Vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence

Abstract

The invention relates to the technical field of image processing, and discloses a vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence, which comprises an ambient light sensing module, a screen content analysis module, a dynamic range acquisition module, an AI enhancement decision module and an image rendering module, through the arrangement of the ambient light sensing module, the screen content analysis module and the dynamic range acquisition module, the dynamic range can be adapted through dual sensing of ambient light and vehicle-mounted screen content, and the ideal dynamic range is corrected based on the semantic mask so as to acquire the optimal dynamic range; the self-adaption of the dynamic range is realized, the problem that fixed parameters cannot adapt to illumination sudden change scenes is solved, meanwhile, a foundation is laid for subsequent image enhancement, and a foundation is laid for subsequent effective reservation of visual priorities of key areas such as the sky and the road surface.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to an artificial intelligence-based image dynamic range adaptive enhancement system for in-vehicle screens. Background Technology

[0002] With the rapid development of technology, in-vehicle displays are being used more and more widely in automobiles. The image quality of in-vehicle displays is of great significance to the driver's visual experience and driving safety. Image quality directly affects driving safety and user experience. However, the in-vehicle environment has the characteristics of drastic changes in lighting. During the day, when the sun is shining directly on the screen, the screen brightness needs to be as high as 1,000 to 1,500 nits to counteract the ambient light, while at night it needs to be reduced to 50 to 100 nits to avoid glare.

[0003] Traditional in-vehicle screens often use fixed gamma correction or simple histogram equalization techniques, which still have some shortcomings. Traditional in-vehicle screens have insufficient dynamic range adaptation, and fixed parameters cannot adapt to scenes with sudden changes in lighting, resulting in overexposure or underexposure. In addition, local contrast is unbalanced, global enhancement algorithms are prone to amplifying noise, and cannot retain the visual priority of key areas such as the sky and the road surface.

[0004] In view of this, the present invention proposes an AI-based in-vehicle screen image dynamic range adaptive enhancement system that can adaptively sense ambient light and screen content in real time. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an artificial intelligence-based in-vehicle screen image dynamic range adaptive enhancement system to solve the problems existing in the background art.

[0006] This invention provides the following technical solution: an AI-based in-vehicle screen image dynamic range adaptive enhancement system, comprising an ambient light perception module, a screen content analysis module, a dynamic range acquisition module, an AI-enhanced decision-making module, and an image rendering module; The ambient light sensing module is used to deploy dual sensors to collect multi-source ambient light data and then perform preprocessing and fusion. The screen content analysis module is used to extract and classify the features of the content of the vehicle screen image by identifying key areas of the vehicle screen image. The dynamic range acquisition module obtains the optimal dynamic range based on data from the ambient light perception module and the screen content analysis module; The AI-enhanced decision-making module generates parameters for a nonlinear brightness mapping function based on the optimal dynamic range, and performs local optimization of the key region using the nonlinear brightness mapping function, while optimizing the non-key region using an enhancement factor. The image rendering module outputs the optimized image to the vehicle screen for display.

[0007] Preferably, the dual sensors include a forward ambient light sensor and a screen surface light sensor, and the preprocessing and fusion specifically involves: Temperature compensation is applied to the data collected by the forward ambient light sensor; the formula for temperature compensation is as follows: ;in, This indicates the external light intensity collected by the forward ambient light sensor after temperature compensation. This represents the original external light intensity collected by the forward ambient light sensor; Indicates the current ambient temperature. Indicates the calibrated temperature; Indicates the temperature coefficient; The overall ambient light intensity is obtained by the weighted average method, expressed by the formula: ;in, Indicates the overall ambient light intensity; This indicates the intensity of reflected light from the screen, collected by the screen surface light sensor. for The weight, for The weights; and satisfy .

[0008] Preferably, the feature extraction and classification of the content of the vehicle screen image specifically involves: The input image is preprocessed; the input image is the image that needs to be displayed on the vehicle screen. If the input image is... The image will Image converted Format, extract the luminance channel ,use The network extracts multi-scale feature maps; after fusing the feature maps through a fully convolutional network, a semantic mask is output. Mark the key regions of the image The set of key region types is represented as .

