hyperparameter configuration space obtained in Step 1, the hyperparameters in the training process of the maritime target recognition model to be optimized are optimized using a multi-fidelity hyperparameter optimization framework to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; interpretably, subsequent hyperparameter optimization operations, including sampling and selection, are limited to the preset hyperparameter configuration space constructed above. The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the current optimization strategy for subsequent evaluation and comparison. The performance evaluation and early stopping module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set, and obtain the evaluation results; among them, the module terminates the poorly performing hyperparameter configurations early by combining comprehensive evaluation indicators. The fidelity budget allocation module is used to dynamically allocate computing resources (such as training time, number of samples, image resolution, etc.) to different hyperparameter configurations based on the current evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized.

[0024] Explained by example, the multi-fidelity hyperparameter optimization method of this invention is applicable to the automatic optimization process of model hyperparameters in maritime target identification tasks. The optimization object is a deep learning model (such as a convolutional neural network, residual network, etc.) used for maritime target identification. The image data used for model training are RGB images (Red-Green-Blue images) or grayscale images collected during marine monitoring. The images contain various small-sized maritime target features, such as small boats, ships, buoys, etc.

[0025] In the technical solution provided by the embodiments of the present invention, a multi-fidelity hyperparameter optimization framework is used to optimize the hyperparameters of the maritime target recognition model to be optimized. This framework does not simply rely on the performance in the early stage to determine the selection of hyperparameter configurations, but rather comprehensively considers the performance of hyperparameter configurations throughout the entire optimization process through a series of integrated modules and strategies, effectively avoiding the problem of "false elimination" caused by poor initial performance. In the multi-fidelity hyperparameter optimization framework provided in this invention, the sampling and selection module dynamically samples or selects hyperparameter configurations from a preset hyperparameter configuration space according to the current optimization strategy. This dynamic sampling and selection method means that certain hyperparameter configurations are not fixed based solely on limited early data and simple evaluations. Instead, as the optimization process progresses, it continuously explores and discovers hyperparameter combinations that may have later advantages, providing more opportunities for evaluation and optimization of hyperparameter configurations with potential. This solves the problem that early performance cannot accurately reflect final potential. Furthermore, the performance evaluation and early stopping module evaluates the recognition performance of different dynamically sampled or selected hyperparameter configurations on the training and validation sets. Unlike the traditional method of simply eliminating hyperparameter configurations based on initial performance, it combines a comprehensive evaluation index to terminate poorly performing hyperparameter configurations early. This comprehensive evaluation index considers multiple performance factors, not just a single index in the early stages, enabling a more comprehensive and accurate judgment of hyperparameter performance. The system prioritizes the quality of hyperparameter configurations. Configurations that initially perform poorly but possess potential for future development are not prematurely terminated but given further optimization opportunities, effectively preventing "false elimination." Furthermore, the fidelity budget allocation module dynamically allocates computing resources (such as training time, sample size, and image resolution) to different hyperparameter configurations based on current evaluation results. This dynamic allocation method rationally distributes resources according to the actual performance of each hyperparameter configuration, ensuring that promising configurations receive sufficient resources for optimization. For hyperparameter configurations that may have future advantages, more computing resources are allocated, allowing for deeper training and optimization, thereby improving their final performance. Ultimately, the technical solution of this invention outputs the final hyperparameter configuration obtained after a complete multi-fidelity optimization process, ensuring that the maritime target recognition model achieves a superior overall performance level and avoiding suboptimal model performance due to "false elimination." This improved model performance directly enhances the ability to identify key targets, providing more accurate and reliable information support for subsequent maritime monitoring, situational awareness, and decision-making response processes.

[0026] In summary, the technical solution of this invention, through the unique design and synergistic effect of each module in the multi-fidelity hyperparameter optimization framework, avoids the "false elimination" problem existing in existing technical solutions. This allows for the full optimization and development of hyperparameter configurations with later advantages. The resulting optimized hyperparameter configuration significantly improves the overall performance of the maritime target recognition model, enabling it to identify targets more accurately and efficiently in various complex sea conditions and target scenarios, thereby improving key indicators such as accuracy and recall. Furthermore, due to the improvement in the overall performance of the model, the ability to identify key maritime targets (such as ships, buoys, and submarines) is significantly enhanced, enabling more timely and accurate detection of key targets, reducing missed and false detections, and providing a more reliable data foundation for maritime monitoring.

