Dual-band imaging processing system based on multi-intelligent algorithm cooperation

Abstract

The application belongs to the technical field of imaging processing, and is a dual-band imaging processing system based on multi-intelligent algorithm cooperation, comprising an intelligent imaging processing unit, a processing effect dynamic feedback unit and a user terminal; the application realizes accurate timing alignment and targeted denoising through a dual-band source signal synchronous modulation module, enriches feature dimension while ensuring computing efficiency through a heterogeneous feature hierarchical extraction module, generates an optimal processing scheme adapted to different scenes through a multi-algorithm cooperative decision module that integrates deep learning, fuzzy reasoning and reinforcement learning, realizes accurate fusion and resolution improvement based on the decision result through a dual-band image adaptive reconstruction module, the processing effect dynamic feedback unit optimizes parameters in real time, the multi-factor judgment output unit monitors the operating state and timely alarms, and the overall realization of high synchronism, high detail retention rate, low noise characteristics and strong scene adaptability of dual-band imaging, continuously improves the imaging quality, and meets the application requirements of industrial detection, security monitoring and other scenes.

Description

Technical Field

[0001] This invention relates to the field of imaging processing technology, specifically to a dual-band imaging processing system based on the collaborative use of multiple intelligent algorithms. Background Technology

[0002] Dual-band imaging processing technology effectively overcomes the limitations of single-band imaging in detail capture and scene adaptation by integrating complementary information from images of different bands. It is widely used in key fields such as security monitoring, industrial defect detection, and remote sensing. The clarity of the imaging quality, the comprehensiveness of feature extraction, and the stability of system operation directly determine the decision-making efficiency and reliability of the terminal application. As application scenarios expand towards complex lighting, dynamic targets, and long-term continuous operation, higher requirements are placed on the algorithm adaptability, data processing depth, and system control capabilities of dual-band imaging processing.

[0003] Currently, Chinese invention patent CN104835129A discloses "a dual-band image fusion method using local window visual attention extraction". This technology targets visible light and infrared dual-band images, extracts multi-scale local window visual attention maps, and generates enhanced fused images by weighted fusion method, so as to solve the problem of insufficient consideration of target and background information in traditional fusion methods and improve the visual enhancement effect of images.

[0004] However, this scheme still has the following obvious limitations in practical applications. On the one hand, the algorithm design is singular and rigid, relying solely on the multi-scale fusion logic of visual attention maps without introducing a multi-intelligent algorithm collaborative decision-making mechanism. It cannot dynamically adjust the processing strategy according to the imaging characteristics of different scenarios, which leads to problems such as loss of details and artifact residue in complex scenarios. On the other hand, it lacks a full-dimensional feature extraction system, focusing only on the visual attention features of local windows and failing to fully explore the shallow details of single-band and the deep correlation features of dual-band. The limitation of feature dimensions directly affects the accuracy of subsequent fusion processing. Furthermore, it has not built a monitoring and feedback mechanism for the system's operating status, making it impossible to identify and optimize abnormalities such as poor processing results or hardware overload in a timely manner. It is also prone to affecting the stability of the system due to long-term operation failures.

[0005] To address the aforementioned technical shortcomings, a solution is proposed. Summary of the Invention

[0006] The purpose of this invention is to provide a dual-band imaging processing system based on the collaboration of multiple intelligent algorithms, so as to solve the technical defects mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a dual-band imaging processing system based on multi-intelligent algorithm collaboration, comprising an intelligent imaging processing unit, a processing effect dynamic feedback unit, and a user terminal; wherein, the intelligent imaging processing unit consists of a dual-band source signal synchronous modulation module, a heterogeneous feature hierarchical extraction module, a multi-algorithm collaborative decision-making module, and a dual-band image adaptive reconstruction module; The intelligent imaging processing unit is used to perform dual-band imaging processing and output high-quality dual-band fused images; the processing effect dynamic feedback unit receives the dual-band fused images, dynamically evaluates the processing effect of the intelligent imaging processing unit, and feeds back the evaluation results to the user terminal and the intelligent imaging processing unit.