[0009] Preferably, the dynamic range acquisition module acquires the optimal dynamic range by: Based on the fundamental dynamic range benchmark, the logarithmic relationship between the ideal dynamic range and ambient light is obtained, expressed by the formula: ;in, Indicates the ideal dynamic range; Indicates the reference dynamic range. ; Indicates the reference illumination threshold. ; Based on semantic masking The area proportions of each key region are used to correct the ideal dynamic range, resulting in the corrected target dynamic range. That is, the optimal dynamic range.

[0010] Preferably, the local optimization of the key region using a nonlinear brightness mapping function specifically involves: A conditional generative adversarial network is used, taking the original image, the combined ambient light intensity, and the semantic mask as input, and an enhanced image annotated by experts as input. Train the generator for the goal. The learning mapping relationship is expressed as follows: ; The generator is compressed to 500K parameters through knowledge distillation; Output nonlinear brightness mapping function parameters The Satisfying the formula: ;in, This represents the enhanced image brightness value. This represents the coefficients of the quadratic term in the nonlinear brightness mapping function. This represents the coefficient of the first-order term in the nonlinear brightness mapping function. This represents the constant term of the nonlinear brightness mapping function. Represents the brightness value of the original image; A nonlinear brightness mapping function is used to enhance key areas.

[0011] Preferably, the enhancement factor is the ratio between the optimal dynamic range and the preset dynamic range value; the enhancement factor is expressed by the formula: ;in, Indicates the enhancing factor. Indicates the preset dynamic range value; The optimization of non-critical areas using an enhancement factor is expressed by the following formula: ;in, This represents the optimized pixel values ​​of the non-critical region image; This represents the original image pixel values ​​of non-critical regions after preprocessing. This represents the minimum pixel value of the non-critical region after preprocessing. This represents the maximum pixel value in the non-critical region after preprocessing.

[0012] Preferably, the key region type of the marked image is specifically: An encoder-decoder structure is adopted, where the encoder is responsible for extracting multi-scale features, and the decoder is responsible for restoring the feature map to the original image size and generating a mask; the encoder employs... As an encoder, it extracts multi-scale features from the input image. The decoder progressively upsamples the feature map output by the encoder to the original image size through transposed convolution. Each pixel value output by the decoder corresponds to a region category. The function is normalized to generate a semantic mask. ; Semantic mask After generation, the area ratio of each region is calculated, that is, the ratio of the number of pixels in each region to the total number of pixels, and key regions are filtered according to preset thresholds. Finally, the key region type is generated and stored as a semantic mask.

[0013] Preferably, the area proportion of each region is specifically as follows: Count the number of pixels for each type in the semantic mask and calculate the area percentage: ;in, Indicates key region type The area percentage; Indicates key region type The number of pixels; This represents the total number of pixels in the image.

[0014] Preferably, the generative adversarial network mainly includes a generator, a discriminator, and a loss function; The generator is used to input the integrated ambient light intensity, semantic mask, and luminance channel, and output an enhanced image, including an input layer, a convolutional layer, a residual block, a transposed convolutional layer, and an output layer; the discriminator mainly includes an input layer, a convolutional layer, and a fully connected layer; the loss function consists of adversarial loss and content loss, and the content loss includes perceptual loss and pixel loss.

[0015] Preferably, the loss function is expressed as: ;in, This represents the total loss function; Indicates resistance to loss; Indicates perceived loss; Indicates pixel loss; , , These represent the corresponding weighting coefficients.

[0016] The technical effects and advantages of this invention are as follows: (1) This invention has an ambient light perception module, a screen content analysis module and a dynamic range acquisition module, which is conducive to adapting the dynamic range through dual perception of ambient light and vehicle screen content, and correcting the ideal dynamic range based on semantic mask to obtain the optimal dynamic range; it realizes the adaptive dynamic range, solves the problem that fixed parameters cannot adapt to sudden changes in lighting, and lays the foundation for subsequent image enhancement, and lays the foundation for effectively preserving the visual priority of key areas such as sky and road surface.

