Fast video image adjustment method based on deep learning neural network

Abstract

The embodiment of the present application provides a method for rapid adjustment of video images based on a deep learning neural network. By constructing a parameter mapping database for different lighting scenes, an image quality evaluation system is established based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and a channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically to a method for rapid adjustment of video images based on a deep learning neural network. Background Art

[0002] In the field of video surveillance, automatic adjustment of image quality has always been an important technical challenge. Traditional image parameter adjustment methods mainly rely on fixed algorithm rules and preset thresholds, which are difficult to adapt to complex and changing lighting environments. Although with the development of automatic exposure technology, some parameter adjustment solutions based on scene recognition have emerged, these solutions still have obvious shortcomings when dealing with fast lighting changes and target tracking scenes.

[0003] Existing image adjustment systems generally have problems such as slow response speed and unstable adjustment effect. Under complex lighting conditions, especially when facing scenes such as low light indoors, strong light outdoors, and low light at night, the system often has difficulty in accurately identifying moving targets and making effective parameter adjustments. At the same time, traditional methods perform poorly in dealing with the balance between the target area and the background area, and it is easy to cause the target to be overexposed or the background to be too dark. In addition, the existing system lacks effective use of historical parameter adjustment experience and cannot achieve intelligent parameter prediction and rapid response, which seriously affects the practicality of the video surveillance system.

[0004] Therefore, how to establish an efficient parameter mapping mechanism, achieve precise adjustment of the moving target area, and improve the system's ability to respond quickly to changes in illumination are technical challenges that need to be solved urgently. This is not only related to the image quality of video surveillance, but also the key to improving the practicality of intelligent monitoring systems. Summary of the invention

[0005] In response to the problems in the prior art, the present application provides a method for quickly adjusting video images based on a deep learning neural network, which can improve the image quality in complex lighting environments.

[0006] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides a method for fast adjusting video images based on a deep learning neural network, comprising:

[0008] A parameter mapping database is constructed based on the image adjustment process of parameter feedback control, image sequences are collected in indoor low light, outdoor strong light, and night low light scenes, and parameter change trajectories are recorded, the brightness distribution of the moving target area and the target edge contrast are extracted as image quality evaluation indicators, the corresponding relationship between the image sequence and the final parameters is established, and the final parameters are normalized to obtain a standardized parameter set;

[0009] A residual neural network pruning evaluation mechanism is constructed to analyze the brightness change characteristics of the moving target area, calculate the feature extraction sensitivity of each layer of the network to the moving target area, use the feature extraction sensitivity as the neuron importance score, retain the key neurons for brightness adjustment of the moving target area according to the importance score, use a large-size convolution kernel to enhance the response ability to sudden changes in illumination, introduce a channel attention mechanism in the residual module to highlight the characteristics of the target area, train the pruned network structure based on the image sequence and the standardized parameter set, and design a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area;

[0010] A parameter sensitivity evaluation matrix is ​​constructed to calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, and the parameter adjustment priority is set according to the influence degree. A smooth transition strategy is designed using the parameter adjustment priority, and the output result of the pruned network structure is used through the smooth transition strategy to achieve real-time image parameter adjustment.

[0011] Furthermore, the image adjustment process based on parameter feedback control constructs a parameter mapping database, collects image sequences in indoor low light, outdoor strong light, and night low light scenes and records parameter change trajectories, extracts the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishes a corresponding relationship between the image sequence and the final parameters, and normalizes the final parameters to obtain a standardized parameter set, including:

[0012] Based on the parameter controller, the feedback control curve is obtained, the exposure duration parameter is adjusted in indoor low-light scenes, the exposure gain parameter is adjusted in outdoor strong-light scenes, and the white balance gain parameter is adjusted in night low-light scenes. For each set of parameters, image frames are collected and the parameter combination is recorded. The background image of the video sequence is calculated using a mixed Gaussian model, the foreground area is obtained by background difference, and the moving target area is obtained using connected domain marking.

[0013] The brightness histogram of the moving target area is calculated to obtain the brightness mean of the target area, the Sobel operator is used to calculate the edge gradient value of the target area, the image quality is evaluated based on the brightness mean and the gradient value, the parameter combination with the best evaluation index is selected as the final parameter, the final parameter is normalized to obtain a standardized parameter set, and a corresponding database of image features and standardized parameter sets is established.

[0014] Furthermore, the residual neural network pruning evaluation mechanism is constructed to analyze the brightness change characteristics of the moving target area and calculate the sensitivity of each layer of the network to the feature extraction of the moving target area, including:

[0015] The moving target area is input into the residual neural network, and the feature map output by each layer is recorded during the forward propagation of the network. A small perturbation is performed on the feature map to obtain a perturbed feature map. The output difference between the feature map and the perturbed feature map is calculated, and the sensitivity of each layer of neurons to the feature extraction of the moving target area is calculated based on the output difference.

[0016] The target area is divided into grid blocks, the brightness mean of each grid block is extracted to construct a brightness change feature vector, the response degree of the brightness change feature vector and the feature maps of each layer of the network is calculated, and the response degree is used as the feature extraction sensitivity of the neuron to the moving target area.

[0017] Furthermore, the feature extraction sensitivity is used as a neuron importance score, and the key neurons for adjusting the brightness of the moving target area are retained according to the importance score, a large-size convolution kernel is used to enhance the response ability to sudden changes in illumination, and a channel attention mechanism is introduced in the residual module to highlight the characteristics of the target area, including:

[0018] The neurons in each layer of the network are sorted according to the sensitivity of feature extraction, and a retention ratio threshold is set to remove neurons with a sensitivity lower than the threshold. A 7×7 convolution kernel is used to replace the original 3×3 convolution kernel, and the network's feature extraction capability for illumination changes in the target area is enhanced by expanding the receptive field.

[0019] In the residual module, the global average pooling of the feature map is calculated to obtain the channel descriptor, and the correlation between channels is learned using a two-layer fully connected network to generate channel attention weights. The channel attention weights are multiplied by the feature map to obtain a weighted feature map, and the weighted feature map is added to the input feature through a jump connection.

[0020] Furthermore, the training of the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area, includes:

[0021] Input the image sequence into the pruned network structure to obtain prediction parameters, calculate the mean square error between the prediction parameters and the standardized parameter set as the basic loss, assign a weight coefficient of 1.5 to the moving target area, assign a weight coefficient of 0.5 to the background area, and multiply the basic loss by the weight coefficient to obtain a weighted loss function;

[0022] The stochastic gradient descent method is used to optimize the network parameters, the learning rate is set to 0.001, the gradient of the weighted loss function is calculated in each iteration, the network weights are updated according to the gradient, and the training is stopped when the loss function converges or reaches a preset number of iterations.

[0023] Furthermore, constructing a parameter sensitivity evaluation matrix, calculating the influence of different parameters on the brightness distribution of the moving target area and the contrast of the target edge, and setting the parameter adjustment priority according to the influence degree, includes:

[0024] Construct a parameter sensitivity matrix, where the rows of the matrix represent the exposure time, exposure gain, and white balance gain parameters, and the columns of the matrix represent the brightness mean and edge gradient value of the moving target area. Calculate the change rate of the parameter change to the evaluation index through parameter perturbation, and fill the change rate into the corresponding matrix position;

[0025] The influence coefficient of each parameter is calculated based on the parameter sensitivity matrix. The influence coefficient is the numerical average of the row corresponding to the parameter. The parameters are sorted in descending order according to the influence coefficient, and the sorting result is used as the priority order of parameter adjustment.