[0008] Furthermore, the dual-band source signal synchronization modulation module is used to receive the dual-band raw imaging signal, realize signal timing synchronization calibration and initial noise pre-suppression, and output a dual-band modulation signal with high synchronization and low noise; the heterogeneous feature layer extraction module extracts the heterogeneous features of the dual-band modulation signal through a layered architecture, including single-band shallow detail features and dual-band deep correlation features, and outputs a layered feature set with complete dimensions. The multi-algorithm collaborative decision-making module receives a hierarchical feature set and generates the optimal processing decision result for the current imaging scene through collaborative reasoning of three intelligent algorithms: deep learning, fuzzy reasoning, and reinforcement learning. The dual-band image adaptive reconstruction module generates a high-quality dual-band fused image based on the hierarchical feature set and the optimal processing decision result through adaptive fusion and super-resolution reconstruction.

[0009] Furthermore, the dual-band source signal synchronization modulation module receives the original electrical signal transmitted from the dual-band imaging source through the signal interface, calls the adaptive timing calibration algorithm, uses the infrared signal as the reference timing, and dynamically adjusts the sampling clock frequency of the visible light signal by calculating the timing offset of the visible light signal in real time, so that the frame frequency and sampling points of the dual-band signal are completely aligned. After timing calibration, the wavelet threshold denoising algorithm is started. Based on the noise characteristics of the dual-band signal, the wavelet basis function and threshold parameters are adaptively adjusted to perform hierarchical denoising on the calibrated signal, removing random noise and impulse noise from the signal. Finally, the amplitude of the denoised dual-band signal is normalized to uniformly map the signal amplitude to the range of 0 to 255, generating a synchronously modulated dual-band signal and transmitting it to the heterogeneous feature hierarchical extraction module.

[0010] Furthermore, the heterogeneous feature layer extraction module performs band separation on the synchronous modulation signal, splitting the signal into visible photonic signals and infrared sub-signals. For the visible photonic signals, a lightweight convolutional neural network is invoked, using 3 convolutional layers and 2 pooling layers to extract shallow detail features, including edges and textures, from the signal, generating a visible light shallow feature map. For the infrared sub-signals, a lightweight convolutional neural network is used to extract shallow features, including temperature gradients and thermal target contours, from the signal, generating an infrared shallow feature map. The visible light shallow feature map and the infrared shallow feature map are input into the encoder of the Transformer architecture. The correlation weight between the two band features is calculated through the self-attention mechanism, and the deep correlation features of the two bands are extracted to generate a deep correlation feature map. The visible light shallow feature map, the infrared shallow feature map, and the deep correlation feature map are unified in dimension and feature stitched together to form a hierarchical feature set, which is then transmitted to the multi-algorithm collaborative decision-making module and the dual-band image adaptive reconstruction module, respectively.

[0011] Furthermore, the multi-algorithm collaborative decision-making module inputs the hierarchical feature set into the pre-trained deep learning classification model. The deep learning classification model identifies the current imaging scene type by training on historical scene data and outputs scene category labels and preliminary processing parameters. The fuzzy inference algorithm is invoked, taking scene category labels as input and combining them with a fuzzy rule base constructed from expert experience to perform fuzzy correction on the initial processing parameters, generating corrected processing parameters. The reinforcement learning algorithm is then activated, using the "imaging quality assessment index" as the reward function and the corrected processing parameters as action variables. During real-time processing, the parameter values ​​are dynamically adjusted to generate the optimal processing decision result and transmit it to the dual-band image adaptive reconstruction module.

[0012] Furthermore, the dual-band image adaptive reconstruction module performs pixel-level fusion of the visible light shallow feature map and the infrared shallow feature map in the hierarchical feature set according to the fusion weight matrix in the optimal processing decision result, and generates an initial fused feature map. The initial fused feature map is concatenated with the deep correlation feature map at the channel level to supplement the dual-band correlation information and generate a complete fused feature map. The adaptive super-resolution reconstruction algorithm is called to perform resolution enhancement processing on the complete fused feature map according to the scaling factor in the optimal processing decision result, generating a high-resolution fused feature map. The high-resolution fused feature map is mapped to a grayscale image or a color image through a convolutional layer, and after contrast enhancement and color correction processing, the final dual-band fused image is generated and output.

[0013] Furthermore, the specific operation process of the dynamic feedback unit for processing results is as follows: Objective quality assessment is performed on dual-band fused images by calculating three core indicators: PSNR, SSIM, and information entropy. A subjective perception assessment model is then invoked to score the "visual comfort" and "detail recognition" of the fused images by simulating human visual characteristics. The objective indicators are combined with the subjective scores to generate a comprehensive assessment result. If the assessment result is "needs optimization," the root cause of the problem is analyzed. If the assessment result is "excellent," the current processing parameters are kept unchanged.