[0017] (2) By providing an AI-enhanced decision-making module, this invention can distinguish between key and non-key areas, adopt different enhancement methods, and use non-linear brightness mapping to finely optimize the key areas, which can more accurately improve the detail clarity and contrast of the key areas, meeting the requirement of "prioritizing the clarity of key information" in vehicle scenarios. The non-key areas are optimized with enhancement factors to ensure the overall effect and avoid the computational time consumption caused by the complex processing of the whole image, adapting to the performance limitations of vehicle hardware; at the same time, it effectively preserves the visual priority of key areas such as the sky and the road surface. Attached Figure Description

[0018] Figure 1 This is a structural diagram of the AI-based in-vehicle screen image dynamic range adaptive enhancement system of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. In addition, the forms of the various structures described in the following embodiments are merely illustrative. The AI-based vehicle screen image dynamic range adaptive enhancement system involved in the present invention is not limited to the structures described in the following embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figure 1 As shown, the present invention provides an AI-based in-vehicle screen image dynamic range adaptive enhancement system, including an ambient light perception module, a screen content analysis module, a dynamic range acquisition module, an AI enhancement decision module, and an image rendering module; the modules communicate with each other via an in-vehicle bus. The ambient light sensing module is used to deploy dual sensors to collect multi-source ambient light data, and then preprocess and fuse the data to obtain the light intensity in the real environment, thereby providing a data basis and basis for dynamic range acquisition. The screen content analysis module is used to extract and classify the features of the content of the vehicle screen image. By identifying the key areas of the vehicle screen image, it lays the foundation for subsequent local enhancement. The dynamic range acquisition module obtains the optimal dynamic range based on data from the ambient light perception module and the screen content analysis module; The AI-enhanced decision module generates parameters for a nonlinear brightness mapping function based on the optimal dynamic range. It then uses the nonlinear brightness mapping function for local optimization of key regions and enhancement factors for optimization of non-key regions. Non-key regions refer to parts of the image that are not critical. Distinguishing between key and non-key regions aims to employ different enhancement methods. Using nonlinear brightness mapping for fine-tuning of key regions more accurately improves the detail clarity and contrast, meeting the requirement of "prioritizing clarity of key information" in automotive scenarios. Lightweight optimization using enhancement factors for non-key regions ensures overall quality while avoiding the computational time consumption caused by complex full-image processing, thus adapting to the performance limitations of automotive hardware. The image rendering module outputs the optimized image to the vehicle screen for display.

[0021] In this embodiment, it should be specifically noted that the dual sensors include a forward ambient light sensor and a screen surface light sensor. The forward ambient light sensor is installed on the inside of the windshield and is used to detect the external light intensity. The screen surface light sensor is attached to the edge of the screen and is used to detect the actual reflected light intensity of the screen. The preprocessing and fusion process specifically involves: Temperature compensation is applied to the data collected by the forward ambient light sensor to eliminate the effects of chip temperature drift; the formula for temperature compensation is as follows: ;in, This indicates the external light intensity collected by the forward ambient light sensor after temperature compensation. This represents the original external light intensity collected by the forward ambient light sensor; Indicates the current ambient temperature. Indicates the calibration temperature, select ; Represents the temperature coefficient, with a value of ; The overall ambient light intensity, i.e., the light intensity in the real environment, is obtained by weighted averaging, and can be expressed by the formula: ;in, Indicates the overall ambient light intensity; This indicates the intensity of reflected light from the screen, collected by the screen surface light sensor. for The weight, for The weights; and satisfy The weights can be dynamically adjusted based on sensor confidence levels; for example, they can be set under strong light. The value is 0.8, which can be adjusted in low light conditions. It is 0.7.

[0022] In this embodiment, it should be specifically explained that the feature extraction and classification of the content of the vehicle screen image specifically involves: The input image is preprocessed; the input image is the image that needs to be displayed on the vehicle screen. If the input image is... The image will Image converted Format, extract the luminance channel ,use The network extracts multi-scale feature maps; after fusing the feature maps through a fully convolutional network, a semantic mask is output. Mark the key regions of the image The set of key region types is represented as .

