Industrial vision adaptive illumination compensation system and method based on Bayesian optimization

Abstract

The invention relates to an industrial vision adaptive illumination compensation system and method based on Bayesian optimization, and the system comprises an environment perception module which is used for collecting the multi-dimensional environment state data of an industrial site in real time, and collecting an original image; the intelligent preprocessing module is used for enhancing the original image according to the parameter combination output by the Bayesian optimization module to generate an enhanced image; the model reasoning module is used for identifying and detecting the enhanced image through a bimodal algorithm to generate a detection result; the Bayesian optimization module is used for performing objective function calculation to generate an optimization objective score, performing iterative optimization according to the optimization objective score and a historical observation data set, generating a candidate parameter combination, and generating a global optimal parameter and a configuration template based on the candidate parameter combination and a current environment feature; and the industrial interface module is used for transmitting the global optimal parameter, the detection result and the multi-dimensional environment state data to an industrial control system. According to the invention, the efficiency and adaptability of illumination compensation are improved, and the data calculation amount is reduced.

Description

Technical Field

[0001] The present application relates to the field of intelligent manufacturing technology, and in particular to an industrial vision adaptive lighting compensation system and method based on Bayesian optimization. Background Art

[0002] Amid the rapid development of intelligent manufacturing technology, industrial visual inspection systems, as a key component of modern automated production lines, are facing unprecedented demand for technological upgrades. Taking the electronics manufacturing sector as an example, with the continuous reduction in the size of electronic components and the continuous increase in assembly density, traditional quality control methods relying on manual visual inspection can no longer meet the industry's dual demands for inspection accuracy and efficiency. Deep learning-based intelligent vision technology, by constructing multi-layer neural network models, enables the precise identification and classification of minute defects, providing a reliable quality assurance tool for modern industrial production. However, when these inspection systems, which perform well in laboratory environments, are deployed in actual industrial sites, complex and variable environmental factors (such as light fluctuations and mechanical vibrations) often lead to a significant decline in system performance, with lighting conditions having a particularly significant impact. Light interference in industrial sites presents multi-faceted technical challenges. First, the mix of natural and artificial light in production environments leads to dynamic fluctuations in light intensity and color temperature. This unstable lighting condition directly alters the appearance of surface features on inspected objects. Second, due to differences in equipment installation position and angle, imaging systems may experience varying degrees of geometric distortion, including lens distortion and object scale distortion. Third, the overlapping shadows and reflection interference in multi-light environments create complex interference patterns. These factors combine to cause laboratory-trained vision models to frequently experience feature extraction bias, reduced recognition confidence, and even classification errors in real-world applications. This dual negative impact (feature distortion and decision errors) severely tests the reliability of industrial vision systems.

[0003] The current mainstream technical solutions for addressing lighting challenges in the industry face significant technical bottlenecks. Traditional manual parameter tuning methods not only rely on repeated debugging by experienced technicians, but also make it difficult to establish a quantifiable correlation model between lighting conditions and imaging quality. Although existing adaptive enhancement algorithms can achieve certain results in specific scenarios, they generally adopt independent parameter optimization strategies and cannot effectively solve the problem of interactive effects between multiple parameters. Especially in complex multi-parameter scenarios that require the simultaneous optimization of optical parameters, geometric correction parameters, and spatial filtering parameters, conventional grid search or random search methods have difficulty finding the global optimal solution under limited computational resources, and consume huge computing resources. Summary of the Invention

[0004] The purpose of the embodiments of the present application is to propose an industrial vision adaptive lighting compensation system and method based on Bayesian optimization to improve the efficiency of lighting compensation and adaptability in complex environments and reduce the amount of data calculation.

[0005] In order to solve the above technical problems, the embodiment of the present application provides an industrial vision adaptive illumination compensation system based on Bayesian optimization, including: The environmental perception module is used to collect multi-dimensional environmental status data of the industrial site in real time and collect original images; The intelligent preprocessing module is used to perform optical compensation, geometric correction and spatial filtering on the original image in sequence according to the parameter combination output by the Bayesian optimization module to generate an enhanced image; The model inference module is used to identify and detect the enhanced image using a bimodal algorithm to generate a detection result; The Bayesian optimization module is configured to calculate an objective function based on the batch of enhanced images and the corresponding detection results to generate an optimization target score, perform iterative optimization based on the optimization target score and a historical observation dataset to generate candidate parameter combinations, and generate global optimal parameters and configuration templates based on the candidate parameter combinations and current environmental characteristics; The industrial interface module is used to transmit the global optimal parameters, the detection results and the multi-dimensional environmental status data to the industrial control system through protocol conversion and data encapsulation technology.

[0006] In order to solve the above technical problems, the embodiment of the present application provides an industrial vision adaptive illumination compensation method based on Bayesian optimization, including: Real-time collection of multi-dimensional environmental status data of industrial sites and original images; determining a parameter combination based on the multi-dimensional environmental state data, and performing optical compensation, geometric correction, and spatial filtering on the original image in sequence according to the parameter combination to generate an enhanced image; Performing recognition and detection on the enhanced image using a dual-modal algorithm to generate a detection result; performing objective function calculation based on the batch of enhanced images and the corresponding detection results to generate an optimization target score, performing iterative optimization based on the optimization target score and a historical observation dataset to generate candidate parameter combinations, and generating a global optimal parameter and configuration template based on the candidate parameter combinations and current environmental characteristics; The global optimal parameters, the detection results and the multi-dimensional environmental status data are transmitted to the industrial control system through protocol conversion and data encapsulation technology.