[0027] As a specific embodiment of the technical solution of the present invention, the exploration space definition module is used to construct a unified search space, and the hyperparameter configuration space includes various types of hyperparameters related to maritime target recognition models, including but not limited to: Categorical hyperparameters (such as optimizer type, activation function type, normalization method, etc.); Discrete hyperparameters (such as batch size, number of convolutional layers, number of feature map channels, etc.); Continuous hyperparameters (such as initial learning rate, weight decay coefficient, momentum coefficient, etc.); Explained, the above modules ensure that the optimization process samples and selects only on legal configuration combinations, thereby meeting the accuracy and efficiency requirements of maritime target identification tasks.

[0028] As a specific embodiment of the technical solution of the present invention, the sampling and selection module supports a variety of hyperparameter sampling and selection strategies, including but not limited to: random sampling; grid search; Bayesian optimization; genetic algorithm; particle swarm optimization, etc. Explained, the above modules can generate new candidate configurations based on historical evaluation results and select from the legal configuration range; among them, the Bayesian optimization strategy predicts the performance of the new configuration by constructing a surrogate model, while the genetic algorithm and particle swarm optimization strategy simulate natural evolution to generate better configurations, so as to improve the search efficiency and accuracy in maritime target identification tasks.

[0029] As a specific embodiment of the technical solution of the present invention, the performance evaluation and early shutdown module includes: The validation accuracy is obtained, which represents the model's recognition accuracy on a validation set of maritime target images. The comprehensive evaluation metric TLC is calculated, which combines: the current validation accuracy; the trend slope based on the observed learning curve; and the predicted performance under the maximum budget based on the LC-PFN method (an efficient Bayesian learning curve extrapolation method based on Prior-Data Fitted Networks). Specifically, the formula for calculating the TLC index is as follows: ; In the formula, To verify accuracy, This is the slope value calculated in the learning curve analysis. This represents the predicted value obtained using the LC-PFN learning curve prediction method at the maximum budget B. , and These are the corresponding adaptive coefficients. To ensure fidelity, The number of data points needed in learning curve prediction; Explanatoryly, all hyperparameter configurations are ranked based on the TLC metric, and configurations with poor TLC metrics are terminated early to avoid mistakenly eliminating high-quality configurations with late-stage potential in marine applications.

[0030] As a specific embodiment of the technical solution of the present invention, the fidelity budget allocation module is used to dynamically allocate resources (such as computation budget, experimental time, etc.) to different hyperparameter configurations based on the historical evaluation results of the hyperparameter configuration. This module adopts the Hyperband algorithm, which internally uses a continuous halving algorithm to gradually allocate resources.

[0031] Specifically, the Hyperband algorithm explores the hyperparameter configuration space through multiple rounds of parallel search. In each round, a successive halving algorithm is used to gradually reduce the budget, allocating more resources to the better-performing configurations in each round. Hyperband first performs initial resource allocation on multiple candidate configurations, and then, through an inner loop's successive halving strategy, concentrates the budget on the best-performing configurations in subsequent rounds, gradually eliminating those poorly performing configurations.

[0032] Finally, the configuration selection module can receive the final optimized hyperparameter configuration and output it as the optimal recommendation result for deployment and operation of the maritime target recognition model, so as to achieve accurate identification and efficient classification of target types.

[0033] Please see Figures 1 to 3 In this embodiment of the invention, the optimization process of the above method may specifically include the following steps: Step 1: Input the hyperparameter configuration space of the maritime target recognition model to be optimized into the exploration space definition module to construct a legal search range; at the same time, input the corresponding maritime target image training set and validation set into the performance evaluation and early stopping module, and obtain candidate hyperparameter configurations in conjunction with the sampling and selection module and perform performance evaluation, and perform early stopping processing on inefficient configurations in combination with the TLC index; Step 2: The fidelity budget allocation module dynamically allocates computing resources based on the historical evaluation results of each configuration, and iteratively selects high-quality configurations using a continuous halving strategy; the final configuration selection module outputs the optimal hyperparameters to guide the deployment and practical application of the maritime target recognition model. like Figure 3 As shown, the TLC metric is a comprehensive performance evaluation metric that combines validation accuracy, learning curve trend analysis, and learning curve prediction results. The metric consists of three parts: the original validation accuracy, the slope based on learning curve trend analysis, and the predicted value based on learning curve prediction. By integrating the model performance at the current fidelity level with the estimation of its future performance trend, the metric can more accurately evaluate the merits of hyperparameter configuration. In a specific embodiment of the present invention, the trend analysis of the learning curve is performed using the least squares method on the last observed... n The data points are fitted together; the prediction of the learning curve is achieved using the LC-PFN method.