[0026] Furthermore, the smooth transition strategy is designed by using the parameter adjustment priority, and the output result of the pruned network structure is adjusted in real time using the smooth transition strategy, including:

[0027] The parameter update time interval is divided based on the parameter adjustment priority. The parameters with high priority are updated once per frame, the parameters with medium priority are updated once every two frames, and the parameters with low priority are updated once every three frames. The prediction parameters output by the network are adjusted in sequence according to the update time interval.

[0028] The difference between the current parameter and the predicted parameter is calculated, and the difference is divided by the preset number of frames to obtain the single-frame parameter adjustment step. The parameter value is adjusted frame by frame according to the single-frame parameter adjustment step until the predicted parameter value is reached to complete the parameter smooth transition.

[0029] In a second aspect, the present application provides a video image fast adjustment device based on a deep learning neural network, comprising:

[0030] A preprocessing module is used to build a parameter mapping database based on the image adjustment process of parameter feedback control, collect image sequences in indoor low light, outdoor strong light, and night low light scenes and record parameter change trajectories, extract the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establish a corresponding relationship between the image sequence and the final parameters, and normalize the final parameters to obtain a standardized parameter set;

[0031] An image analysis module is used to construct a residual neural network pruning evaluation mechanism, analyze the brightness change characteristics of the moving target area, calculate the feature extraction sensitivity of each layer of the network to the moving target area, use the feature extraction sensitivity as the neuron importance score, retain the key neurons for brightness adjustment of the moving target area according to the importance score, use a large-size convolution kernel to enhance the response ability to sudden changes in illumination, introduce a channel attention mechanism in the residual module to highlight the characteristics of the target area, train the pruned network structure based on the image sequence and the standardized parameter set, and design a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area;

[0032] The image adjustment module is used to construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the contrast of the target edge, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0033] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the method for rapid adjustment of video images based on a deep learning neural network are implemented.

[0034] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for rapid adjustment of video images based on a deep learning neural network.

[0035] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the method for rapid adjustment of video images based on a deep learning neural network.

[0036] It can be seen from the above technical solution that the present application provides a method for rapid adjustment of video images based on a deep learning neural network. By constructing a parameter mapping database for different lighting scenes, an image quality evaluation system is established based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and a channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is one of the flow charts of the method for rapid adjustment of video images based on deep learning neural network in the embodiment of the present application;

[0039] Figure 2 This is a second flow chart of a method for rapid adjustment of video images based on a deep learning neural network in an embodiment of the present application;

[0040] Figure 3 This is a flowchart of a method for quickly adjusting video images based on a deep learning neural network in an embodiment of the present application;

[0041] Figure 4 This is a fourth flow chart of a method for rapid adjustment of video images based on a deep learning neural network in an embodiment of the present application;

[0042] Figure 5 This is a fifth flow chart of a method for rapid adjustment of video images based on a deep learning neural network in an embodiment of the present application;

[0043] Figure 6 This is a sixth flow chart of a method for rapid adjustment of video images based on a deep learning neural network in an embodiment of the present application;

[0044] Figure 7 This is a flowchart of the method for rapid adjustment of video images based on a deep learning neural network in an embodiment of the present application;

[0045] Figure 8 This is a structural diagram of a video image rapid adjustment device based on a deep learning neural network in an embodiment of the present application;

[0046] Fig. 9 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application.

[0047] Reference numerals:

[0048] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0051] Taking into account the problems existing in the prior art, the present application provides a method for rapid adjustment of video images based on a deep learning neural network. By constructing a parameter mapping database for different lighting scenes, an image quality evaluation system is established based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and a channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0052] In order to improve the image quality in complex lighting environments, the present application provides an embodiment of a method for fast adjustment of video images based on a deep learning neural network. Figure 1 The method for fast adjusting video images based on deep learning neural network specifically includes the following contents:

[0053] Step S101: constructing a parameter mapping database based on the image adjustment process of parameter feedback control, collecting image sequences in indoor low light, outdoor strong light, and night low light scenes and recording parameter change trajectories, extracting the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishing a corresponding relationship between the image sequence and the final parameters, and normalizing the final parameters to obtain a standardized parameter set;

[0054] Optionally, this embodiment proposes a method for adaptively adjusting image quality based on parameter mapping, which realizes intelligent optimization of image quality by establishing a multi-scene parameter database.

[0055] This embodiment first designs a parameter feedback control strategy for different lighting scenarios. In indoor low-light environments, the image brightness is mainly improved by adjusting the exposure duration parameter; in outdoor strong-light scenes, the exposure gain parameter is focused on suppressing overexposure; in low-light conditions at night, the white balance gain parameter is optimized to improve the color reproduction effect. These parameters are adjusted in real time through the parameter controller, and the image sequence and corresponding parameter combination during the parameter change process are recorded.

[0056] In terms of target region extraction, this embodiment uses a mixed Gaussian model to construct a scene background model. Through statistical analysis of multiple consecutive frames of images, the pixel distribution characteristics of the static background area are accurately identified. The foreground moving target is extracted using the background difference method, and the complete target region outline is obtained through the connected domain labeling algorithm.

[0057] This embodiment innovatively designs a dual-index evaluation mechanism. By calculating the brightness histogram of the target area, the brightness distribution characteristics are obtained; at the same time, the Sobel operator is used to extract the target edge gradient information and evaluate the clarity of the target contour. The combination of these two indicators ensures both the appropriate brightness of the target area and the clarity of the target edge.

[0058] In the parameter optimization process, this embodiment implements an adaptive parameter search strategy, by continuously adjusting the parameter combination and evaluating the changing trend of the image quality index, and finally screening out the optimal parameter combination that makes the brightness distribution of the target area uniform and the edge clear.

[0059] The parameter normalization process uses a piecewise linear mapping method. According to the effective value range of different parameters, the parameter values ​​are mapped to a unified standard range, which is convenient for the subsequent training and optimization of the deep learning network. This standardization process improves the generalization ability of parameter mapping.

[0060] This embodiment is also optimized for different scene characteristics in practical applications. For example, in indoor low-light environments, priority is given to ensuring the brightness uniformity of the target area; in outdoor strong-light scenes, the focus is on controlling the detail retention of the highlight area; in low-light conditions at night, the focus is on balancing the relationship between brightness enhancement and noise suppression.

[0061] By establishing a complete parameter mapping database, this embodiment solves the problem that traditional fixed parameter solutions are difficult to adapt to complex lighting environments. This solution can intelligently adjust image parameters according to the characteristics of different scenes, significantly improving the imaging quality of moving targets.

[0062] The parameter mapping mechanism of this embodiment provides reliable training data for the subsequent deep learning network. By accumulating a large amount of parameter optimization experience in real scenarios, the neural network can accurately learn the optimal parameter configuration rules under different scenario conditions.

[0063] In practical applications, this embodiment demonstrates strong environmental adaptability. Whether in indoor low-light environments, outdoor high-light scenes, or low-light conditions at night, it can quickly find the appropriate parameter configuration to ensure the clarity and recognizability of moving target imaging, providing high-quality image data for subsequent video analysis and processing.

[0064] Step S102: constructing a residual neural network pruning evaluation mechanism, analyzing the brightness change characteristics of the moving target area, calculating the feature extraction sensitivity of each network layer to the moving target area, using the feature extraction sensitivity as a neuron importance score, retaining key neurons for brightness adjustment of the moving target area according to the importance score, using a large-size convolution kernel to enhance the response capability to sudden changes in illumination, introducing a channel attention mechanism in the residual module to highlight the characteristics of the target area, training the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function to make the brightness distribution weight coefficient of the moving target area greater than that of the background area;

[0065] Optionally, this embodiment proposes an adaptive pruning method based on a residual network, which realizes intelligent optimization of the network structure by analyzing the brightness characteristics of the target area.