[0014] Furthermore, the dynamic feedback unit for processing effect is connected to the multi-factor judgment output unit. The dynamic feedback unit for processing effect sends the comprehensive evaluation results to the multi-factor judgment output unit. The multi-factor judgment output unit performs multi-factor judgment analysis on the operating status of the intelligent imaging processing unit during the detection period. Through analysis, it determines whether to generate a processing alarm signal. When a processing alarm signal is generated, it is sent to the user terminal. When the user terminal receives the processing alarm signal, it issues a warning.

[0015] Furthermore, the specific analysis process for the multi-factor judgment output unit includes: The system acquires all comprehensive evaluation results output by the dynamic feedback unit of the processing effect during the detection period, marks the proportion of evaluation results as "to be optimized" as the imaging processing anomaly value, compares the imaging processing anomaly value with the preset imaging processing anomaly threshold, and generates a processing alarm signal if the imaging processing anomaly value exceeds the preset imaging processing anomaly threshold.

[0016] Furthermore, if the abnormal measurement value of the imaging processing does not exceed the preset abnormal measurement threshold of the imaging processing, then 20 frames are used as a processing round, the maximum number of rounds in which the "to be optimized" evaluation result appears consecutively within the detection period is counted, and the ratio of the maximum number of rounds to the preset threshold for consecutive rounds to be optimized is calculated to obtain the wheel spoke value to be optimized; and the processing time of all single frames within the detection period is obtained, and the average processing time of all single frames within the detection period is compared with the preset processing time threshold to calculate the efficiency value at the frame. Furthermore, the CPU and memory usage during the monitoring process are recorded. A Cartesian coordinate system is established with time as the X-axis and CPU usage as the Y-axis. The CPU usage change curve during the monitoring period is plotted in the first quadrant of the Cartesian coordinate system. A CPU pressure ray parallel to the X-axis and with its endpoint on the Y-axis is drawn in the first quadrant. The area enclosed by the portion of the CPU usage change curve above the CPU pressure ray and the CPU pressure ray is marked as the CPU overpressure region. The area of ​​all overpressure regions is obtained and marked as the CPU overpressure value. Similarly, the memory overpressure value is obtained. The multi-factor decision coefficient is obtained by weighted summation of the imaging processing anomaly value, the wheel spoke value to be optimized, the frame efficiency value, the CPU overvoltage value, and the memory overvoltage value. The multi-factor decision coefficient is then compared with a preset multi-factor decision coefficient threshold. If the multi-factor decision value exceeds the preset multi-factor decision threshold, a processing alarm signal is generated.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. In this invention, a dual-band source signal synchronous modulation module is used to achieve precise timing alignment and targeted denoising. A heterogeneous feature layer extraction module enriches the feature dimensions while ensuring computational efficiency. A multi-algorithm collaborative decision-making module generates the optimal processing scheme adapted to different scenarios. A dual-band image adaptive reconstruction module achieves precise fusion and resolution improvement based on the decision results. A dynamic feedback unit for processing effect optimizes parameters in real time, making the imaging more in line with actual visual needs and helping to continuously improve imaging quality.

[0018] 2. In this invention, a multi-factor judgment output unit is used to achieve a comprehensive and accurate assessment of the operating status of the intelligent imaging processing unit during the detection period. It can not only identify the continuous potential problems of the processing effect through imaging-related indicators, but also capture the hardware load risk of the system operation through resource consumption and processing efficiency indicators. This allows managers to quickly know about system anomalies and take timely optimization measures, significantly improving the stability, reliability and controllability of the dual-band imaging processing system. Attached Figure Description

[0019] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a system block diagram of Embodiment 1 of the present invention; Figure 2 This is a system block diagram of Embodiment 2 of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Example 1: As Figure 1As shown, the dual-band imaging processing system based on multi-intelligent algorithm collaboration proposed in this invention includes an intelligent imaging processing unit, a dynamic feedback unit for processing effect, and a user terminal. The intelligent imaging processing unit is used to perform dual-band imaging processing and output a high-quality dual-band fused image. The dynamic feedback unit for processing effect receives the dual-band fused image, dynamically evaluates the processing effect of the intelligent imaging processing unit, and feeds back the evaluation results to the user terminal and the intelligent imaging processing unit.