[0007] An embodiment of the present invention provides an industrial vision adaptive illumination compensation system and method based on Bayesian optimization. The system includes an environmental perception module, an intelligent preprocessing module, a model inference module, a Bayesian optimization module, and an industrial interface module. The environmental perception module is used to collect multi-dimensional environmental status data of the industrial site in real time, as well as to collect original images. The intelligent preprocessing module is used to sequentially perform optical compensation, geometric correction, and spatial filtering on the original images based on the parameter combination output by the Bayesian optimization module to generate an enhanced image. The model inference module is used to identify and detect the enhanced images using a bimodal algorithm to generate detection results. The Bayesian optimization module is used to calculate an objective function based on a batch of enhanced images and the corresponding detection results to generate an optimization target score, and iteratively optimize based on the optimization target score and a historical observation data set to generate candidate parameter combinations. Furthermore, the industrial interface module is used to transmit the global optimal parameters, the detection results, and the multi-dimensional environmental status data to an industrial control system through protocol conversion and data encapsulation technology. The embodiment of the present invention effectively enhances the image by performing optical compensation, geometric correction and spatial filtering on the original image in sequence. At the same time, the enhanced image is recognized and detected through a dual-modal algorithm, which is conducive to detecting dynamic and static features in the image. In addition, the iterative optimization of the model through the Bayesian optimization module realizes the configuration of global optimal parameters, which is conducive to improving the efficiency of illumination compensation and adaptability in complex environments, and reducing the amount of data calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] In order to more clearly illustrate the solutions in this application, a brief introduction will be given below to the drawings required for use in the description of the embodiments of this application. Obviously, the drawings described below are some embodiments of this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0009] Figure 1 Schematic diagram of an industrial vision adaptive illumination compensation system based on Bayesian optimization provided in an embodiment of the present application; Figure 2 This is a schematic diagram of the structure of the environment perception module provided in an embodiment of the present application; Figure 3 Schematic diagram of the structure of the intelligent preprocessing module provided in the embodiment of the present application; Figure 4 This is a schematic diagram of the structure of the model reasoning module provided in the embodiment of the present application; Figure 5 Schematic diagram of the structure of the Bayesian optimization module provided in the embodiment of the present application; Figure 6This is a flowchart of the implementation process of the industrial vision adaptive illumination compensation method based on Bayesian optimization provided in an embodiment of the present application; Figure 7 This is a flowchart of a sub-process in the industrial vision adaptive illumination compensation method based on Bayesian optimization provided in an embodiment of the present application; Figure 8 This is another implementation flowchart of the sub-process in the industrial vision adaptive illumination compensation method based on Bayesian optimization provided in an embodiment of the present application. DETAILED DESCRIPTION

[0010] Unless otherwise defined, all technical and scientific terms used herein have the same meanings as commonly understood by those skilled in the art to which this application belongs. The terms used in the specification of the application are for the purpose of describing specific embodiments only and are not intended to limit this application. The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of this application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0011] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0012] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0013] The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0014] Please refer to Figure 1 , this application provides an embodiment of an industrial vision adaptive illumination compensation system based on Bayesian optimization. Figure 1 As shown, the industrial vision adaptive illumination compensation system based on Bayesian optimization of this embodiment includes: an environment perception module 10, an intelligent preprocessing module 20, a model reasoning module 30, a Bayesian optimization module 40 and an industrial interface module 50, wherein: The environmental perception module 10 is used to collect multi-dimensional environmental status data of the industrial site in real time, and to collect original images; the intelligent preprocessing module 20 is used to perform optical compensation, geometric correction and spatial filtering on the original image in sequence according to the parameter combination output by the Bayesian optimization module 40 to generate an enhanced image; the model reasoning module 30 is used to identify and detect the enhanced image through a bimodal algorithm to generate a detection result; the Bayesian optimization module 40 is used to calculate the objective function based on the batch of enhanced images and the corresponding detection results to generate an optimization target score, and to perform iterative optimization based on the optimization target score and the historical observation data set to generate candidate parameter combinations, and to generate global optimal parameters and configuration templates based on the candidate parameter combinations and current environmental characteristics; the industrial interface module 50 is used to transmit the global optimal parameters, the detection results and multi-dimensional environmental status data to the industrial control system through protocol conversion and data encapsulation technology.

[0015] Because traditional methods consume significant computing resources and lack adaptability in complex environments, this application proposes an adaptive illumination compensation system for industrial vision based on Bayesian optimization. By building a dynamically adaptive Bayesian optimization framework, this system achieves a breakthrough improvement in the model's ability to rapidly adapt to environmental changes at significantly reduced hardware costs. This system decouples environmental adaptability from resource consumption control through an intelligent optimization algorithm. This allows the industrial vision system to maintain detection accuracy while significantly reducing computing resource requirements. The system can also automatically adapt to typical industrial scenarios such as fluctuating light intensity and mechanical vibration interference. In actual industrial applications, this system demonstrates excellent environmental adaptability. Tests under different lighting conditions demonstrate that the system can automatically learn optimal image enhancement strategies, significantly improving the generalization performance of the detection model. Furthermore, the parameter configuration schemes generated by the system are highly portable, enabling rapid deployment across different production lines and significantly shortening equipment commissioning cycles. Furthermore, by building a complete adaptive optimization chain, this system provides an intelligent environmental adaptation solution for industrial vision inspection systems, effectively overcoming the challenge of model generalization under complex lighting conditions and providing key technical support for improving the reliability of industrial vision systems.

[0016] Specifically, this system employs a layered modular design, building a complete "perception-processing-decision-execution" closed-loop system specifically designed to address the deployment and adaptation of mainstream deep learning object detection models in industrial inspection scenarios. The environmental perception module 10 primarily collects multi-dimensional environmental status data from industrial sites in real time, providing a decision basis for the dynamic adaptive adjustment of subsequent image preprocessing algorithms. By continuously monitoring key parameters of the production environment, such as light intensity distribution, color temperature characteristics, and equipment vibration frequency, the environmental perception module 10 constructs an environment-image quality correlation model, ensuring that the visual inspection system maintains detection accuracy and stability in complex industrial scenarios. The intelligent preprocessing module 20 primarily performs adaptive enhancement processing on raw images captured by industrial cameras. Its core function is to eliminate the impact of environmental factors on image quality through multimodal parameterization, thereby ensuring that the image feature distribution input to the model is highly consistent with the training data. This intelligent preprocessing module 20 utilizes a layered processing architecture, employing a three-stage process flow consisting of optical compensation, geometric correction, and spatial filtering, to systematically address common industrial field issues such as uneven lighting, perspective deviation, and noise interference. As the core execution unit of the intelligent detection system for retractable triggers, model inference module 30 is specifically responsible for the dual tasks of dynamic target tracking and static scene analysis in industrial sites. This module achieves efficient decoupling of detection tasks through a dual-modal algorithm architecture. Its full detection efficiency is highly dependent on the high-quality standardized image data (i.e., the output enhanced image) output by the intelligent preprocessing module 20. The Bayesian optimization module 40 is used to search for the optimal preprocessing parameter combination, maximize the image enhancement effect through an intelligent optimization algorithm, and establish a quantitative relationship model between preprocessing parameters and model detection performance. It also dynamically adjusts the parameter search strategy to balance the exploration of new parameter combinations with the development of known high-quality areas. The Bayesian optimization module 40 can also achieve adaptive alignment of image feature distributions with training data in complex industrial environments, reducing the cost of manual parameter adjustment and improving the system's environmental adaptability.