[0066] This embodiment first constructs a feature sensitivity evaluation mechanism. After the moving target area is input into the residual neural network, the feature map output of each layer is recorded during the forward propagation process. By applying random perturbations to the feature map and analyzing the degree of change in the output response, the sensitivity of each layer of neurons to the feature extraction of the target area is quantified.

[0067] In terms of target area brightness feature extraction, this embodiment adopts a grid division strategy. The target area is evenly divided into multiple grid blocks, and the brightness mean of each grid block is extracted to construct a feature vector. By calculating the response relationship between the feature vector and the feature map of each layer of the network, the sensitivity of the neuron to the brightness change of the target area is further evaluated.

[0068] This embodiment innovatively designs a sensitivity-based pruning mechanism. The feature extraction sensitivity is used as an evaluation index of the importance of neurons, and a reasonable retention ratio threshold is set to give priority to neurons that contribute most to the brightness adjustment of the target area, effectively reducing network redundancy.

[0069] In terms of network structure optimization, this embodiment adopts a large-size convolution kernel replacement strategy. The original 3×3 convolution kernel is updated to a 7×7 convolution kernel, which significantly expands the receptive field range and enhances the network's feature extraction capability for drastic changes in illumination.

[0070] This embodiment introduces a channel attention mechanism in the residual module. The channel feature description is extracted through global average pooling, and the correlation between channels is learned using a two-layer fully connected network to generate discriminative channel weights. The channel weights are multiplied by the feature map to highlight the key features of the target area.

[0071] In the network training phase, this embodiment designs a weighted loss function. Different weight coefficients are assigned to the moving target area and the background area to ensure that the network pays more attention to the brightness distribution optimization of the target area. At the same time, the original feature information is maintained through the residual connection, which improves the training stability of the network.

[0072] This embodiment is also optimized for different scene characteristics. For example, in indoor low-light environments, the network focuses more on extracting dark detail features; in outdoor strong-light scenes, it focuses on suppressing overexposure in highlight areas; and in low-light conditions at night, it focuses on analyzing noise distribution characteristics.

[0073] Through the feature sensitivity driven network pruning strategy, this embodiment solves the problem of computational redundancy in traditional fixed structure networks. This solution can adaptively optimize the network structure according to actual scenario requirements, significantly improving the real-time performance of parameter adjustment.

[0074] The network optimization mechanism of this embodiment provides an efficient processing framework for practical applications. By retaining key neurons, expanding the receptive field range, introducing attention mechanisms and other innovative designs, the network can quickly respond to changes in illumination and accurately adjust image parameters.

[0075] In practical applications, this embodiment shows excellent environmental adaptability. The optimized network structure can quickly capture the brightness change characteristics of the target area and output appropriate parameter configuration in real time, ensuring clear imaging of moving targets under different lighting conditions.

[0076] Step S103: construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0077] Optionally, this embodiment proposes a sensitivity-based parameter priority adjustment method, which achieves smooth adjustment of image quality by evaluating the degree of parameter influence.

[0078] This embodiment first establishes a parameter sensitivity evaluation matrix. For key parameters such as exposure time, gain, and white balance, small changes are made within a certain range, and the changes in brightness distribution and edge contrast of the moving target area are recorded. By calculating the ratio of the parameter change to the image quality index change, the sensitivity of each parameter to the image quality is quantified.

[0079] In terms of brightness distribution evaluation, this embodiment adopts a regional statistical method. The target area is divided into multiple sub-areas, the brightness mean and variance of each sub-area are calculated, and a brightness distribution feature vector is constructed. By analyzing the difference in the brightness distribution feature vector before and after the parameter changes, the influence of the parameter on the brightness uniformity is evaluated.

[0080] This embodiment innovatively introduces multi-scale analysis in edge contrast evaluation. Sobel operators of different scales are used to extract target edge gradient information and calculate the contrast value of edge pixels. By comparing the edge clarity changes before and after parameter adjustment, the influence of the parameters on the target contour preservation is quantified.

[0081] Based on the sensitivity evaluation results, this embodiment designs a parameter adjustment priority mechanism. The parameters are sorted according to the degree of influence, and the parameters that have a significant impact on the image quality are adjusted first. For example, in an indoor low-light environment, the exposure time adjustment priority is higher; in an outdoor strong light scene, the gain parameter adjustment priority is higher.

[0082] This embodiment innovatively proposes a smooth transition strategy. By designing a time window mechanism, parameter gradient constraints are introduced between consecutive frames. Different adjustment steps are set for parameters of different priorities. High-priority parameters are allowed to have a larger adjustment range, while low-priority parameters are fine-tuned with a small step size to avoid image flickering caused by sudden changes in parameters.

[0083] In the actual adjustment process, this embodiment realizes adaptive smooth control. When the scene illumination changes slowly, a smaller adjustment step is used to achieve a smooth transition; when the illumination changes suddenly, the adjustment step of the high-priority parameter is appropriately increased to speed up the response speed while maintaining the smooth change of the low-priority parameter.

[0084] This embodiment is optimized for different scene characteristics. In indoor low-light environments, the focus is on the smoothness of brightness enhancement; in outdoor strong-light scenes, the focus is on controlling the stability of overexposure suppression; in low-light conditions at night, the balance between noise reduction effect and detail preservation is achieved.

[0085] Through the parameter sensitivity driven priority adjustment strategy, this embodiment solves the problem of slow response of the traditional fixed step adjustment scheme. The scheme can intelligently adjust the parameter change rate according to the characteristics of scene changes, significantly improving the adaptability of image quality adjustment.

[0086] The smooth transition mechanism of this embodiment provides a stable processing effect for practical applications. Through multi-level parameter linkage adjustment, it not only ensures a rapid response to illumination changes, but also avoids image instability caused by drastic parameter changes.

[0087] In practical applications, this embodiment demonstrates excellent adjustment performance. The optimized parameter adjustment scheme can achieve smooth transition while ensuring image quality, and provide stable and reliable image sequences for video acquisition in various lighting scenarios.

[0088] From the above description, it can be seen that the method for rapid video image adjustment based on deep learning neural network provided by the embodiment of the present application can establish an image quality evaluation system based on the brightness distribution and edge contrast of the moving target area by constructing a parameter mapping database for different lighting scenes. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and the channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is adopted to realize real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0089] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 2 , and can also include the following:

[0090] Step S201: Based on the parameter controller, a feedback control curve is obtained, the exposure duration parameter is adjusted in the indoor low-light scene, the exposure gain parameter is adjusted in the outdoor strong-light scene, and the white balance gain parameter is adjusted in the night low-light scene. For each set of parameters, image frames are collected and parameter combinations are recorded. A mixed Gaussian model is used to calculate the background image of the video sequence, the foreground area is obtained by background difference, and the moving target area is obtained by connected domain marking.

[0091] Step S202: Calculate the brightness histogram of the moving target area to obtain the brightness mean of the target area, use the Sobel operator to calculate the edge gradient value of the target area, evaluate the image quality based on the brightness mean and the gradient value, select the parameter combination with the best evaluation index as the final parameter, normalize the final parameter to obtain a standardized parameter set, and establish a corresponding database of image features and standardized parameter sets.

[0092] Optionally, this embodiment proposes an image quality assessment method based on parameter feedback control, which realizes the construction of a mapping relationship between image features and parameters through adaptive parameter adjustment and target quality evaluation.