[0022] Specifically, the dynamic feedback unit for processing effects enables the system to continuously adjust parameters based on the actual processing results. Image quality can be effectively improved after multiple rounds of processing. Furthermore, by combining objective indicators with subjective perception evaluation, it avoids the problem that a single objective indicator cannot reflect the human visual experience, making the optimized image more in line with the visual needs of actual application scenarios. The specific operation process of the dynamic feedback unit for processing effects is as follows: First, an objective quality assessment is performed on the dual-band fused image, calculating three core indicators: PSNR, SSIM, and information entropy. PSNR reflects the image noise level, SSIM reflects the image structural similarity, and information entropy reflects the image information richness. Then, a subjective perception assessment model is invoked, simulating human visual characteristics (such as higher sensitivity to edges and textures than to smooth areas), to score the "visual comfort" and "detail recognition" of the fused image (out of 10). Then, objective indicators are combined with subjective scores to generate a comprehensive evaluation result (e.g., "Excellent" or "Needs Optimization"). If the evaluation result is "Needs Optimization" (e.g., PSNR < 30dB, subjective score < 7 points), the root cause of the problem is analyzed. For example, if PSNR is low, it means that the signal noise has not been effectively suppressed, and the noise reduction threshold of the dual-band source signal synchronous modulation module needs to be adjusted; if SSIM is low, it means that the feature fusion effect is poor, and the Transformer attention weight of the heterogeneous feature hierarchical extraction module needs to be adjusted; if the subjective score is low, it means that the reconstruction details are insufficient, and the contrast enhancement parameters of the dual-band image adaptive reconstruction module need to be adjusted. Finally, the adjusted parameters are used to generate feedback signals, which are then transmitted to each module in the intelligent imaging processing unit. After receiving the feedback signals, the relevant modules update their own algorithm parameters to optimize the next round of processing. If the evaluation result is "excellent", the current parameters are kept unchanged, and only the current processing log is stored for reference in subsequent scene adaptation.

[0023] It should be noted that the intelligent imaging processing unit consists of a dual-band source signal synchronous modulation module, a heterogeneous feature hierarchical extraction module, a multi-algorithm collaborative decision-making module, and a dual-band image adaptive reconstruction module; The dual-band source signal synchronization modulation module is used to receive the original dual-band imaging signals, realize signal timing synchronization calibration and initial noise pre-suppression, and output a dual-band modulation signal with high synchronization and low noise. This solves the problem of timing deviation of dual-band signals in traditional systems, and improves the timing synchronization accuracy from the 10ms level of existing technologies to the 1ms level, laying the foundation for subsequent image registration and feature fusion. In addition, it performs targeted denoising for the differentiated noise characteristics of dual-band signals. Compared with traditional general denoising algorithms, the noise suppression rate is significantly improved, while preserving the detailed information of the signal (such as the edge details of visible light signals and the temperature gradient of infrared signals).

[0024] Specifically, the dual-band source signal synchronization modulation module first receives the raw electrical signal transmitted by the dual-band imaging source through a signal interface (such as a visible light camera or an infrared thermal imager). The visible light signal is a high-frequency detail signal, and the infrared signal is a low-frequency temperature field signal. There is a millisecond-level timing deviation between the two. Then, the adaptive timing calibration algorithm is invoked. Using the infrared signal as the reference timing, the sampling clock frequency of the visible light signal is dynamically adjusted by calculating the timing offset of the visible light signal in real time, so that the frame rate and sampling points of the dual-band signal are completely aligned. After timing calibration, the wavelet threshold denoising algorithm is started. Based on the noise characteristics of the dual-band signal (visible light signal is mainly Gaussian noise, and infrared signal is mainly salt-and-pepper noise), the wavelet basis function and threshold parameters are adaptively adjusted to perform hierarchical denoising on the calibrated signal, removing random noise and impulse noise from the signal. Finally, the amplitude of the denoised dual-band signal is normalized to uniformly map the signal amplitude to the range of 0 to 255, generating a synchronously modulated dual-band signal and transmitting it to the heterogeneous feature hierarchical extraction module.