[0017] The original image data is RGB or grayscale images captured by industrial cameras, which may be affected by uneven lighting, shadows, and color casts. The optimization parameters in this application include multidimensional vectors such as illumination compensation parameters (brightness, contrast, and color temperature), geometric transformation parameters (rotation and scaling), and spatial processing parameters (blur). Bayesian optimization is an intelligent parameter search method that predicts the effects of different parameter combinations by building a probabilistic model and gradually approaches the optimal solution. Compared to traditional trial-and-error methods, Bayesian optimization can find the optimal parameters with fewer attempts, achieving the core goal of "improving environmental adaptability at a low cost." Industrial visual inspection is an automated inspection technology based on computer vision. Industrial cameras capture target images (such as PCBs, robotic arms, and retractable board machines) and use image processing algorithms to identify their state, detect defects, or classify them. This system is used for real-time monitoring of object status on the production line, and its accuracy is significantly affected by lighting conditions. Illumination compensation is an image enhancement technology that addresses varying lighting conditions in industrial environments. By adjusting optical parameters such as brightness, contrast, and color temperature, the distribution of captured images is aligned with the training data. This system uses multi-parameter joint optimization (such as light intensity, color shift, blur, etc.) to improve the inference stability of the model under different lighting conditions.

[0018] See also Figure 2 , Figure 2 The schematic diagram of the structure of the environment perception module 10 is shown. The environment perception module 10 includes an optical environment monitoring unit 101, an imaging device state monitoring unit 102, a mechanical environment monitoring unit 103 and an image acquisition unit 104, wherein: The optical environment monitoring unit 101 is used to collect the lighting data and color characteristic data of the industrial site operation area through the built-in sensor of the industrial camera to generate a multi-dimensional optical parameter set; the imaging device status monitoring unit 102 is used to dynamically obtain the underlying imaging parameters of the camera by continuously tracking the imaging parameter status of the camera; the mechanical environment monitoring unit 103 is used to detect the mechanical vibration data of the equipment during operation through a vibration sensing device; the image acquisition unit 104 is used to collect the original image through the industrial camera.

[0019] Specifically, the optical environment monitoring unit 101 uses the built-in sensors of industrial cameras to collect real-time data on the lighting conditions (e.g., ambient light intensity distribution) and color characteristics (e.g., color temperature) of the work area, generating a multidimensional optical parameter set. The imaging device status monitoring unit 102 continuously tracks the camera's imaging parameter status to dynamically acquire underlying imaging parameters such as exposure time, analog gain, and digital gain. The mechanical environment monitoring unit 103 uses a vibration sensor to detect mechanical vibration data during equipment operation and quantitatively assess the impact of vibration on imaging stability.

[0020] The embodiments of the present application can adopt a high-frequency sampling mechanism to ensure millisecond-level response capabilities to changes in environmental conditions and achieve high real-time detection; integrate monitoring data from different sensors through intelligent algorithms, construct a quantitative correlation model between environmental factors and image quality characteristics, and realize multi-source data fusion; dynamically adjust monitoring strategies according to environmental changes, optimize the pertinence and effectiveness of data collection, and realize adaptive perception.

[0021] See also Figure 3 , Figure 3 FIG. 2 shows a schematic diagram of the structure of the intelligent pre-processing module 20, which includes a data acquisition unit 201, an optical compensation unit 202, a geometric correction unit 203, and a spatial filtering unit 204, wherein: The data acquisition unit 201 is used to acquire the original image output by the environmental perception module 10, and to acquire the parameter combination generated by the Bayesian optimization module 40 based on the multi-dimensional environmental state data; the optical compensation unit 202 is used to adjust the illumination distribution data of the original image according to the brightness and color temperature parameters in the parameter combination using a preset compensation algorithm to generate a first enhanced image, wherein the preset compensation algorithm includes a gamma correction algorithm, a white balance algorithm and a histogram equalization algorithm; the geometric correction unit 203 is used to correct the first enhanced image according to the underlying imaging parameters in the parameter combination using a preset correction algorithm to generate a second enhanced image; the spatial filtering unit 204 is used to perform spatial filtering processing on the second enhanced image using an adaptive bilateral filter to generate the enhanced image.

[0022] Specifically, the Bayesian optimization module 40 generates parameter combinations based on multi-dimensional environmental state data and historical observation data. Based on this parameter combination, the intelligent preprocessing module 20 performs a three-stage processing process on the original image: optical compensation, geometric correction, and spatial filtering. This systematically addresses common industrial field issues such as uneven lighting, perspective deviation, and noise interference. Table 1 below details the technical implementation solutions, supporting acceleration strategies, and performance indicators for each enhancement unit: Table 1

[0023] In this embodiment of the present application, the optical compensation unit 202 performs optical compensation on the original image, reshaping the illumination intensity distribution through gamma correction and eliminating ambient light color casts by combining a color temperature compensation matrix. The geometric correction unit 203 uses high-precision camera calibration technology to correct lens distortion and implements perspective transformation by combining feature point matching. The spatial filtering unit 204 employs an adaptive bilateral filter to suppress noise while preserving edge information.

[0024] This embodiment of the application utilizes a parameterized processing pipeline design that supports dynamic combination and adjustment of multiple core parameters. GPU hardware acceleration is also used to achieve efficient parallel computing, ensuring the real-time performance of industrial inspections. The intelligent preprocessing module 20 is deeply integrated with the Bayesian optimization system to form a closed-loop optimization mechanism. This application is environmentally adaptive and can automatically adjust processing strategies based on field conditions.