[0093] This embodiment first designs a parameter control strategy according to the characteristics of different scenes. In indoor low-light environments, the overall brightness of the image is mainly improved by increasing the exposure time. The controller uses a slow start and fast stop adjustment curve to avoid overexposure. In outdoor strong light scenes, the exposure gain is adjusted to suppress overexposure. The controller uses a fast response curve to reduce the gain value in time. Under low-light conditions at night, the white balance gain is optimized to improve color reproduction. The controller uses a progressive adjustment curve to maintain color temperature balance.

[0094] In terms of target detection, this embodiment uses a mixed Gaussian model to construct a background model. Multiple Gaussian distributions are established for each pixel, and the weight update mechanism is used to adapt to scene changes. The current frame is differentially operated with the background model to extract the foreground target area. After eliminating noise through morphological processing, the connected domain labeling algorithm is used to obtain the complete moving target area.

[0095] This embodiment innovatively designs a dual quality evaluation mechanism. First, the brightness histogram of the target area is calculated to analyze the concentration and uniformity of the brightness distribution. At the same time, the Sobel operator is used to extract edge gradient information in the horizontal and vertical directions to evaluate the clarity of the target contour. The combination of these two indicators ensures both the appropriate brightness level of the target area and the clear visibility of the edge details.

[0096] In the parameter optimization process, this embodiment implements an adaptive parameter search strategy, by continuously adjusting the parameter combination and evaluating the change trend of the brightness mean and edge gradient value, and finally screening out the parameter configuration that optimizes the imaging quality of the target area.

[0097] The parameter normalization process uses a piecewise linear mapping method. According to the effective value range of different parameters, the corresponding mapping function is designed to normalize the parameter values ​​to a unified range. This standardization process improves the versatility of parameter mapping and facilitates the subsequent training of deep learning networks.

[0098] This embodiment is also optimized for different scene characteristics. For example, in indoor low-light environments, the controller pays more attention to the stability of brightness increase; in outdoor strong-light scenes, it focuses on controlling the detail retention of highlight areas; in low-light conditions at night, it balances the relationship between noise reduction effect and brightness increase.

[0099] By establishing a complete parameter mapping database, this embodiment solves the problem that traditional fixed parameter solutions are difficult to adapt to complex lighting environments. This solution can intelligently adjust image parameters according to the characteristics of different scenes, significantly improving the imaging quality of moving targets.

[0100] The parameter optimization mechanism of this embodiment provides reliable training data for the subsequent deep learning network. By accumulating a large amount of parameter optimization experience in real scenarios, the neural network can accurately learn the optimal parameter configuration rules under different scenario conditions.

[0101] In practical applications, this embodiment demonstrates strong environmental adaptability. Whether in indoor low-light environments, outdoor high-light scenes, or low-light conditions at night, it can quickly find the appropriate parameter configuration to ensure the clarity and recognizability of moving target imaging, providing high-quality image data for subsequent video analysis and processing.

[0102] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 3 , and can also include the following:

[0103] Step S301: input the moving target region into the residual neural network, record the feature map output by each layer during the network forward propagation process, perform a small perturbation on the feature map to obtain a perturbed feature map, calculate the output difference between the feature map and the perturbed feature map, and calculate the sensitivity of each layer of neurons to the feature extraction of the moving target region based on the output difference;

[0104] Step S302: Divide the target area into grid blocks, extract the brightness mean of each grid block to construct a brightness change feature vector, calculate the response degree of the brightness change feature vector and the feature map of each layer of the network, and use the response degree as the feature extraction sensitivity of the neuron to the moving target area.

[0105] Optionally, this embodiment proposes a neural network optimization method based on feature sensitivity analysis, which achieves accurate pruning of the network structure by evaluating the response characteristics of neurons to the target area.

[0106] This embodiment first designs a dual-path feature sensitivity evaluation mechanism. The first path applies random Gaussian noise to the feature map of each layer during the network forward propagation through feature perturbation analysis to generate a perturbation feature map. By calculating the response difference between the original feature map and the perturbation feature map at the output of the network, the sensitivity of each layer of neurons to the target feature extraction is quantified. The perturbation amplitude is selected using an adaptive strategy to dynamically adjust the perturbation intensity according to the numerical distribution of the feature map.

[0107] In terms of feature response analysis, this embodiment innovatively introduces a grid processing strategy. The moving target area is divided into multiple grid blocks of fixed size, and each grid block independently extracts brightness statistical features. By calculating the brightness mean of the grid blocks, a feature vector reflecting the brightness distribution change of the target area is constructed. This localized feature extraction method can more finely characterize the brightness change characteristics of the target area.

[0108] This embodiment proposes a feature-response mapping analysis method in the second path. The constructed brightness change feature vector is used as a reference benchmark to calculate the correlation with the feature map of each layer of the network. By analyzing the response sensitivity of the feature map to brightness changes, the importance of neurons in the feature extraction of the target area is further evaluated.

[0109] In the process of grid feature extraction, this embodiment adopts a multi-scale analysis strategy. By setting grid division schemes of different sizes, the brightness change characteristics of the target area are captured from coarse to fine. Larger grid blocks reflect the overall brightness distribution trend, while smaller grid blocks retain local detail information.

[0110] This embodiment innovatively designs a method for calculating feature responsiveness. By calculating the correlation coefficient between the brightness feature vector and the feature map, the sensitivity of each layer of neurons to the brightness change of the target area is evaluated. The responsiveness calculation takes into account the spatial position information and assigns higher weights to the key parts of the target area.

[0111] In practical applications, this embodiment is optimized for different scene characteristics. For example, in indoor low-light environments, more emphasis is placed on evaluating the ability of neurons to extract dark details; in outdoor strong-light scenes, the focus is on analyzing the response characteristics of neurons to highlight areas; and in low-light conditions at night, the focus is on evaluating the noise reduction performance of neurons.

[0112] Through dual-path feature sensitivity analysis, this embodiment solves the problem that traditional network pruning methods are difficult to accurately evaluate the importance of neurons. This solution can comprehensively evaluate the contribution of neurons from two dimensions: feature perturbation and response mapping, providing a reliable basis for subsequent network optimization.

[0113] The feature evaluation mechanism of this embodiment provides an accurate importance metric for network pruning. By combining perturbation analysis and response mapping, the robustness of neurons is considered and their ability to extract target features is evaluated.

[0114] In practical applications, this embodiment shows excellent evaluation accuracy. The optimized feature extraction network can accurately identify key neurons that significantly affect the image quality of the target area, provide reliable feature support for subsequent parameter adjustment, and ensure the image quality of moving targets under different lighting conditions.

[0115] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 4 , and can also include the following:

[0116] Step S401: sorting neurons in each layer of the network according to feature extraction sensitivity, setting a retention ratio threshold, removing neurons with sensitivity lower than the threshold, replacing the original 3×3 convolution kernel with a 7×7 convolution kernel, and enhancing the network's feature extraction capability for illumination changes in the target area by expanding the receptive field;

[0117] Step S402: Calculate the global average pooling of the feature map in the residual module to obtain a channel descriptor, use a two-layer fully connected network to learn the correlation between channels, generate channel attention weights, multiply the channel attention weights with the feature map to obtain a weighted feature map, and add the weighted feature map to the input feature through a jump connection.

[0118] Optionally, this embodiment proposes a sensitivity-based network optimization method, which achieves lightweight network structure and improved feature extraction capability through neuron pruning and attention enhancement.