[0025] The heterogeneous feature layered extraction module extracts heterogeneous features of dual-band modulated signals through a layered architecture, including single-band shallow detail features and dual-band deep correlation features. It outputs a complete layered feature set, breaking through the limitation of traditional single algorithms that can only extract single-type features. It realizes layered feature extraction of "shallow details + deep correlation", increasing the feature dimension by 2-3 times compared with existing technologies. It can cover the multi-dimensional information requirements of dual-band imaging. Moreover, it adopts an architecture that combines lightweight convolutional neural networks and Transformers, which improves computational efficiency while ensuring feature extraction accuracy and meeting the real-time processing requirements of the system.

[0026] Specifically, the heterogeneous feature layer extraction module first performs band separation on the synchronous modulation signal, splitting the signal into visible photonic signals and infrared sub-signals. For the visible photonic signals, a lightweight convolutional neural network (such as MobileNetV3) is invoked, using 3 convolutional layers and 2 pooling layers to extract shallow detail features such as edges and textures from the signal, generating a shallow visible light feature map. For the infrared sub-signals, a similar lightweight convolutional neural network is used to extract shallow features such as temperature gradients and thermal target contours from the signal, generating a shallow infrared feature map. Subsequently, the visible light shallow feature map and the infrared shallow feature map are input into the encoder of the Transformer architecture. The association weights between the two-band features are calculated through the self-attention mechanism (such as the position association between the thermal target and the visible light target, and the attribute association between temperature and texture). The deep association features of the two bands are extracted to generate a deep association feature map. Finally, the visible light shallow feature map, infrared shallow feature map, and deep correlation feature map are dimension-unified and feature stitched together to form a hierarchical feature set (forming a hierarchical feature set with dimensions of H×W×C, where H is the feature map height, W is the width, and C is the number of feature channels) and transmitted to the multi-algorithm collaborative decision-making module and the dual-band image adaptive reconstruction module, respectively.

[0027] The multi-algorithm collaborative decision-making module receives a hierarchical feature set and generates the optimal processing decision result (including image fusion weights, reconstruction parameters, etc.) for the current imaging scene through collaborative reasoning of three intelligent algorithms: deep learning, fuzzy reasoning, and reinforcement learning. This provides decision guidance for dual-band image reconstruction, avoids the limitations of traditional single-algorithm decision-making, significantly improves decision accuracy in complex scenes, and can adapt to different imaging environments and application requirements. Furthermore, the introduction of reinforcement learning enables dynamic optimization of decision results, allowing processing parameters to match scene changes in real time.

[0028] Specifically, the multi-algorithm collaborative decision-making module first inputs the hierarchical feature set into a pre-trained deep learning classification model (such as ResNet50). The deep learning classification model can identify the current imaging scene type by training on historical scene data (such as daytime strong light scene, nighttime weak light scene, and complex background scene) and output scene category labels (such as "daytime strong light - industrial inspection" and "nighttime weak light - security monitoring") and preliminary processing parameters (such as fusion weight range and reconstructed resolution benchmark). Subsequently, the fuzzy inference algorithm is invoked, taking the scene category label as input and combining it with the fuzzy rule library built by expert experience (such as "nighttime low light scene → infrared feature weight increased by 20%" and "industrial inspection scene → visible light detail weight increased by 30%) to perform fuzzy correction on the preliminary processing parameters, solve the parameter deviation problem of deep learning model in complex and uncertain scenes (such as cloudy weather and dynamic target scenes), and generate the corrected processing parameters. Finally, the reinforcement learning algorithm is launched, using "imaging quality evaluation indicators (such as peak signal-to-noise ratio PSNR, structural similarity SSIM)" as the reward function, and the corrected processing parameters as action variables. During real-time processing, the parameter values ​​are dynamically adjusted (for example, if the PSNR and SSIM of the subsequent reconstructed image improve, the selection probability of the current parameter is increased; if the indicators decrease, the parameters are adjusted to the optimal range). Finally, the optimal processing decision result (including the specific dual-band fusion weight matrix, super-resolution reconstruction scaling factor, etc.) is generated and transmitted to the dual-band image adaptive reconstruction module.

[0029] The dual-band image adaptive reconstruction module generates high-quality dual-band fused images through adaptive fusion and super-resolution reconstruction based on hierarchical feature sets and optimal processing decision results. The adaptive fusion and reconstruction based on decision results avoids the "detail loss" or "artifact generation" problems caused by traditional fixed fusion methods. The detail retention rate of the fused image is improved and the artifact occurrence rate is reduced. Moreover, the adaptive super-resolution reconstruction algorithm can adjust the scaling factor according to the scene requirements. Compared with fixed resolution output, the system has stronger adaptability and can meet the needs of different display devices and detection accuracy (such as 4K resolution for industrial inspection and 1080P resolution for security monitoring).