[0034] In a specific embodiment of this invention, the method is applied to a maritime target identification task, using a subset of the SeaShips dataset. A total of 3000 images are selected through uniform sampling from the original SeaShips dataset, ensuring coverage of six types of ship targets (ore carriers, bulk carriers, container ships, general cargo ships, fishing vessels, and passenger ships), with a balanced number of images per category (500 images). To standardize the model input size, all images are scaled to a uniform resolution of 640×640 pixels. To facilitate model training and performance evaluation, the dataset is divided into a training set and a validation set. 80% (400 images) of each category are selected as the training set, and the remaining 20% ​​(100 images) are used as the validation set.

[0035] The hyperparameter search space of the ResNet18 model is defined as follows: batch_size: The value range is [8, 64]; momentum: The value range is [0, 0.99]; weight_decay: The value range is [1e-5, 1e-3]; learning_rate: The value range is [1e-4, 0.01].

[0036] In this embodiment of the invention, the training budget is measured in epochs, with a maximum budget set at 27 epochs, representing the highest fidelity in multi-fidelity optimization. The fidelity budget allocation employs the Hyperband algorithm, which uses a continuous halving strategy internally. The halving factor η is set to 3, meaning that the budget is divided into 3 parts in each iteration, and the worst-performing 2 / 3 configurations are eliminated.

[0037] The operating environment is a computer system equipped with deep learning frameworks such as PyTorch, featuring a Gold 6626R @2.90GHz CPU, 32GB of memory, and an NVIDIA GeForce RTX 3090 GPU, capable of supporting the reading of maritime target images, model building, training, and optimization. In this embodiment, the running time is approximately 1800 seconds.

[0038] The specific implementation steps include: first, inputting the defined hyperparameter configuration space of the ResNet18 model into the exploration space definition module for sampling or selecting configurations during hyperparameter optimization; then, iteratively sampling and selecting these configurations based on the Hyperband multi-fidelity optimization algorithm; calculating the TLC index for each configuration, which, combined with validation accuracy, learning curve trend, and prediction results, is used to evaluate the configuration's merits and determine whether to stop early; the fidelity budget allocation module allocates more budget to potential configurations based on the TLC evaluation results, and iterates continuously until the preset maximum budget is reached; finally, the configuration selection module outputs the hyperparameter combination with the optimal optimization performance. This embodiment compares the optimization performance of the original Hyperband method and this method in maritime target recognition tasks. Please refer to... Figure 4 ,based on Figure 4 As shown in the illustrations, the technical solution of this invention outperforms traditional methods in both evaluation efficiency and final accuracy. In summary, this invention discloses a multi-fidelity hyperparameter optimization method for maritime target identification. This method, by introducing trend analysis and prediction mechanisms for learning curves, provides more accurate evaluation information even when the model has not completed full training, thus improving the rationality and reliability of hyperparameter ranking. The proposed TLC index significantly improves ranking accuracy, and the algorithm framework has strong compatibility, allowing for wide integration into various multi-fidelity optimization algorithms, demonstrating high applicability and general scalability at sea.

[0039] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0040] Please see Figure 5 In this embodiment of the invention, a multi-fidelity hyperparameter optimization system for maritime target identification is provided, comprising: The data acquisition module is used to acquire the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; The optimization module is used to optimize the hyperparameters in the training process of the maritime target recognition model to be optimized based on the acquired image dataset and the hyperparameter configuration space, using a multi-fidelity hyperparameter optimization framework, to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive the acquired hyperparameter configuration space and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy; The performance evaluation and early termination module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set according to a comprehensive evaluation index, and obtain the evaluation results. Among them, the module terminates the poorly performing hyperparameter configurations early by ranking them according to the comprehensive evaluation index. The comprehensive evaluation index combines the validation accuracy, the trend slope based on the observed learning curve, and the predicted value of the performance under the maximum budget based on the LC-PFN method. The fidelity budget allocation module is used to dynamically allocate computing resources to different hyperparameter configurations based on the evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized.

[0041] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or 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 and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to execute the operation of a multi-fidelity hyperparameter optimization method for maritime target identification.

[0042] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the multi-fidelity hyperparameter optimization method for maritime target identification in the above embodiments.