[0119] This embodiment first designs an adaptive pruning strategy based on feature extraction sensitivity. The sensitivity of neurons in each layer of the network is normalized and sorted, and a dynamic threshold is set to determine the retention ratio. The selection of the threshold takes into account the feature extraction tasks at different levels. The shallow network focuses on retaining basic feature extraction units such as edges and textures, while the deep network focuses more on retaining semantic feature extraction units.

[0120] In terms of convolution kernel optimization, this embodiment innovatively adopts a large convolution kernel replacement strategy. The original 3×3 convolution kernel is replaced with a 7×7 convolution kernel, which significantly expands the receptive field. The larger receptive field enables the network to simultaneously focus on the illumination change characteristics of the target area and its surrounding environment, improving the network's ability to adapt to local illumination unevenness.

[0121] This embodiment introduces a channel attention mechanism in the residual module. First, the spatial dimension is compressed by global average pooling to obtain a channel descriptor that reflects the feature distribution of each channel. A two-layer fully connected network is designed to learn the dependencies between channels, where the first layer performs feature dimensionality reduction to reduce the amount of calculation, and the second layer restores the dimension and generates channel weights.

[0122] In the process of generating attention weights, this embodiment uses a nonlinear activation function to enhance feature expression. The weight value is normalized to between 0 and 1 through the sigmoid function to ensure the stability of the attention mechanism. The product operation of the weight and the feature map adaptively adjusts the feature response intensity of each channel to highlight the feature channels that are more important for the evaluation of the light quality of the target area.

[0123] This embodiment innovatively designs a residual connection structure. The weighted feature map is added element by element to the input feature, which not only retains the original feature information, but also introduces the feature enhancement effect brought by the channel attention mechanism. This skip connection structure effectively alleviates the gradient vanishing problem of deep networks.

[0124] In practical applications, this embodiment is optimized for different lighting scenarios. In indoor low-light environments, the attention mechanism pays more attention to feature channels related to brightness improvement; in outdoor strong-light scenes, it focuses on enhancing feature responses related to contrast preservation; in low-light conditions at night, it focuses on strengthening feature expressions related to noise reduction and detail preservation.

[0125] Through sensitivity-driven network optimization, this embodiment solves the problem of redundant traditional network structure and limited feature extraction capability. This solution can accurately identify and retain key neurons, while enhancing feature expression through the attention mechanism, significantly improving the feature extraction efficiency of the network.

[0126] The structural optimization mechanism of this embodiment provides an efficient and reliable processing framework for practical applications. The computational complexity is reduced through neuron pruning, and the feature extraction accuracy is improved through attention enhancement, achieving dual optimization of network performance and efficiency.

[0127] In practical applications, this embodiment shows excellent adaptability. The optimized network structure can quickly respond to scene illumination changes, accurately extract quality-related features of moving target areas, provide reliable feature support for subsequent parameter adjustments, and ensure the processing quality of video images under different illumination conditions.

[0128] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 5 , and can also include the following:

[0129] Step S501: input the image sequence into the pruned network structure to obtain prediction parameters, calculate the mean square error between the prediction parameters and the standardized parameter set as the basic loss, assign a weight coefficient of 1.5 to the moving target area, assign a weight coefficient of 0.5 to the background area, and multiply the basic loss by the weight coefficient to obtain a weighted loss function;

[0130] Step S502: Use the stochastic gradient descent method to optimize the network parameters, set the learning rate to 0.001, calculate the gradient of the weighted loss function in each iteration, update the network weights according to the gradient, and stop training when the loss function converges or reaches a preset number of iterations.

[0131] Optionally, this embodiment proposes a network training method based on weighted loss, which achieves precise control of the image quality of the moving target area through a differentiated loss calculation strategy.

[0132] This embodiment first constructs a training framework for the parameter prediction network. The image sequence is input into the pruned and optimized network structure to obtain parameter prediction values ​​for the current scene. The prediction parameters include key imaging parameters such as exposure time, exposure gain, and white balance gain. The mean square error is calculated with the standardized parameter set established in the early stage, and the basic loss function is constructed to provide an optimization target for network training.

[0133] In terms of loss calculation, this embodiment innovatively introduces a regional weighting mechanism. Considering that the moving target area is the focus area of ​​image quality assessment, a weight coefficient of 1.5 times is assigned to it; relatively speaking, the image quality requirements of the background area are lower, and a weight coefficient of 0.5 times is assigned. This differentiated weight allocation strategy enables the network to pay more attention to parameter optimization of the moving target area during training.

[0134] This embodiment uses the stochastic gradient descent method for network optimization. The learning rate is set to 0.001 to ensure a faster convergence speed while ensuring training stability. In each iteration, the gradient of the weighted loss function to the network parameters is first calculated, and then the network weights are updated according to the gradient direction to gradually reduce the difference between the predicted parameters and the standard parameters.

[0135] During the training process, this embodiment designs an adaptive learning rate adjustment strategy. When the loss decreases slowly over multiple rounds of iterations, the learning rate is appropriately reduced to achieve more precise parameter adjustment; when the loss fluctuates, the learning rate is temporarily increased to jump out of the local optimal solution.

[0136] This embodiment adopts differentiated training strategies in different lighting scenarios. When training in indoor low-light environments, more attention is paid to the optimization of exposure duration parameters; when training in outdoor strong-light scenes, the focus is on the adjustment of exposure gain; when training in low-light conditions at night, the focus is on optimizing white balance gain parameters.

[0137] In order to improve the training efficiency, this embodiment adopts a batch training mechanism. Each batch randomly selects image sequences under different lighting conditions for training to enhance the network's ability to adapt to scene changes. At the same time, data enhancement technology is introduced to expand the training samples through random cropping, rotation and other operations to improve the generalization performance of the network.

[0138] Through weighted loss driven network training, this embodiment solves the problem that traditional training methods are difficult to balance the image quality of the target area and the background area. This solution can accurately control the parameter optimization degree of the moving target area and significantly improve the image quality of the target area.

[0139] The training mechanism of this embodiment provides a reliable parameter prediction model for practical applications. Through differentiated loss calculation and adaptive optimization strategy, the directional improvement of network performance is achieved, ensuring the priority of image quality in the moving target area.

[0140] In practical applications, this embodiment demonstrates excellent parameter prediction capabilities. The trained network can quickly predict the appropriate imaging parameter combination based on the scene characteristics of the input image sequence, achieving clear imaging of moving targets under different lighting conditions, and providing high-quality image data support for subsequent visual analysis tasks.

[0141] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 6 , and can also include the following:

[0142] Step S601: construct a parameter sensitivity matrix, where the rows of the matrix represent the exposure time, exposure gain and white balance gain parameters, and the columns of the matrix represent the brightness mean and edge gradient value of the moving target area. The change rate of the parameter change to the evaluation index is calculated by parameter perturbation, and the change rate is filled into the corresponding matrix position;

[0143] Step S602: Calculate the influence coefficient of each parameter based on the parameter sensitivity matrix, where the influence coefficient is the numerical average of the row corresponding to the parameter, sort the parameters in descending order according to the influence coefficient, and use the sorting result as the priority order of parameter adjustment.

[0144] Optionally, this embodiment proposes a parameter optimization method based on sensitivity analysis, which realizes precise control of imaging parameters by constructing a parameter-index response relationship matrix.

[0145] This embodiment first designs a method for constructing a parameter sensitivity matrix. The row vectors of the matrix correspond to key imaging parameters, including exposure time, exposure gain, and white balance gain; the column vectors correspond to quality evaluation indicators of the target area, including brightness mean and edge gradient value. By applying a small disturbance to each parameter, the change response of the evaluation index is calculated, thereby quantifying the degree of influence of parameter adjustment on image quality.