[0030] Specifically, the dual-band image adaptive reconstruction module first performs pixel-level fusion of the visible light shallow feature map and the infrared shallow feature map in the hierarchical feature set based on the fusion weight matrix in the optimal processing decision result. For each pixel, the weighted sum of the visible light feature pixel and the infrared feature pixel (such as the visible light pixel value with a weight of 0.6 + the infrared pixel value with a weight of 0.4) is calculated based on the corresponding weight value in the fusion weight matrix to generate the initial fused feature map. Subsequently, the initial fused feature map and the deep correlation feature map are concatenated at the channel level to supplement dual-band correlation information and generate a complete fused feature map. Then, an adaptive super-resolution reconstruction algorithm (such as the EDSR algorithm) is called to perform resolution enhancement processing on the complete fused feature map according to the scaling factor (such as 2x or 4x) in the optimal processing decision result. The algorithm dynamically adjusts the number of residual blocks (selecting 8-16 residual blocks according to feature complexity) to repair the loss of detail in the reconstruction process and generate a high-resolution fused feature map. Finally, the high-resolution fused feature map is mapped to a grayscale image or a color image (selected according to application requirements) through a convolutional layer. After contrast enhancement and color correction, the final dual-band fused image is generated and output.

[0031] Example 2: Figure 2As shown, the difference between this embodiment and Embodiment 1 is that the processing effect dynamic feedback unit is communicatively connected to the multi-factor judgment output unit. The processing effect dynamic feedback unit sends the comprehensive evaluation result to the multi-factor judgment output unit. The multi-factor judgment output unit performs multi-factor judgment analysis on the operating status of the intelligent imaging processing unit during the detection period. Through analysis, it determines whether a processing alarm signal is generated. When a processing alarm signal is generated, it is sent to the user terminal. When the user terminal receives the processing alarm signal, it issues a warning. The multi-factor judgment output unit overcomes the limitations of single-indicator judgment, enabling a comprehensive and accurate assessment of the intelligent imaging processing unit's operational status during the detection period. It not only identifies persistent risks to processing effectiveness through imaging-related indicators but also captures hardware load risks in system operation through resource consumption and processing efficiency indicators. This ensures the assessment covers all dimensions of "imaging quality, operational efficiency, and resource consumption," allowing managers to quickly identify system anomalies and take timely optimization measures. This effectively avoids system failures caused by poor imaging quality, low processing efficiency, or hardware overload, significantly improving the stability, reliability, and controllability of the dual-band imaging processing system. The specific analysis process of the multi-factor judgment output unit is as follows: All comprehensive evaluation results output by the dynamic feedback unit of processing effect during the detection period are obtained. The proportion of evaluation results marked as "to be optimized" is marked as the imaging processing anomaly value. The imaging processing anomaly value is compared with the preset imaging processing anomaly threshold. If the imaging processing anomaly value exceeds the preset imaging processing anomaly threshold, it indicates that the operation of the intelligent imaging processing unit is poor during the detection period. It is necessary to strengthen operation supervision and take reasonable improvement measures, and then generate a processing alarm signal.