[0025] See also Figure 4 , Figure 4 FIG. 3 shows a schematic diagram of the structure of the model reasoning module 30 , which includes a dynamic recognition unit 301 and a static recognition unit 302 , wherein: The dynamic recognition unit 301 is used to identify and detect the preprocessed regular image based on a dynamic recognition algorithm using a lightweight neural network architecture to output the spatial position coordinates and the confidence level of the key points of the robotic arm; the static recognition unit 302 is used to perform state judgment on the preset area in the preprocessed uniform illumination image through a multi-scale feature analysis network, generate the category label, and perform semantic analysis on the preprocessed uniform illumination image to identify the abnormal pattern.

[0026] Specifically, the dual-modal algorithm comprises a dynamic recognition algorithm and a static recognition algorithm. The enhanced image consists of a preprocessed regularized image and a preprocessed uniformly illuminated image. Detection results include spatial position coordinates, confidence scores, category labels, and anomaly patterns. The dynamic recognition algorithm, based on the preprocessed regularized image, employs a lightweight network architecture to enable real-time state monitoring and trajectory analysis of moving components, such as robotic arms. The elimination of geometric distortion and motion blur during preprocessing significantly improves the algorithm's tracking stability and positioning accuracy, enabling it to accurately capture the motion characteristics of fast-moving targets. The static recognition algorithm, on the other hand, utilizes a multi-scale feature analysis network to simultaneously perform two key tasks: accurately identifying the state of predefined key areas and semantically analyzing the entire original image to identify potential anomaly patterns. The elimination of illumination interference and noise during preprocessing significantly enhances the algorithm's ability to perceive subtle features. The two algorithms work together to build a complete detection and cognitive system, ultimately outputting structured data containing detection results and confidence scores, providing critical data support for predictive maintenance, fault warnings, and operational status monitoring of the robotic arm.

[0027] In an embodiment of the present application, the dynamic recognition unit 301 adopts a lightweight neural network architecture and develops a high-precision state recognition algorithm for moving parts such as the trigger retractor robot arm. This high-precision state recognition algorithm directly outputs the spatial position coordinates and detection confidence of the robot arm's key points through an innovative three-dimensional coordinate regression network, and combines with the temporal motion analysis module to achieve continuous tracking of the motion trajectory. The present application provides a suppressive adaptive target matching mechanism, which significantly improves the detection accuracy of fast-moving small-sized parts by dynamically adjusting the size and proportion of the detection frame, ensuring stable positioning performance in complex industrial environments. The static recognition unit 302 adopts a multi-level feature analysis architecture to achieve dual recognition tasks in the trigger retractor working scene: on the one hand, accurate state judgment of predefined key areas, and on the other hand, semantic analysis of the overall image content to identify potential abnormalities. Through the adaptive receptive field adjustment mechanism, it can focus on local key details while maintaining global context perception capabilities. The embodiment of the present application integrates region-specific feature extraction and global semantic analysis modules, and realizes interactive verification of local and global information through a cross-attention mechanism. By adopting a hierarchical feature distillation strategy, while ensuring the recognition accuracy of key areas, the model's detection sensitivity to unknown abnormal patterns is enhanced, realizing comprehensive intelligent monitoring of the working status of the retractable trigger.

[0028] The embodiment of the present application realizes the simultaneous completion of motion target tracking and scene semantic analysis through the parallel computing architecture of dynamic recognition algorithm and static recognition algorithm. The embodiment of the present application forms a closed-loop data flow with the intelligent preprocessing module 20, making full use of the standardized image features after preprocessing, so that the convergence speed of the detection algorithm is significantly improved. Based on the environmental parameter label output by the preprocessing module, the detection threshold is dynamically adjusted so that the system maintains stable detection performance under different lighting conditions. Through the intelligent threshold adjustment mechanism, it automatically adapts to changes in ambient lighting to ensure the consistency of detection results. The parallel computing architecture that adopts CUDA kernel function and TensorRT engine collaborative optimization significantly improves reasoning efficiency. The dynamic recognition branch achieves real-time response through lightweight network design and hardware acceleration, while the static recognition branch uses efficient feature extraction and classification strategies to ensure the efficient operation of the overall detection process. The dual-modal architecture design integrates the standardized image features output by the intelligent preprocessing module 20, and on the basis of eliminating environmental interference, enables the dynamic recognition and static recognition branches to achieve efficient collaboration through asynchronous computing, while meeting the dual requirements of industrial sites for detection speed and accuracy.

[0029] See also Figure 5 , Figure 5 The schematic diagram of the structure of the Bayesian optimization module 40 is shown. The Bayesian optimization module 40 includes a parameter space construction unit 401, an objective function calculation unit 402, an iterative optimization unit 403 and a knowledge transfer unit 404, wherein: The parameter space construction unit 401 is used to construct a parameter space based on the multi-dimensional environmental state data and the parameter combination; the objective function calculation unit 402 is used to use a composite objective function to calculate the confidence improvement item and the classification consistency item based on the batch of enhanced images and the corresponding detection results, and generate the optimization target score; the iterative optimization unit 403 is used to use a Gaussian process to construct a parameter and performance relationship model, and iteratively optimize the parameter and performance relationship model according to the optimization target score and the historical observation data set, and output the candidate parameter combination based on the optimized parameter and performance relationship model; the knowledge transfer unit 404 is used to obtain the current environmental characteristics, and perform parameter mapping and template configuration based on the current environmental characteristics and the candidate parameter combination to generate the global optimal parameters and the configuration template.