[0146] In terms of parameter perturbation analysis, this embodiment adopts an adaptive step size strategy. The perturbation amplitude is dynamically adjusted according to the current value of the parameter. A larger perturbation step size is used for a larger parameter value, and a smaller perturbation step size is used for a smaller parameter value to ensure the accuracy of the change rate calculation. For different lighting scenes, the perturbation range is also adjusted accordingly to adapt to the parameter sensitivity characteristics under different imaging conditions.

[0147] This embodiment innovatively introduces a multi-index evaluation mechanism. The brightness mean reflects the overall exposure level of the target area, and the edge gradient value represents the clarity of the target outline. By considering these two indicators at the same time, the impact of parameter adjustment on image quality is comprehensively evaluated. When calculating the rate of change, normalization is used to eliminate the impact of different indicator dimensions.

[0148] In the process of matrix construction, this embodiment designs a stability enhancement strategy. Multiple positive and negative perturbations are performed on each parameter, and the average change rate is taken to fill in the matrix to reduce the impact of random noise. At the same time, the coupling effect between parameters is considered, and other parameters are kept unchanged when a single parameter is perturbed to ensure the accuracy of sensitivity analysis.

[0149] This embodiment innovatively designs a parameter influence assessment method based on the sensitivity matrix. The mean of the row corresponding to each parameter is calculated as the influence coefficient, which comprehensively reflects the overall influence of parameter adjustment on multiple evaluation indicators. By sorting the influence coefficients in descending order, a priority sequence for parameter adjustment is established to guide the subsequent parameter optimization process.

[0150] In practical applications, this embodiment is optimized for different scene characteristics. In indoor low-light environments, the focus is on analyzing the sensitive characteristics of exposure time and exposure gain; in outdoor strong-light scenes, the focus is on evaluating the impact of white balance gain; under conditions of drastic changes in lighting, the focus is on the impact of various parameters on image quality stability.

[0151] Through sensitivity-driven parameter analysis, this embodiment solves the problem of strong blindness in traditional parameter adjustment methods. This solution can accurately quantify the effect of parameter adjustment and provide a scientific decision-making basis for parameter optimization.

[0152] The sensitivity analysis mechanism of this embodiment provides reliable parameter adjustment guidance for practical applications. By establishing a parameter-index response relationship, accurate control of parameter adjustment is achieved, and the efficiency of parameter optimization is improved.

[0153] In practical applications, this embodiment shows excellent adaptability. Based on the sensitivity analysis results, the most effective parameter adjustment strategy can be quickly determined, achieving clear imaging of moving targets under different lighting conditions and ensuring the acquisition quality of video images. At the same time, the priority sorting mechanism significantly improves the efficiency of parameter adjustment and reduces computing resource consumption.

[0154] In one embodiment of the video image fast adjustment method based on deep learning neural network of the present application, see Figure 7 , and can also include the following:

[0155] Step S701: dividing the parameter update time interval based on the parameter adjustment priority, updating the high priority parameters once per frame, updating the medium priority parameters once every two frames, and updating the low priority parameters once every three frames, and adjusting the prediction parameters output by the network in sequence according to the update time interval;

[0156] Step S702: Calculate the difference between the current parameter and the predicted parameter, divide the difference by the preset number of frames to obtain the single-frame parameter adjustment step, and adjust the parameter value frame by frame according to the single-frame parameter adjustment step until the predicted parameter value is reached to complete the parameter smooth transition.

[0157] Optionally, this embodiment proposes a priority-based adaptive parameter adjustment method, which achieves precise control of imaging parameters through differentiated update strategies and smooth transition mechanisms.

[0158] This embodiment first designs a multi-level update mechanism based on the parameter adjustment priority obtained in the early stage. According to the degree of influence of the parameters on the image quality, the imaging parameters are divided into three priority levels: high, medium, and low. For high-priority parameters with significant impact, such as exposure time, a frame-by-frame update strategy is adopted to ensure a rapid response to lighting changes; for medium-priority parameters, such as exposure gain, updates are performed every other frame to strike a balance between response speed and computational efficiency; for low-priority parameters, such as white balance gain, updates are performed every two frames to reduce the computational load.

[0159] This embodiment innovatively introduces a timing coordination mechanism during the parameter update process. The update of high-priority parameters is performed before other parameters, which enables subsequent parameters to be adjusted based on the updated high-priority parameter status to avoid mutual interference between parameters. The update times of parameters with different priorities are staggered, which effectively reduces the computational load of a single frame.

[0160] In order to achieve a smooth transition of parameters, this embodiment designs an adaptive step control strategy. First, the difference between the current parameter value and the network prediction value is calculated, and then the difference is evenly distributed according to the preset number of transition frames to obtain the parameter adjustment step size for each frame. This progressive adjustment method avoids image quality fluctuations caused by sudden changes in parameters.

[0161] This embodiment takes into account the scene characteristics when calculating the step size. In scenes with drastic changes in lighting, the number of transition frames is appropriately reduced to speed up parameter adjustment; in environments with relatively stable lighting, the number of transition frames is increased to achieve smoother parameter adjustment. At the same time, the number of transition frames is dynamically adjusted according to the difference between the current value of the parameter and the target value to ensure the adaptability of the adjustment process.

[0162] In actual operation, this embodiment adopts a parameter adjustment progress monitoring mechanism. By recording the cumulative adjustment amount of each parameter, it is ensured that the adjustment process is performed as expected. When external interference occurs and causes the parameter adjustment to deviate from the expected value, the step size of the subsequent frame is adjusted in time to ensure that the predicted parameter value is finally reached.

[0163] This embodiment designs specific adjustment strategies for different application scenarios. In indoor scenarios, the focus is on ensuring the smooth transition of exposure parameters; in outdoor scenarios, the focus is on the dynamic adjustment of white balance parameters; in scenarios with sudden changes in lighting, priority is given to ensuring the rapid response of high-priority parameters.

[0164] Through priority-driven parameter adjustment, this embodiment solves the problems of unreasonable parameter update and unstable adjustment in traditional methods. This solution can perform differentiated updates according to parameter importance, and ensure the stability of image quality through a smooth transition mechanism.

[0165] The parameter adjustment mechanism of this embodiment provides an efficient and reliable solution for practical applications. The multi-level update strategy reduces the computational load, the smooth transition mechanism improves the stability of image quality, and achieves the unity of efficiency and effect of parameter adjustment.

[0166] In practical applications, this embodiment shows excellent adaptability. The differentiated update strategy ensures timely response of key parameters, and the smooth transition mechanism effectively suppresses image quality fluctuations, providing stable parameter support for video acquisition in different scenarios. This solution is particularly suitable for dynamic scenes with changing lighting conditions, and can ensure continuous clear imaging of moving targets.