[0032] Furthermore, if the abnormal measurement value of the imaging processing does not exceed the preset abnormal measurement threshold of the imaging processing, then 20 frames are used as a processing round, the maximum number of rounds in which the "to be optimized" evaluation result appears consecutively within the detection period is counted, and the ratio of the maximum number of rounds to the preset threshold for consecutive rounds to be optimized is calculated to obtain the wheel spoke value to be optimized; and the processing time of all single frames within the detection period is obtained, and the average processing time of all single frames within the detection period is compared with the preset processing time threshold to calculate the efficiency value at the frame. Furthermore, the CPU and memory usage during the monitoring process are recorded. A Cartesian coordinate system is established with time as the X-axis and CPU usage as the Y-axis. The CPU usage change curve during the monitoring period is plotted in the first quadrant of the Cartesian coordinate system. A CPU pressure ray parallel to the X-axis and with its endpoint on the Y-axis is drawn in the first quadrant. The area enclosed by the portion of the CPU usage change curve above the CPU pressure ray and the CPU pressure ray is marked as the CPU overpressure region. The area of ​​all overpressure regions is obtained and marked as the CPU overpressure value. Similarly, the memory overpressure value is obtained. The multi-factor decision coefficient is obtained by weighted summation of the imaging processing anomaly value, the spoke value to be optimized, the frame efficiency value, the CPU overpressure value, and the memory overpressure value. Specifically, the corresponding preset weight coefficients are assigned to the imaging processing anomaly value, the spoke value to be optimized, the frame efficiency value, the CPU overpressure value, and the memory overpressure value. The imaging processing anomaly value, the spoke value to be optimized, the frame efficiency value, the CPU overpressure value, and the memory overpressure value are multiplied by the corresponding preset weight coefficients, and the sum of the five sets of results is marked as the multi-factor decision coefficient. It should be noted that the larger the value of the multi-factor decision coefficient, the worse the overall operation of the intelligent imaging processing unit during the detection period. The multi-factor decision coefficient is compared with the preset multi-factor decision coefficient threshold. If the multi-factor decision value exceeds the preset multi-factor decision threshold, it indicates that the overall operation of the intelligent imaging processing unit during the detection period is poor, and it is necessary to strengthen operation supervision and take reasonable improvement measures. In this case, a processing alarm signal is generated.

[0033] The working principle of this invention is as follows: During use, a dual-band source signal synchronous modulation module achieves precise timing alignment and targeted denoising, solving the problems of traditional signal deviation and noise interference. A heterogeneous feature layer extraction module considers both shallow details in single-band and deep correlation features in dual-band, enriching feature dimensions while ensuring computational efficiency. A multi-algorithm collaborative decision-making module integrates deep learning, fuzzy inference, and reinforcement learning to generate optimal processing solutions adapted to different scenarios, improving processing targeting. A dual-band image adaptive reconstruction module achieves precise fusion and resolution improvement based on decision results, reducing detail loss and artifact generation. A dynamic feedback unit for processing effects optimizes parameters in real time through evaluation combining objective indicators and subjective perception, making imaging more aligned with actual visual needs. A multi-factor judgment output unit monitors the operating status and provides timely warnings, ensuring system stability and reliability. Overall, this invention achieves high synchronization, high detail retention, low noise characteristics, and strong scene adaptability in dual-band imaging, continuously improving imaging quality and meeting the application needs of various scenarios such as industrial inspection and security monitoring.

[0034] In this invention, the threshold, preset value, or preset range settings are for result comparison and analysis to determine whether the result is good or bad. The magnitude of these values ​​is determined by a combination of large-scale model analysis of sample data and human experience, and can also be appropriately adjusted based on seasonal or common-sense influence conditions. Similarly, the preset weight coefficients and influence factors are assigned specific values ​​based on the magnitude of each parameter's influence on the result, ultimately reflecting the impact on the result. These settings are also determined by a combination of large-scale model analysis of sample data and human experience, and can also be appropriately adjusted based on seasonal or common-sense influence conditions.

[0035] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, enabling those skilled in the art to better understand and utilize it. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A dual-band imaging processing system based on multi-intelligent algorithm collaboration, characterized in that, It includes an intelligent imaging processing unit, a dynamic feedback unit for processing results, and a user terminal; the intelligent imaging processing unit consists of a dual-band source signal synchronous modulation module, a heterogeneous feature hierarchical extraction module, a multi-algorithm collaborative decision-making module, and a dual-band image adaptive reconstruction module; The intelligent imaging processing unit is used to perform dual-band imaging processing and output high-quality dual-band fused images; the processing effect dynamic feedback unit receives the dual-band fused images, dynamically evaluates the processing effect of the intelligent imaging processing unit, and feeds back the evaluation results to the user terminal and the intelligent imaging processing unit.

2. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 1, characterized in that, The dual-band source signal synchronization modulation module is used to receive the dual-band raw imaging signal, realize signal timing synchronization calibration and initial noise pre-suppression, and output the dual-band modulated signal; the heterogeneous feature hierarchical extraction module extracts the heterogeneous features of the dual-band modulated signal through a hierarchical architecture and outputs a hierarchical feature set. The multi-algorithm collaborative decision-making module generates the optimal processing decision result for the current imaging scene through the collaborative reasoning of three intelligent algorithms; the dual-band image adaptive reconstruction module generates a dual-band fused image based on the hierarchical feature set and the optimal processing decision result through adaptive fusion and super-resolution reconstruction.

3. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 2, characterized in that, The dual-band source signal synchronization modulation module receives the original electrical signal transmitted from the dual-band imaging source through the signal interface, calls the adaptive timing calibration algorithm, uses the infrared signal as the reference timing, calculates the timing offset of the visible light signal in real time, and dynamically adjusts the sampling clock frequency of the visible light signal to make the frame frequency and sampling points of the dual-band signal completely aligned. After timing calibration, the wavelet threshold denoising algorithm is started. Based on the noise characteristics of the dual-band signal, the wavelet basis function and threshold parameters are adaptively adjusted to perform hierarchical denoising on the calibrated signal. Finally, the amplitude of the denoised dual-band signal is normalized to map the signal amplitude to the 0-255 range, generate the synchronously modulated dual-band signal and transmit it to the heterogeneous feature hierarchical extraction module.

4. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 3, characterized in that, The heterogeneous feature layer extraction module performs band separation on the synchronous modulation signal, splitting the signal into visible photonic signals and infrared sub-signals. The visible light shallow feature map and the infrared shallow feature map are input into the encoder of the Transformer architecture. The correlation weight between the two band features is calculated through the self-attention mechanism to extract the deep correlation features of the two bands and generate a deep correlation feature map. The visible light shallow feature map, the infrared shallow feature map, and the deep correlation feature map are dimension-unified and feature-stitched to form a layered feature set, which is then transmitted to the multi-algorithm collaborative decision-making module and the dual-band image adaptive reconstruction module, respectively.

5. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 4, characterized in that, The multi-algorithm collaborative decision-making module inputs the hierarchical feature set into the pre-trained deep learning classification model, identifies the current imaging scene type, and outputs scene category labels and preliminary processing parameters; it calls the fuzzy inference algorithm, takes the scene category labels as input, and combines the fuzzy rule base constructed by expert experience to perform fuzzy correction on the preliminary processing parameters and generate the corrected processing parameters. During real-time processing, parameter values ​​are dynamically adjusted to generate the optimal processing decision result and transmit it to the dual-band image adaptive reconstruction module.

6. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 5, characterized in that, The dual-band image adaptive reconstruction module performs pixel-level fusion of the visible light shallow feature map and the infrared shallow feature map in the layered feature set to generate an initial fused feature map. The initial fused feature map and the deep correlation feature map are concatenated at the channel level to supplement the dual-band correlation information and generate a complete fused feature map. The complete fused feature map is then subjected to resolution enhancement to generate a high-resolution fused feature map. The high-resolution fused feature map is then mapped to a grayscale or color image through a convolutional layer. After contrast enhancement and color correction, a dual-band fused image is generated and output.

7. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 1, characterized in that, The specific operation process of the dynamic feedback unit for processing effect is as follows: objective quality assessment is performed on the dual-band fused image, and the three core indicators of the image, PSNR, SSIM, and information entropy, are calculated; the subjective perception assessment model is called, and the "visual comfort" and "detail recognition" of the fused image are scored by simulating the visual characteristics of the human eye; and the objective indicators are combined with the subjective scores to generate a comprehensive evaluation result.

8. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 7, characterized in that, The dynamic feedback unit for processing effect sends the comprehensive evaluation results to the multi-factor judgment output unit. The multi-factor judgment output unit performs multi-factor judgment analysis on the operation status of the intelligent imaging processing unit during the detection period and sends it to the user terminal when generating a processing alarm signal.

9. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 8, characterized in that, The specific analysis process of the multi-factor judgment output unit includes: All comprehensive evaluation results output by the dynamic feedback unit of the processing effect during the detection period are obtained. The percentage of evaluation results marked as "to be optimized" is marked as the imaging processing abnormal value. If the imaging processing abnormal value exceeds the preset imaging processing abnormal value threshold, a processing alarm signal is generated.

10. The dual-band imaging processing system based on multi-intelligent algorithm collaboration according to claim 9, characterized in that, If the abnormal measurement value of imaging processing does not exceed the preset abnormal measurement threshold of imaging processing, the multi-factor decision coefficient is obtained by weighted summation of the abnormal measurement value of imaging processing, the wheel spoke value to be optimized, the frame efficiency value, the CPU overvoltage value and the memory overvoltage value. If the multi-factor decision value exceeds the preset multi-factor decision threshold, a processing alarm signal is generated.

Images