[0030] Specifically, the parameter space construction unit 401 is used to construct a parameter space based on multi-dimensional environmental state data and parameter combinations. Parameter space definition: Construct a multi-dimensional continuous parameter space covering optical property compensation, geometric distortion correction, and spatial domain filtering, establish a coupling relationship between each parameter dimension through a nonlinear mapping function, and form a constrained optimization model for industrial scenarios. The parameter space design strictly follows the physical laws of industrial visual inspection and constructs a continuously adjustable parameter domain in 10 key dimensions such as gamma correction coefficient, color temperature compensation matrix, and affine transformation parameters. The parameter definitions and engineering constraints of each dimension are shown in Table 2 below: Table 2

[0031] In the embodiment of the present application, in the objective function calculation unit 402, the composite objective function needs to comprehensively consider the dual indicators of model confidence improvement and classification result stability. Through the weighted combination of probability logarithm and classification consistency, it is ensured that the optimization direction takes into account both single frame quality and system stability. The composite optimization objective function is: ; in, is the confidence improvement term, which represents the average logarithmic probability of the correct category predicted by the enhanced model, reflecting the improvement in the detection reliability of a single frame image. The confidence improvement term is calculated as follows: ;in, It is a parameter The enhanced feature vector of the i-th image, is the target detection model, parameter is a vector to be trained in the aforementioned parameter space, and c is a set of possible category results obtained by the target detection model. is the classification consistency term, which quantifies the difference between the model prediction results before and after enhancement, reflecting the stability of the system output. The calculation formula of the classification consistency term is: , is the indicator function (1 if there is a difference, 0 otherwise), is the feature vector of the i-th image without enhancement. is the dynamic weight coefficient, which uses a progressive adjustment strategy to balance the contributions of the two items. The dynamic weight coefficient calculation formula is: , iter is the number of optimizations. Table 3 below is a detailed description of the above formula: Table 3

[0032] Specifically, the iterative optimization unit 403 uses a Gaussian process to construct a parameter-performance relationship model. It iteratively optimizes the parameter-performance relationship model based on the optimization target score and historical observation dataset, and outputs candidate parameter combinations based on the optimized parameter-performance relationship model. Gaussian process modeling uses a kernel function with strong generalization capabilities to construct a probabilistic proxy model of the objective function. The posterior distribution of the parameter-performance relationship is continuously updated through Bayesian inference. The modeling process incorporates an adaptive hyperparameter optimization mechanism to dynamically adjust model complexity to meet the needs of different scenarios. During the candidate parameter combination search step, an improved acquisition function strategy guides the selection of the most promising candidate points in the parameter space. Each iteration generates multiple sets of candidate parameter configurations, and their performance is simultaneously evaluated through a parallel computing architecture, significantly improving optimization efficiency. The knowledge transfer unit 404 is used to obtain current environment characteristics and perform parameter mapping and template configuration based on these characteristics and candidate parameter combinations, generating globally optimal parameters and configuration templates. Within the knowledge transfer unit 404, an optimization experience database is established to support the rapid migration and application of historically optimal parameter configurations in new scenarios, significantly shortening the system's adaptation period to new environments.

[0033] Gaussian process regression is a nonparametric Bayesian model used to probabilistically model the target function. As a core component of Bayesian optimization in this system, it updates the posterior distribution using the prior distribution and observed data to guide parameter search. Parameter space exploration is a technique in Bayesian optimization that uses surrogate models to guide parameter search. This system uses Gaussian processes to model the parameter space and, combined with acquisition functions, allows for targeted exploration of high-potential areas, avoiding inefficient searches.

[0034] Furthermore, in the embodiment of the present application, the system adopts a two-stage optimization strategy: an offline optimization stage and an online adaptation stage. The offline optimization stage is based on the historical data set, and the optimal parameter combination is searched through Bayesian optimization to establish a lighting-parameter mapping model. The online adaptation stage is to fine-tune the parameters in real time to adapt to dynamic lighting changes and ensure that the distribution of model input data is consistent with the training environment. The key performance indicators of the system include: optimization efficiency, generalization ability and stability. Among them, optimization efficiency means that the computational time of a single parameter evaluation must meet the real-time requirements of the production line. Generalization ability means that the optimized parameter configuration must be applicable to different lighting scenarios of similar production lines. Stability refers to the drastic fluctuations in model output caused by improper parameter adjustment.

[0035] The embodiments of the present application achieve joint optimization of optical compensation, geometric correction, and spatial filtering parameters. By constructing a high-dimensional continuous parameter space and corresponding proxy models, the complex interactions between parameters are effectively captured. The embodiments of the present application develop a hybrid acquisition function that integrates expected improvement and confidence bounds, intelligently balancing the relationship between parameter space exploration and high-quality area development. Model compression and computational acceleration technologies are used to achieve efficient parameter optimization with limited hardware resources.

[0036] Furthermore, the system's industrial interface module 50 serves as a key hub connecting the visual inspection system with the industrial control environment, enabling deep integration between the inspection system and production line control equipment. Built based on industrial communication standards, this module utilizes intelligent protocol conversion and data encapsulation technologies to transmit the optimal parameter configuration output by the Bayesian optimization system, the detection results from the model inference module 30, and monitoring data or multi-dimensional environmental status data from the environmental perception module 10 in real time to industrial control systems such as PLCs and MESs, forming a complete "perception-processing-decision-execution" closed loop. The module's communication architecture utilizes a layered isolation approach: the physical layer supports mainstream industrial bus protocols, enabling plug-and-play connectivity with devices from different manufacturers. The data layer uses standardized interface specifications to convert internal data structures into a format recognizable by industrial control systems. The application layer provides three core interfaces: parameter configuration, detection results, and system status. The parameter configuration interface uses XML format to encapsulate parameter combinations generated by the Bayesian optimization system, such as illumination compensation and geometric correction, enabling rapid deployment of parameter solutions across production lines. The detection result interface transmits structured data containing target status, position coordinates, and confidence scores in JSON format, providing decision-making support for production line control.

[0037] In the embodiment of the present application, the industrial interface module 50 integrates an environmental adaptive communication mechanism, which can dynamically adjust the communication strategy according to the environmental perception data: a high-frequency data transmission mode is adopted under stable working conditions to ensure real-time control; when strong electromagnetic interference or network fluctuations are detected, it automatically switches to a redundant transmission mode, and ensures communication reliability through a data verification and retransmission mechanism. At the same time, the module's built-in heartbeat monitoring unit continuously evaluates the communication link status, and immediately triggers a hierarchical recovery strategy when an anomaly is detected: minor anomalies start parameter retransmission, intermediate anomalies switch to a backup communication channel, and serious anomalies enable local cache mode to ensure production continuity. The industrial interface module 50 is not only responsible for sending the optimized parameter configuration to the preprocessing module, but also for real-time feedback of the parameter application effect to the optimization system, forming a closed-loop learning mechanism. This design enables the system to continuously accumulate the optimal parameter combinations under different lighting conditions, gradually build a parameter knowledge base that adapts to a variety of industrial scenarios, and significantly improve the deployment efficiency of new production lines.