[0167] In order to improve the image quality in complex lighting environments, the present application provides an embodiment of a video image fast adjustment device based on a deep learning neural network for implementing all or part of the content of the video image fast adjustment method based on a deep learning neural network, see Figure 8 The video image fast adjustment device based on deep learning neural network specifically includes the following contents:

[0168] The preprocessing module 10 is used to build a parameter mapping database based on the image adjustment process of parameter feedback control, collect image sequences in indoor low light, outdoor strong light, and night low light scenes and record parameter change trajectories, extract the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establish the corresponding relationship between the image sequence and the final parameters, and normalize the final parameters to obtain a standardized parameter set;

[0169] The image analysis module 20 is used to construct a residual neural network pruning evaluation mechanism, analyze the brightness change characteristics of the moving target area, calculate the feature extraction sensitivity of each layer of the network to the moving target area, use the feature extraction sensitivity as the neuron importance score, retain the key neurons for brightness adjustment of the moving target area according to the importance score, use a large-size convolution kernel to enhance the response ability to sudden changes in illumination, introduce a channel attention mechanism in the residual module to highlight the characteristics of the target area, train the pruned network structure based on the image sequence and the standardized parameter set, and design a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area;

[0170] The image adjustment module 30 is used to construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the contrast of the target edge, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0171] From the above description, it can be seen that the video image rapid adjustment device based on deep learning neural network provided by the embodiment of the present application can establish an image quality evaluation system based on the brightness distribution and edge contrast of the moving target area by constructing a parameter mapping database for different lighting scenes. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and the channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0172] From a hardware perspective, in order to improve the image quality in a complex lighting environment, the present application provides an embodiment of an electronic device for implementing all or part of the content of the method for fast adjusting video images based on a deep learning neural network, and the electronic device specifically includes the following content:

[0173] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the video image rapid adjustment device based on deep learning neural network and related equipment such as core business system, user terminal and related database; the logic controller can be a desktop computer, tablet computer and mobile terminal, etc., and the present embodiment is not limited thereto. In the present embodiment, the logic controller can be implemented with reference to the embodiment of the video image rapid adjustment method based on deep learning neural network and the embodiment of the video image rapid adjustment device based on deep learning neural network in the embodiment, and the contents thereof are incorporated herein, and the repeated parts are not repeated.

[0174] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0175] In practical applications, part of the method for rapid adjustment of video images based on deep learning neural networks can be executed on the electronic device side as described above, or all operations can be completed in the client device. The selection can be made based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0176] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0177] Fig. 9 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Fig. 9 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Fig. 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0178] In one embodiment, the function of the method for fast adjusting video images based on deep learning neural network can be integrated into the central processing unit 9100. The central processing unit 9100 can be configured to perform the following control:

[0179] Step S101: constructing a parameter mapping database based on the image adjustment process of parameter feedback control, collecting image sequences in indoor low light, outdoor strong light, and night low light scenes and recording parameter change trajectories, extracting the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishing a corresponding relationship between the image sequence and the final parameters, and normalizing the final parameters to obtain a standardized parameter set;

[0180] Step S102: constructing a residual neural network pruning evaluation mechanism, analyzing the brightness change characteristics of the moving target area, calculating the feature extraction sensitivity of each network layer to the moving target area, using the feature extraction sensitivity as a neuron importance score, retaining key neurons for brightness adjustment of the moving target area according to the importance score, using a large-size convolution kernel to enhance the response capability to sudden changes in illumination, introducing a channel attention mechanism in the residual module to highlight the characteristics of the target area, training the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function to make the brightness distribution weight coefficient of the moving target area greater than that of the background area;

[0181] Step S103: construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0182] From the above description, it can be seen that the electronic device provided by the embodiment of the present application builds a parameter mapping database for different lighting scenes, and establishes an image quality evaluation system based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and the channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0183] In another embodiment, the video image rapid adjustment device based on deep learning neural network can be configured separately from the central processing unit 9100. For example, the video image rapid adjustment device based on deep learning neural network can be configured as a chip connected to the central processing unit 9100, and the function of the video image rapid adjustment method based on deep learning neural network can be realized through the control of the central processing unit.

[0184] like Fig. 9 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Fig. 9 In addition, the electronic device 9600 may also include Fig. 9 For components not shown, reference may be made to the prior art.

[0185] like Fig. 9 As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0186] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0187] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0188] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0189] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0190] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0191] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0192] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps of the method for quickly adjusting video images based on a deep learning neural network in the above-mentioned embodiments, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all the steps of the method for quickly adjusting video images based on a deep learning neural network in the above-mentioned embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0193] Step S101: constructing a parameter mapping database based on the image adjustment process of parameter feedback control, collecting image sequences in indoor low light, outdoor strong light, and night low light scenes and recording parameter change trajectories, extracting the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishing a corresponding relationship between the image sequence and the final parameters, and normalizing the final parameters to obtain a standardized parameter set;

[0194] Step S102: constructing a residual neural network pruning evaluation mechanism, analyzing the brightness change characteristics of the moving target area, calculating the feature extraction sensitivity of each network layer to the moving target area, using the feature extraction sensitivity as a neuron importance score, retaining key neurons for brightness adjustment of the moving target area according to the importance score, using a large-size convolution kernel to enhance the response capability to sudden changes in illumination, introducing a channel attention mechanism in the residual module to highlight the characteristics of the target area, training the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function to make the brightness distribution weight coefficient of the moving target area greater than that of the background area;

[0195] Step S103: construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0196] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application builds a parameter mapping database for different lighting scenes, and establishes an image quality evaluation system based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and the channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0197] The embodiments of the present application also provide a computer program product capable of implementing all the steps of the method for quickly adjusting video images based on a deep learning neural network in the above embodiments, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the method for quickly adjusting video images based on a deep learning neural network are implemented. For example, the computer program / instruction implements the following steps:

[0198] Step S101: constructing a parameter mapping database based on the image adjustment process of parameter feedback control, collecting image sequences in indoor low light, outdoor strong light, and night low light scenes and recording parameter change trajectories, extracting the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishing a corresponding relationship between the image sequence and the final parameters, and normalizing the final parameters to obtain a standardized parameter set;

[0199] Step S102: constructing a residual neural network pruning evaluation mechanism, analyzing the brightness change characteristics of the moving target area, calculating the feature extraction sensitivity of each network layer to the moving target area, using the feature extraction sensitivity as a neuron importance score, retaining key neurons for brightness adjustment of the moving target area according to the importance score, using a large-size convolution kernel to enhance the response capability to sudden changes in illumination, introducing a channel attention mechanism in the residual module to highlight the characteristics of the target area, training the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function to make the brightness distribution weight coefficient of the moving target area greater than that of the background area;

[0200] Step S103: construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

[0201] From the above description, it can be seen that the computer program product provided by the embodiment of the present application builds a parameter mapping database for different lighting scenes, and establishes an image quality evaluation system based on the brightness distribution and edge contrast of the moving target area. Through the residual neural network pruning evaluation mechanism, the feature extraction sensitivity of each layer of the network to the target area is calculated, key neurons are retained and the channel attention mechanism is introduced to enhance the response ability to sudden changes in illumination. At the same time, a parameter sensitivity evaluation matrix is ​​constructed, and the adjustment priority is set according to the degree of influence of different parameters on the target area, and a smooth transition strategy is used to achieve real-time parameter adjustment. This method effectively solves the problems of slow response speed and unstable adjustment effect of traditional adjustment schemes, and improves the image quality in complex lighting environments.

[0202] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0203] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0204] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0205] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0206] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A method for fast video image adjustment based on deep learning neural network, characterized in that: The method comprises: A parameter mapping database is constructed based on the image adjustment process of parameter feedback control, image sequences are collected in indoor low light, outdoor strong light, and night low light scenes, and parameter change trajectories are recorded, the brightness distribution of the moving target area and the target edge contrast are extracted as image quality evaluation indicators, the corresponding relationship between the image sequence and the final parameters is established, and the final parameters are normalized to obtain a standardized parameter set; A residual neural network pruning evaluation mechanism is constructed to analyze the brightness change characteristics of the moving target area, calculate the feature extraction sensitivity of each layer of the network to the moving target area, use the feature extraction sensitivity as the neuron importance score, retain the key neurons for brightness adjustment of the moving target area according to the importance score, use a large-size convolution kernel to enhance the response ability to sudden changes in illumination, introduce a channel attention mechanism in the residual module to highlight the characteristics of the target area, train the pruned network structure based on the image sequence and the standardized parameter set, and design a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area; A parameter sensitivity evaluation matrix is ​​constructed to calculate the influence of different parameters on the brightness distribution of the moving target area and the target edge contrast, and the parameter adjustment priority is set according to the influence degree. A smooth transition strategy is designed using the parameter adjustment priority, and the output result of the pruned network structure is used through the smooth transition strategy to achieve real-time image parameter adjustment.