[0043] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0044] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0045] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0046] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0047] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A multi-fidelity hyperparameter optimization method for maritime target identification, characterized in that, Includes the following steps: Obtain the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; Based on the acquired image dataset and the hyperparameter configuration space, a multi-fidelity hyperparameter optimization framework is used to optimize the hyperparameters during the training process of the maritime target recognition model to be optimized, so as to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive the acquired hyperparameter configuration space and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy; The performance evaluation and early termination module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set according to a comprehensive evaluation index, and obtain the evaluation results. Among them, the module terminates the poorly performing hyperparameter configurations early by ranking them according to the comprehensive evaluation index. The comprehensive evaluation index combines the validation accuracy, the trend slope based on the observed learning curve, and the predicted value of the performance under the maximum budget based on the LC-PFN method. The fidelity budget allocation module is used to dynamically allocate computing resources to different hyperparameter configurations based on the evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized. The comprehensive evaluation indicators The calculation expression is: ； In the formula, To verify accuracy, it is used to represent the model's recognition accuracy on a validation set of maritime target images; This is the slope value calculated in the learning curve analysis; The predicted value obtained using the LC-PFN learning curve prediction method at the maximum budget B; , and These are the corresponding adaptive coefficients; To ensure fidelity; This represents the number of data points needed in learning curve prediction.

2. The multi-fidelity hyperparameter optimization method for maritime target identification according to claim 1, characterized in that, In the step of obtaining the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized, the maritime target recognition model to be optimized is a deep learning model for maritime target recognition.

3. The multi-fidelity hyperparameter optimization method for maritime target identification according to claim 1, characterized in that, In the step of obtaining the image dataset and hyperparameter configuration space corresponding to the marine target recognition model to be optimized, the image dataset is an RGB image or grayscale image collected during marine monitoring; the hyperparameter configuration space includes relevant hyperparameters of various preset types of marine target recognition models to be optimized.

4. The multi-fidelity hyperparameter optimization method for maritime target identification according to claim 1, characterized in that, In the sampling and selection module, the step of dynamically sampling or selecting hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy is one or more of random sampling, grid search, Bayesian optimization, genetic algorithm and particle swarm optimization algorithm.

5. The multi-fidelity hyperparameter optimization method for maritime target identification according to claim 1, characterized in that, In the fidelity budget allocation module, the step of dynamically allocating computing resources to different hyperparameter configurations based on the evaluation results uses the Hyperband algorithm to explore the hyperparameter configuration space through multiple rounds of parallel search. In each round, a continuous halving algorithm is used to gradually reduce the budget, and more resources are allocated to the better-performing configurations in each round.

6. A multi-fidelity hyperparameter optimization system for maritime target identification, characterized in that, include: The data acquisition module is used to acquire the image dataset and hyperparameter configuration space corresponding to the maritime target recognition model to be optimized; The optimization module is used to optimize the hyperparameters in the training process of the maritime target recognition model to be optimized based on the acquired image dataset and the hyperparameter configuration space, using a multi-fidelity hyperparameter optimization framework, to obtain the optimized hyperparameter configuration. The multi-fidelity hyperparameter optimization framework includes: Explore the space definition module, which is used to receive the acquired hyperparameter configuration space and construct the preset hyperparameter configuration space of the maritime target recognition model to be optimized; The sampling and selection module is used to dynamically sample or select hyperparameter configurations from the preset hyperparameter configuration space according to the optimization strategy; The performance evaluation and early termination module is used to evaluate the recognition performance of different hyperparameter configurations with dynamic sampling or selection on the training set and validation set according to a comprehensive evaluation index, and obtain the evaluation results. Among them, the module terminates the poorly performing hyperparameter configurations early by ranking them according to the comprehensive evaluation index. The comprehensive evaluation index combines the validation accuracy, the trend slope based on the observed learning curve, and the predicted value of the performance under the maximum budget based on the LC-PFN method. The fidelity budget allocation module is used to dynamically allocate computing resources to different hyperparameter configurations based on the evaluation results, and output the final hyperparameter configuration obtained after a complete multi-fidelity optimization process for deployment of the maritime target recognition model to be optimized. The comprehensive evaluation indicators The calculation expression is: ； In the formula, To verify accuracy, it is used to represent the model's recognition accuracy on a validation set of maritime target images; This is the slope value calculated in the learning curve analysis; The predicted value obtained using the LC-PFN learning curve prediction method at the maximum budget B; , and These are the corresponding adaptive coefficients; To ensure fidelity; This represents the number of data points needed in learning curve prediction.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of claims 1 to 5.

9. A computer program product, comprising a computer program or instructions, characterized in that, When a computer program or instruction is executed by a processor, it implements the multi-fidelity hyperparameter optimization method for maritime target identification as described in any one of claims 1 to 5.

Images