[0038] Furthermore, the system also includes: a performance compensation module, which is used to call the intelligent preprocessing module 20 to load the currently verified configuration information if the confidence of the continuous frames in the detection result is lower than the dynamic threshold, and to link the environmental perception module 10 to re-collect new multi-dimensional environmental status data, and fine-tune the parameter combination; an adaptive recovery module, which is used to retrieve historical similar fault scenarios from the parameter knowledge base when the performance degradation result is determined, and activate the emergency search mode of the Bayesian optimization module 40, and perform multi-stage test verification through the model reasoning module 30; an emergency guarantee module, which switches to the standby computing unit, enables the degradation detection mode, and initiates cloud-based remote diagnosis when the system fails.

[0039] Specifically, the system of the present application adopts a three-level linkage exception handling architecture, which deeply cooperates with the aforementioned modules to build a complete system resilience guarantee system. If the continuous frame confidence in the detection result is lower than the dynamic threshold (the threshold is dynamically calculated by the Bayesian optimization system according to the environmental state), the intelligent pre-processing module 20 is called to load the currently verified configuration information, and the environmental perception module 10 is linked to re-collect new multi-dimensional environmental state data, and fine-tune the parameter combination within a safe range within the restricted parameter space. If the performance deteriorates continuously (such as a 20% drop in the average confidence over 30 minutes), the performance degradation result is determined, historical similar fault scenarios are retrieved from the parameter knowledge base, and the emergency search mode of the Bayesian optimization module 40 is activated, and multi-stage test verification is performed through the model reasoning module 30. If there is a hardware failure / communication interruption signal, the system failure is determined, the backup computing unit is switched, the degradation detection mode is enabled, and cloud-based remote diagnosis is initiated.

[0040] Furthermore, the system further comprises: The short-term adjustment module is used to call the real-time parameter adjustment interface of the environment perception module 10 to adjust the parameters of the multi-dimensional environment status data when the illumination mutation or vibration data in the multi-dimensional environment status data exceeds the threshold.

[0041] Specifically, this system adopts an innovative dual-loop optimization architecture, including a short-term rapid adjustment loop and a long-term deep optimization loop, to achieve intelligent adaptation in different time dimensions. The short-term rapid adjustment loop is implemented through the short-term adjustment module. The short-term rapid adjustment loop is designed to address sudden environmental changes on the production line (such as instantaneous strong light interference, equipment vibration, etc.). Based on the real-time monitoring data of the environmental perception module 10, the system conducts a rapid local search within the parameter feasible domain established by the Bayesian optimization system. This mechanism has the following characteristics: (1) Second-level response: directly calling the real-time parameter adjustment interface of the preprocessing module; (2) Safety constraints: following the stability boundary conditions of the model reasoning module 30; (3) Precise compensation: specifically optimizing the ability to suppress instantaneous interference such as sudden changes in illumination.

[0042] The long-term deep optimization loop, designed to continuously improve system performance, periodically initiates a complete Bayesian optimization process, featuring global exploration, multi-objective balancing, and energy efficiency management. Global exploration involves seeking optimal performance within the full parameter space supported by the intelligent preprocessing module 20; multi-objective balancing involves simultaneously optimizing key metrics such as model confidence and classification stability; and energy efficiency management dynamically adjusts computing resource allocation based on device operating status.

[0043] In an embodiment of the present application, the system includes: an environmental perception module 10, an intelligent preprocessing module 20, a model reasoning module 30, a Bayesian optimization module 40 and an industrial interface module 50, wherein: the environmental perception module 10 is used to collect multi-dimensional environmental status data of the industrial site in real time, and collect original images; the intelligent preprocessing module 20 is used to perform optical compensation, geometric correction and spatial filtering on the original image in sequence according to the parameter combination output by the Bayesian optimization module 40 to generate an enhanced image; the model reasoning module 30 is used to identify and detect the enhanced image through a bimodal algorithm to generate a detection result; the Bayesian optimization module 40 is used to calculate the objective function based on the batch of enhanced images and the corresponding detection results, generate an optimization target score, and perform iterative optimization based on the optimization target score and the historical observation data set to generate a candidate parameter combination, and generate a global optimal parameter and configuration template based on the candidate parameter combination and the current environmental characteristics; the industrial interface module 50 is used to transmit the global optimal parameters, the detection results and the multi-dimensional environmental status data to the industrial control system through protocol conversion and data encapsulation technology. The embodiment of the present invention effectively enhances the image by performing optical compensation, geometric correction and spatial filtering on the original image in sequence, and at the same time recognizes and detects the enhanced image through a dual-modal algorithm, which is conducive to detecting dynamic features and static features in the image. In addition, the iterative optimization of the model by the Bayesian optimization module 40 realizes the configuration of global optimal parameters, which is conducive to improving the efficiency of illumination compensation and adaptability in complex environments, and reducing the amount of data calculation.

[0044] See also Figure 6 , Figure 6 A specific implementation of the industrial vision adaptive illumination compensation method based on Bayesian optimization is shown. The industrial vision adaptive illumination compensation method based on Bayesian optimization is implemented by the above industrial vision adaptive illumination compensation system based on Bayesian optimization.

[0045] It should be noted that the method of the present invention is not limited to the method of Figure 1 The process sequence shown is limited to the following steps: S1: Real-time collection of multi-dimensional environmental status data of industrial sites and original images.

[0046] See also Figure 7 , Figure 7 A specific implementation of step S1 is shown, which is described in detail as follows: S11: Using the built-in sensors of an industrial camera, the industrial site's illumination and color characteristic data are collected to generate a multidimensional optical parameter set. S12: By continuously tracking the camera's imaging parameter status, the underlying imaging parameters of the camera are dynamically acquired. S13: Using a vibration sensor to detect mechanical vibration data during equipment operation. S14: The original image is captured by the industrial camera.

[0047] Specifically, the environmental perception module collects multi-dimensional environmental status data of the industrial site in real time, as well as original images.

[0048] S2: Determine a parameter combination based on the multi-dimensional environmental state data, and perform optical compensation, geometric correction, and spatial filtering on the original image in sequence according to the parameter combination to generate an enhanced image.