2. The method for fast video image adjustment based on deep learning neural network according to claim 1, characterized in that: The image adjustment process based on parameter feedback control builds a parameter mapping database, collects image sequences in indoor low light, outdoor strong light, and night low light scenes and records parameter change trajectories, extracts the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establishes a corresponding relationship between the image sequence and the final parameters, and normalizes the final parameters to obtain a standardized parameter set, including: Based on the parameter controller, the feedback control curve is obtained, the exposure duration parameter is adjusted in indoor low-light scenes, the exposure gain parameter is adjusted in outdoor strong-light scenes, and the white balance gain parameter is adjusted in night low-light scenes. For each set of parameters, image frames are collected and the parameter combination is recorded. The background image of the video sequence is calculated using a mixed Gaussian model, the foreground area is obtained by background difference, and the moving target area is obtained using connected domain marking. The brightness histogram of the moving target area is calculated to obtain the brightness mean of the target area, the Sobel operator is used to calculate the edge gradient value of the target area, the image quality is evaluated based on the brightness mean and the gradient value, the parameter combination with the best evaluation index is selected as the final parameter, the final parameter is normalized to obtain a standardized parameter set, and a corresponding database of image features and standardized parameter sets is established.

3. The method for fast video image adjustment based on deep learning neural network according to claim 1, characterized in that: The residual neural network pruning evaluation mechanism is constructed, the brightness change characteristics of the moving target area are analyzed, and the sensitivity of each layer of the network to the feature extraction of the moving target area is calculated, including: The moving target area is input into the residual neural network, and the feature map output by each layer is recorded during the forward propagation of the network. A small perturbation is performed on the feature map to obtain a perturbed feature map. The output difference between the feature map and the perturbed feature map is calculated, and the sensitivity of each layer of neurons to the feature extraction of the moving target area is calculated based on the output difference. The target area is divided into grid blocks, the brightness mean of each grid block is extracted to construct a brightness change feature vector, the response degree of the brightness change feature vector and the feature maps of each layer of the network is calculated, and the response degree is used as the feature extraction sensitivity of the neuron to the moving target area.

4. The method for rapid video image adjustment based on deep learning neural network according to claim 1, characterized in that: The feature extraction sensitivity is used as a neuron importance score, and the key neurons for adjusting the brightness of the moving target area are retained according to the importance score. A large-size convolution kernel is used to enhance the response ability to sudden changes in illumination, and a channel attention mechanism is introduced in the residual module to highlight the characteristics of the target area, including: The neurons in each layer of the network are sorted according to the sensitivity of feature extraction, and a retention ratio threshold is set to remove neurons with a sensitivity lower than the threshold. A 7×7 convolution kernel is used to replace the original 3×3 convolution kernel, and the network's feature extraction capability for illumination changes in the target area is enhanced by expanding the receptive field. In the residual module, the global average pooling of the feature map is calculated to obtain the channel descriptor, and the correlation between channels is learned using a two-layer fully connected network to generate channel attention weights. The channel attention weights are multiplied by the feature map to obtain a weighted feature map, and the weighted feature map is added to the input feature through a jump connection.

5. The method for fast video image adjustment based on deep learning neural network according to claim 1, characterized in that: The training of the pruned network structure based on the image sequence and the standardized parameter set, and designing a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area, includes: Input the image sequence into the pruned network structure to obtain prediction parameters, calculate the mean square error between the prediction parameters and the standardized parameter set as the basic loss, assign a weight coefficient of 1.5 to the moving target area, assign a weight coefficient of 0.5 to the background area, and multiply the basic loss by the weight coefficient to obtain a weighted loss function; The stochastic gradient descent method is used to optimize the network parameters, the learning rate is set to 0.001, the gradient of the weighted loss function is calculated in each iteration, the network weights are updated according to the gradient, and the training is stopped when the loss function converges or reaches a preset number of iterations.

6. The method for fast video image adjustment based on deep learning neural network according to claim 1, characterized in that: The step of constructing a parameter sensitivity evaluation matrix, calculating the influence of different parameters on the brightness distribution of the moving target area and the contrast of the target edge, and setting the parameter adjustment priority according to the influence degree, includes: Construct a parameter sensitivity matrix, where the rows of the matrix represent the exposure time, exposure gain, and white balance gain parameters, and the columns of the matrix represent the brightness mean and edge gradient value of the moving target area. Calculate the change rate of the parameter change to the evaluation index through parameter perturbation, and fill the change rate into the corresponding matrix position; The influence coefficient of each parameter is calculated based on the parameter sensitivity matrix. The influence coefficient is the numerical average of the row corresponding to the parameter. The parameters are sorted in descending order according to the influence coefficient, and the sorting result is used as the priority order of parameter adjustment.

7. The method for fast video image adjustment based on deep learning neural network according to claim 1, characterized in that: The method of designing a smooth transition strategy by using the parameter adjustment priority and implementing real-time image parameter adjustment on the output result of the pruned network structure through the smooth transition strategy includes: The parameter update time interval is divided based on the parameter adjustment priority. The parameters with high priority are updated once per frame, the parameters with medium priority are updated once every two frames, and the parameters with low priority are updated once every three frames. The prediction parameters output by the network are adjusted in sequence according to the update time interval. The difference between the current parameter and the predicted parameter is calculated, and the difference is divided by the preset number of frames to obtain the single-frame parameter adjustment step. The parameter value is adjusted frame by frame according to the single-frame parameter adjustment step until the predicted parameter value is reached to complete the parameter smooth transition.

8. A video image fast adjustment device based on deep learning neural network, characterized in that: The device comprises: A preprocessing module is used to build a parameter mapping database based on the image adjustment process of parameter feedback control, collect image sequences in indoor low light, outdoor strong light, and night low light scenes and record parameter change trajectories, extract the brightness distribution of the moving target area and the target edge contrast as image quality evaluation indicators, establish a corresponding relationship between the image sequence and the final parameters, and normalize the final parameters to obtain a standardized parameter set; An image analysis module is used to construct a residual neural network pruning evaluation mechanism, analyze the brightness change characteristics of the moving target area, calculate the feature extraction sensitivity of each layer of the network to the moving target area, use the feature extraction sensitivity as the neuron importance score, retain the key neurons for brightness adjustment of the moving target area according to the importance score, use a large-size convolution kernel to enhance the response ability to sudden changes in illumination, introduce a channel attention mechanism in the residual module to highlight the characteristics of the target area, train the pruned network structure based on the image sequence and the standardized parameter set, and design a weighted loss function so that the brightness distribution weight coefficient of the moving target area is greater than that of the background area; The image adjustment module is used to construct a parameter sensitivity evaluation matrix, calculate the influence of different parameters on the brightness distribution of the moving target area and the contrast of the target edge, set the parameter adjustment priority according to the influence degree, design a smooth transition strategy using the parameter adjustment priority, and use the output result of the pruned network structure through the smooth transition strategy to achieve real-time image parameter adjustment.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the method for rapid adjustment of video images based on deep learning neural network described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for rapid adjustment of video images based on a deep learning neural network as described in any one of claims 1 to 7 are implemented.

Images