[0023] In this embodiment, it should be specifically explained that the dynamic range acquisition module acquires the optimal dynamic range in the following way: Based on the fundamental dynamic range benchmark, the logarithmic relationship between the ideal dynamic range and ambient light is obtained, expressed by the formula: ;in, Indicates the ideal dynamic range; This represents the reference dynamic range, which is the reference dynamic range value under typical indoor lighting conditions. ; Indicates the reference illumination threshold. ; Based on semantic masking The area proportion of each key region in the image is used to adjust the ideal dynamic range. For example, when the pedestrian area accounts for more than 10%, the dynamic range of the shadows is increased to highlight the pedestrian outlines; when the sky area accounts for more than 30%, the dynamic range of the highlights is limited to avoid overexposure. This can be expressed by the following formula: ;in, This represents the corrected target dynamic range, i.e., the optimal dynamic range; This indicates the percentage of the pedestrian area. Indicates the percentage of the sky area; and This represents the corresponding empirical coefficient, selected... , .

[0024] In this embodiment, it should be specifically explained that the local optimization of the key region using a nonlinear brightness mapping function specifically refers to: A conditional generative adversarial network is used, taking the original image, the combined ambient light intensity, and the semantic mask as input, and an enhanced image annotated by experts as input. Train the generator for the goal. The learning mapping relationship is expressed as follows: ; The generator is compressed to 500K parameters through knowledge distillation; the generator with 500K parameters is extremely small and can be easily deployed on edge devices such as vehicle terminals to meet localized real-time generation needs. Output nonlinear brightness mapping function parameters The Satisfying the formula: ;in, This represents the enhanced image brightness value. The coefficients of the quadratic term in the nonlinear brightness mapping function are used to control the curvature of the nonlinear transformation. The coefficients of the first-order term of the nonlinear brightness mapping function represent the slope of the corresponding linear part. The constant term of the nonlinear brightness mapping function corresponds to the intercept of the linear part. Represents the brightness value of the original image, satisfying ; A nonlinear brightness mapping function is used to enhance key regions while ensuring that the dynamic range of the enhanced image equals the optimal dynamic range. .

[0025] In this embodiment, it should be specifically noted that the enhancement factor is the ratio between the optimal dynamic range and the preset dynamic range value; the preset dynamic range value can be set by those skilled in the art or reasonably set based on historical data. The enhancement factor is expressed by the formula: ;in, Indicates the enhancing factor. Indicates the preset dynamic range value; The optimization of non-critical areas using an enhancement factor is expressed by the following formula: ;in, This represents the optimized pixel values ​​of the non-critical region image; This represents the original image pixel values ​​of non-critical regions after preprocessing. This represents the minimum pixel value of the non-critical region after preprocessing. This represents the maximum pixel value in the non-critical region after preprocessing.

[0026] In this embodiment, it should be specifically noted that the type of key region in the marked image is as follows: An encoder-decoder structure is adopted, where the encoder is responsible for extracting multi-scale features, and the decoder is responsible for restoring the feature map to the original image size and generating a mask; the encoder employs... As an encoder, it extracts multi-scale features from the input image. The decoder progressively upsamples the feature map output by the encoder to the original image size through transposed convolution. Each pixel value output by the decoder corresponds to a region category. The function is normalized to generate a semantic mask. ; Semantic mask After generation, the area ratio of each region is calculated, that is, the ratio of the number of pixels in the region to the total number of pixels, and key regions are selected according to preset thresholds. The preset thresholds can be specifically set by those skilled in the art according to different vehicle screens, or preset after analysis based on prior knowledge. For example, when the area ratio of the pedestrian area is greater than 10%, it is selected as a key region, and when the area ratio of the sky area is greater than 30%, it is selected as a key region. The specific area proportions of each region mentioned in the statistics are as follows: Count the number of pixels for each type in the semantic mask and calculate the area percentage: ;in, Indicates key region type The area percentage; Indicates key region type The number of pixels; Indicates the total number of pixels in the image; when the key region type It can be any of the following: roads, pedestrians, sky, and dashboard. When calculating the road area percentage, the key area type is... For roads; Finally, key region types are generated and stored as semantic masks; for example, a pixel value of 1 in the semantic mask indicates that the location is a pedestrian area, and a pixel value of 2 indicates that it belongs to the dashboard area.