[0049] See also Figure 2 , Figure 2 A specific implementation of step S2 is shown, which is described in detail as follows: S21: Acquire the original image, and generate the parameter combination based on the multi-dimensional environmental state data; S22: Using a preset compensation algorithm to adjust the illumination distribution data of the original image according to the brightness and color temperature parameters in the parameter combination to generate a first enhanced image, wherein the preset compensation algorithm includes a gamma correction algorithm, a white balance algorithm, and a histogram equalization algorithm; S23: Correcting the first enhanced image according to the underlying imaging parameters in the parameter combination using a preset correction algorithm to generate a second enhanced image; S44: Performing spatial filtering processing on the second enhanced image using an adaptive bilateral filter to generate the enhanced image.

[0050] Specifically, after collecting multi-dimensional environmental status data, environmental feature matching is performed based on the multi-dimensional environmental status data to generate parameter combinations. The environmental feature matching process is as follows: input the environmental feature data set and the pre-built parameter knowledge base (storing the mapping relationship between historical optimal parameters and environmental features), then based on the Gaussian process similarity model, retrieve the historical parameter combination that best matches the environmental features in the knowledge base; screen out candidate parameter configurations (including 10 parameters such as brightness, color temperature correction, and rotation angle). Then, a security verification test is performed on the candidate parameter configurations. This includes: (1) accuracy verification: applying candidate parameters to preprocess the image and calculating the average confidence improvement value of the model; (2) stability verification: statistically analyzing the difference ratio of model classification results before and after enhancement; (3) compatibility testing: verifying the adaptability of parameters to current hardware (such as GPU model). Finally, a parameter combination is generated. The verification process uses a combination of offline simulation and online testing to ensure that the new parameters meet the technical requirements of "supporting rapid deployment across production lines." The original image is subjected to optical compensation, geometric correction, and spatial filtering in sequence through the parameter combination to generate an enhanced image.

[0051] S3: Perform recognition and detection on the enhanced image using a dual-modal algorithm to generate a detection result.

[0052] Specifically, dynamic recognition tracks the robot's motion trajectory and outputs coordinates and confidence levels. Static recognition analyzes product defects and outputs category labels and confidence levels. During the recognition and detection process, model confidence levels and classification results must be recorded in real time.

[0053] After step S3, if sudden environmental change data (such as a strong light alarm) is received, short-term rapid adjustment is required. When the illumination suddenly changes or the vibration exceeds the threshold, the parameters are fine-tuned within a safe range (such as brightness ±0.1), with the constraint that the increase in the difference rate of the classification results is ≤5%. Finally, the fine-tuned parameter configuration is output.

[0054] S4: Calculate the objective function based on the batch of enhanced images and the corresponding detection results to generate an optimized target score, perform iterative optimization based on the optimized target score and the historical observation data set, generate candidate parameter combinations, and generate global optimal parameters and configuration templates based on the candidate parameter combinations and current environmental characteristics.

[0055] Specifically, confidence improvement and classification consistency are calculated based on the batch of enhanced images and the corresponding detection results to generate an optimization target score. Proxy modeling, intelligent search, and parallel evaluation are then performed sequentially to iteratively optimize and generate candidate parameter combinations. Proxy modeling uses a Gaussian process to model the parameter-performance relationship; intelligent search uses a hybrid strategy that integrates expected improvement and confidence bounds to select candidate parameters; and parallel evaluation simultaneously tests the performance of multiple sets of candidate parameters. This is followed by knowledge accumulation and transfer, which includes verifying parameter validity, storing optimized parameters in a knowledge base, establishing a mapping between environmental characteristics and optimal parameters, and generating a standardized XML configuration template.

[0056] Furthermore, after step S4, mechanical energy anomaly processing and system recovery are performed through the performance compensation module, the adaptive recovery module and the emergency guarantee module.

[0057] S5: Transmitting the global optimal parameters, the detection results and the multi-dimensional environmental status data to the industrial control system through protocol conversion and data encapsulation technology.

[0058] Specifically, through intelligent protocol conversion and data encapsulation technology, the optimal parameter configuration output by the Bayesian optimization system, the detection results of the model reasoning module, and the monitoring data of the environmental perception module are transmitted in real time to industrial control systems such as PLC and MES, forming a complete "perception-processing-decision-execution" closed loop.

[0059] This application constructs a multi-dimensional parameter combination space containing optical characteristic parameters, geometric transformation parameters and image processing parameters, and uses dynamic modeling technology to dynamically analyze the complex interactive relationship between parameters. This method breaks through the limitations of traditional discrete parameter space and achieves a more efficient comprehensive optimization search. At the same time, this application incorporates model confidence improvement and classification result stability into a unified objective function framework. Through the organic combination of probability logarithm and maximization with classification consistency constraints, it is ensured that the optimized parameter configuration can not only improve the recognition quality of single-frame images, but also maintain the overall logical coherence of the system. In addition, the Bayesian optimization algorithm guides the parameter search direction by constructing a dynamically updated probabilistic agent model. This application integrates multi-task learning strategies and hybrid acquisition functions to achieve an intelligent balance between exploring unknown areas and developing known optimal solutions, greatly improving optimization efficiency.

[0060] Obviously, the embodiments described above are only some of the embodiments of the present application, rather than all of the embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions described in the aforementioned specific embodiments, or to make equivalent replacements for some of the technical features therein. Any equivalent structure made using the contents of the present application specification and the accompanying drawings, directly or indirectly used in other related technical fields, is also within the scope of protection of the present application.

Claims

1. An industrial vision adaptive illumination compensation system based on Bayesian optimization, characterized in that: include: Environmental perception module, intelligent preprocessing module, model reasoning module, Bayesian optimization module and industrial interface module, including: The environmental perception module is used to collect multi-dimensional environmental status data of the industrial site in real time and collect original images; The intelligent preprocessing module is used to perform optical compensation, geometric correction and spatial filtering on the original image in sequence according to the parameter combination output by the Bayesian optimization module to generate an enhanced image; The model inference module is used to identify and detect the enhanced image using a bimodal algorithm to generate a detection result; The Bayesian optimization module is configured to calculate an objective function based on the batch of enhanced images and the corresponding detection results to generate an optimization target score, perform iterative optimization based on the optimization target score and a historical observation dataset to generate candidate parameter combinations, and generate global optimal parameters and configuration templates based on the candidate parameter combinations and current environmental characteristics; The industrial interface module is used to transmit the global optimal parameters, the detection results and the multi-dimensional environmental status data to the industrial control system through protocol conversion and data encapsulation technology.

2. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The environment perception module includes: An optical environment monitoring unit, which uses the built-in sensor of an industrial camera to collect illumination and color characteristic data of the industrial work area and generate a multi-dimensional optical parameter set; An imaging device state monitoring unit 102 is configured to dynamically obtain the underlying imaging parameters of the camera by continuously tracking the imaging parameter state of the camera; Mechanical environment monitoring unit, used to detect mechanical vibration data during equipment operation through a vibration sensor device; An image acquisition unit is used to acquire the original image through the industrial camera.

3. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The intelligent preprocessing module includes: a data acquisition unit, configured to acquire the original image output by the environment perception module, and to acquire the parameter combination generated by the Bayesian optimization module based on the multi-dimensional environment state data; an optical compensation unit, configured to adjust the illumination distribution data of the original image according to the brightness and color temperature parameters in the parameter combination using a preset compensation algorithm to generate a first enhanced image, wherein the preset compensation algorithm includes a gamma correction algorithm, a white balance algorithm, and a histogram equalization algorithm; a geometric correction unit, configured to correct the first enhanced image according to the underlying imaging parameters in the parameter combination using a preset correction algorithm to generate a second enhanced image; The spatial filtering unit is configured to perform spatial filtering processing on the second enhanced image by using an adaptive bilateral filter to generate the enhanced image.

4. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The enhanced image includes a pre-processed regularized image and a pre-processed uniformly illuminated image. The detection result includes spatial position coordinates, confidence, category label, and abnormal pattern. The model inference module includes: A dynamic recognition unit, configured to recognize and detect the pre-processed regularized image based on a dynamic recognition algorithm using a lightweight neural network architecture, so as to output the spatial position coordinates and the confidence level of the key points of the robotic arm; A static recognition unit is used to perform state discrimination on a preset area in the preprocessed uniform illumination image through a multi-scale feature analysis network, generate the category label, and perform semantic analysis on the preprocessed uniform illumination image to identify the abnormal pattern.

5. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The Bayesian optimization module includes: A parameter space construction unit, configured to construct a parameter space based on the multi-dimensional environmental state data and the parameter combination; an objective function calculation unit, configured to calculate a confidence improvement term and a classification consistency term based on a batch of the enhanced images and the corresponding detection results using a composite objective function, and generate the optimization target score; an iterative optimization unit, configured to construct a parameter-performance relationship model using a Gaussian process, iteratively optimize the parameter-performance relationship model according to the optimization target score and the historical observation data set, and output the candidate parameter combination based on the optimized parameter-performance relationship model; The knowledge transfer unit is used to obtain current environment characteristics, and perform parameter mapping and template configuration based on the current environment characteristics and the candidate parameter combination to generate the global optimal parameters and the configuration template.

6. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The system further comprises: A performance compensation module is configured to call the intelligent preprocessing module to load the currently verified configuration information if the confidence level of consecutive frames in the detection result is lower than a dynamic threshold, and to link the environmental perception module to re-collect new multi-dimensional environmental status data and fine-tune the parameter combination; An adaptive recovery module, configured to retrieve historical similar failure scenarios from a parameter knowledge base when a performance degradation result is determined, activate an emergency search mode of the Bayesian optimization module, and perform multi-stage testing and verification through the model reasoning module; The emergency support module, if it determines that the system fails, switches to the backup computing unit, enables the degradation detection mode, and triggers cloud-based remote diagnosis.

7. The industrial vision adaptive illumination compensation system based on Bayesian optimization according to claim 1, characterized in that: The system further comprises: The short-term adjustment module is used to call the real-time parameter adjustment interface of the environment perception module to adjust the parameters of the multi-dimensional environment status data when the illumination mutation or vibration data in the multi-dimensional environment status data exceeds the threshold.

8. An industrial vision adaptive illumination compensation method based on Bayesian optimization, characterized in that: include: Real-time collection of multi-dimensional environmental status data of industrial sites and original images; determining a parameter combination based on the multi-dimensional environmental state data, and performing optical compensation, geometric correction, and spatial filtering on the original image in sequence according to the parameter combination to generate an enhanced image; Performing recognition and detection on the enhanced image using a dual-modal algorithm to generate a detection result; performing objective function calculation based on the batch of enhanced images and the corresponding detection results to generate an optimization target score, performing iterative optimization based on the optimization target score and a historical observation dataset to generate candidate parameter combinations, and generating a global optimal parameter and configuration template based on the candidate parameter combinations and current environmental characteristics; The global optimal parameters, the detection results and the multi-dimensional environmental status data are transmitted to the industrial control system through protocol conversion and data encapsulation technology.

9. The industrial vision adaptive illumination compensation method based on Bayesian optimization according to claim 8, characterized in that: The real-time acquisition of multi-dimensional environmental status data of the industrial site and the acquisition of original images include: The built-in sensor of the industrial camera collects the lighting data and color characteristic data of the industrial field operation area to generate a multi-dimensional optical parameter set; By continuously tracking the imaging parameter status of the camera, the underlying imaging parameters of the camera are dynamically obtained; Detect mechanical vibration data of the equipment during operation through a vibration sensor device; The original image is collected by the industrial camera.

10. The industrial vision adaptive illumination compensation method based on Bayesian optimization according to claim 8, characterized in that: The method of determining a parameter combination based on the multi-dimensional environmental state data, and sequentially performing optical compensation, geometric correction, and spatial filtering on the original image according to the parameter combination to generate an enhanced image includes: Acquiring the original image, and generating the parameter combination based on the multi-dimensional environmental state data; Using a preset compensation algorithm to adjust the illumination distribution data of the original image according to the brightness and color temperature parameters in the parameter combination to generate a first enhanced image, wherein the preset compensation algorithm includes a gamma correction algorithm, a white balance algorithm, and a histogram equalization algorithm; Correcting the first enhanced image according to the underlying imaging parameters in the parameter combination using a preset correction algorithm to generate a second enhanced image; An adaptive bilateral filter is used to perform spatial filtering on the second enhanced image to generate the enhanced image.

Images