[0027] In this embodiment, it should be specifically noted that the generative adversarial network mainly includes a generator, a discriminator, and a loss function; The generator takes as input the integrated ambient light intensity, semantic mask, and luminance channel, and outputs an enhanced image to learn a non-linear mapping from input to output, thereby generating an enhanced image that conforms to the characteristics of human vision. It includes an input layer, convolutional layers, residual blocks, transposed convolutional layers, and an output layer. The input layer takes as input the integrated ambient light intensity, semantic mask, and luminance channel. The convolutional layers use depthwise separable convolution to extract features. The residual blocks introduce residual connections to solve the gradient vanishing problem in deep networks. Each residual block contains two 3×3 convolutional layers and one skip connection, expressed by the formula: ;in, This indicates the output of the residual block; Indicates input, This represents the output of the convolutional layer; the transposed convolutional layer restores the feature map to its original size through transpose convolution, generating an enhanced brightness channel; the output layer uses... The function normalizes the luminance channel to between 0 and 1 to ensure that the output meets the range requirements of the luminance channel. The discriminator mainly comprises an input layer, a convolutional layer, and a fully connected layer. The input layer is used to input data, mainly the enhanced image and conditional information, including the comprehensive ambient light intensity and semantic mask. The convolutional layer uses depthwise separable convolution to extract features, gradually reducing the feature map size. The fully connected layer outputs the discrimination result, which includes true and false. The true result is the real enhanced image, i.e., the enhanced image annotated by experts, and the false result is the enhanced image generated by the generator, thereby guiding the generator to generate images that better meet the requirements. The loss function consists of adversarial loss and content loss, used to guide the generator to produce images that meet both the discriminator's and content requirements; the content loss includes perceptual loss and pixel loss, thereby ensuring that the enhanced image generated by the generator is consistent with the real enhanced image in terms of content and pixel values; the loss function is expressed as: ;in, This represents the total loss function; Indicates resistance to loss; Indicates perceived loss; Indicates pixel loss; , , These represent the corresponding weighting coefficients; in this embodiment, we take... , , ; , , The value can be adjusted and modified according to specific circumstances; ; in, Represents the mathematical expectation; Represents the brightness channel of the original image; Indicates conditional information; This represents the distribution of real augmented images, i.e., the probability distribution of real samples in the training data; Indicates the discriminator; This represents the distribution of the original image, i.e., the probability distribution of the original input samples in the training data; Indicates a generator; ; in, Indicates the number of pixels. ; The first part represents the extraction of real-world enhanced images. Feature values ​​of each pixel; The image extracted from the generator output represents the first... Feature values ​​of each pixel; ; in, Indicates the image height. Indicates the image width; Indicates the generator output image at the th Line number The pixel values ​​of the column; Indicates the real-image enhancement at the 1st Line number The pixel values ​​of the column.

[0028] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0029] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An AI-based in-vehicle screen image dynamic range adaptive enhancement system, characterized in that: It includes an ambient light perception module, a screen content analysis module, a dynamic range acquisition module, an AI-enhanced decision-making module, and an image rendering module; The ambient light sensing module is used to deploy dual sensors to collect multi-source ambient light data and then perform preprocessing and fusion. The screen content analysis module is used to extract and classify the features of the content of the vehicle screen image by identifying key areas of the vehicle screen image. The dynamic range acquisition module obtains the optimal dynamic range based on data from the ambient light perception module and the screen content analysis module; The AI-enhanced decision-making module generates parameters for a nonlinear brightness mapping function based on the optimal dynamic range, and performs local optimization of the key region using the nonlinear brightness mapping function, while optimizing the non-key region using an enhancement factor. The image rendering module outputs the optimized image to the vehicle screen for display.

2. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 1, characterized in that: The dual sensors include a forward ambient light sensor and a screen surface light sensor. The preprocessing and fusion process specifically involves: Temperature compensation is applied to the data collected by the forward ambient light sensor; the formula for temperature compensation is as follows: ;in, This indicates the external light intensity collected by the forward ambient light sensor after temperature compensation. This represents the original external light intensity collected by the forward ambient light sensor; Indicates the current ambient temperature. Indicates the calibrated temperature; Indicates the temperature coefficient; The overall ambient light intensity is obtained by the weighted average method, expressed by the formula: ;in, Indicates the overall ambient light intensity; This indicates the intensity of reflected light from the screen, collected by the screen surface light sensor. for The weight, for The weights; and satisfy .

3. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 2, characterized in that: The specific steps for feature extraction and classification of the content of the in-vehicle screen image are as follows: The input image is preprocessed; the input image is the image that needs to be displayed on the vehicle screen. If the input image is... The image will Image converted Format, extract the luminance channel ,use The network extracts multi-scale feature maps; after fusing the feature maps through a fully convolutional network, a semantic mask is output. Mark the key regions of the image. The set of key region types is represented as .

4. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 3, characterized in that: The dynamic range acquisition module obtains the optimal dynamic range specifically as follows: Based on the fundamental dynamic range benchmark, the logarithmic relationship between the ideal dynamic range and ambient light is obtained, expressed by the formula: ;in, Indicates the ideal dynamic range; Indicates the reference dynamic range. ; Indicates the reference illumination threshold. ; Based on semantic masking The area proportions of each key region are used to correct the ideal dynamic range, resulting in the corrected target dynamic range. That is, the optimal dynamic range.

5. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 4, characterized in that: The specific steps for local optimization of the key region using a nonlinear brightness mapping function are as follows: A conditional generative adversarial network is used, taking the original image, the combined ambient light intensity, and the semantic mask as input, and an enhanced image annotated by experts as input. Train the generator for the goal. The learning mapping relationship is expressed as follows: ; The generator is compressed to 500K parameters through knowledge distillation; Output nonlinear brightness mapping function parameters The Satisfying the formula: ;in, This represents the enhanced image brightness value. This represents the coefficients of the quadratic term in the nonlinear brightness mapping function. This represents the coefficient of the first-order term in the nonlinear brightness mapping function. The constant term of the nonlinear brightness mapping function is represented. Represents the brightness value of the original image; A nonlinear brightness mapping function is used to enhance key areas.

6. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 5, characterized in that: The enhancement factor is the ratio between the optimal dynamic range and the preset dynamic range value; The enhancement factor is expressed by the formula: ;in, Indicates the enhancing factor. Indicates the preset dynamic range value; The optimization of non-critical areas using an enhancement factor is expressed by the following formula: ;in, This represents the optimized pixel values ​​of the non-critical region image; This represents the original image pixel values ​​of non-critical regions after preprocessing. This represents the minimum pixel value of the non-critical region after preprocessing. This represents the maximum pixel value in the non-critical region after preprocessing.

7. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 6, characterized in that: The specific types of key regions in the marked image are: An encoder-decoder structure is adopted, where the encoder is responsible for extracting multi-scale features, and the decoder is responsible for restoring the feature map to the original image size and generating a mask; the encoder employs... As an encoder, it extracts multi-scale features from the input image. The decoder progressively upsamples the feature map output by the encoder to the original image size through transposed convolution. Each pixel value output by the decoder corresponds to a region category. The function is normalized to generate a semantic mask. ; Semantic mask After generation, the area ratio of each region is calculated, that is, the ratio of the number of pixels in each region to the total number of pixels, and key regions are filtered according to preset thresholds. Finally, the key region type is generated and stored as a semantic mask.

8. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 7, characterized in that: The specific area proportions of each region mentioned in the statistics are as follows: Count the number of pixels for each type in the semantic mask and calculate the area percentage: ;in, Indicates key region type The area percentage; Indicates key region type The number of pixels; This represents the total number of pixels in the image.

9. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 8, characterized in that: The generative adversarial network mainly includes a generator, a discriminator, and a loss function; The generator is used to input the integrated ambient light intensity, semantic mask, and luminance channel, and output an enhanced image, including an input layer, a convolutional layer, a residual block, a transposed convolutional layer, and an output layer; the discriminator mainly includes an input layer, a convolutional layer, and a fully connected layer; the loss function consists of adversarial loss and content loss, and the content loss includes perceptual loss and pixel loss.

10. The vehicle-mounted screen image dynamic range adaptive enhancement system based on artificial intelligence according to claim 9, characterized in that: The loss function is expressed as: ;in, This represents the total loss function; Indicates resistance to loss; Indicates perceived loss; Indicates pixel loss; , , These represent the corresponding weighting coefficients.

Images