Image defogging method for visibility change scene and related device

Abstract

The invention discloses an image defogging method for a visibility change scene and a related device, and the method comprises the steps: selecting an adaptive defogging model according to the scene radiance of an image at a current moment, carrying out the overall defogging of a low-resolution image layer in the defogging model, carrying out the mesh-division defogging of a high-resolution image layer, and carrying out the overall defogging of the low-resolution image layer and the mesh-division defogging of the high-resolution image layer. Compared with a traditional fixed model, the method can adapt to a scene with violent visibility change.

Description

Technical Field

[0001] This invention relates to an image dehazing method and related apparatus for scenes with varying visibility, belonging to the fields of intelligent sensing, image enhancement and machine vision technology. Background Technology

[0002] In environments with drastic visibility changes, such as biomass power plants, mining operations, tunnel construction, and fire rescue, high concentrations of dust, smoke, and complex lighting conditions can severely interfere with visual sensors, causing problems such as decreased image contrast, color shift, and loss of detail. These factors not only reduce the equipment's ability to visualize the environment but also affect the path planning, hazard identification, and operational decisions of robots or monitoring systems. Existing image dehazing methods mostly rely on fixed models, making them difficult to adapt to scenarios with drastic visibility changes. Summary of the Invention

[0003] This invention provides an image dehazing method and related apparatus for addressing changes in visibility, thus solving the problems disclosed in the background art.

[0004] According to one aspect of this application, an image dehazing method for scenes with varying visibility is provided, comprising: In response to the current moment when an image dehazing topic is detected from any topic channel, obtain the current moment image corresponding to the image dehazing topic; Based on the current image, determine the scene radiance of the current image. Based on the scene radiance, select a dehazing model that matches the current image from a pre-built set of dehazing models. Use the selected dehazing model to dehaze the current image to obtain the dehazed current image. Then publish the dehazed current image as a topic. In the dehazing model, the current image is processed with multi-resolution to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into grids. An appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed and used as the dehazed current image.

[0005] Furthermore, based on the image at the current moment, the scene radiometer of the image acquisition environment at the current moment is determined, including: Estimate atmospheric light and transmittance based on the image at the current moment; The scene radiance of the image acquisition environment at the current moment is determined based on air light and transmittance.

[0006] Furthermore, the scene radiometer of the image acquisition environment at the current moment is determined by the following formula: ; In the formula, J(x) is the scene radiance, I(x) is the image at the current time, A is the air light, t(x) is the transmittance, t0 is the lower limit threshold to prevent noise amplification caused by excessively low transmittance, and max(t(x),t0) means taking the maximum value of t(x) and t0.

[0007] Furthermore, for the first image layer in the image pyramid with a resolution less than a threshold, dehazing is performed on the entire first image layer, including: For the first image layer in the image pyramid with a resolution less than a threshold, brightness correction is performed on the entire first image layer.

[0008] Furthermore, the second image layer is divided into a grid, and an adapted dehazing strategy is used to dehaze the sub-image layers within the grid, including: The second image layer is divided into grids. For each sub-image layer in the grid, the atmospheric light, transmittance, and contrast are calculated. Based on the atmospheric light, transmittance, and contrast, the grid state is determined. Based on the grid state, a suitable dehazing strategy is selected from a pre-built set of dehazing strategies. The suitable dehazing strategy is then used to dehaze the sub-image layers in the grid.

[0009] Furthermore, the grid status includes bright areas, dark areas, high-dust areas, or low-dust areas; For bright areas, the dehazing strategy selected from the pre-built dehazing strategy set is to reduce overexposure and enhance details; for dark areas, the dehazing strategy selected from the pre-built dehazing strategy set is to increase brightness and improve local contrast; for high-dust areas, the dehazing strategy selected from the pre-built dehazing strategy set is to use a strong dehazing mode; for low-dust areas, the dehazing strategy selected from the pre-built dehazing strategy set is to use a weak dehazing mode.

[0010] Furthermore, based on all the dehazed sub-image layers and the first image layer, a new image at the current moment is reconstructed, including: Based on the dehazed sub-image layers and their weights within the same layer, the dehazed second image layer of the corresponding layer is reconstructed. Based on the dehazed second image layer and the dehazed first image layer, a new image at the current moment is reconstructed; The weights of the sub-image layers are determined based on the average brightness, average dark channel, and average transmittance of the sub-image layers, using the following formula: ; In the formula, w i Let f be the weight of the i-th sub-image layer, and f be a monotonically increasing function. Let D be the dust index of the i-th sub-image layer, α and β be empirical weights, and D be the dust index of the i-th sub-image layer. i T is the mean of the dark channel of the i-th sub-image layer. i Let L be the average transmittance of the i-th sub-image layer. i Let be the average brightness value of the i-th sub-image layer.

[0011] According to another aspect of this application, an image dehazing device for scenes with varying visibility is provided, comprising: The image acquisition module, in response to the image dehazing topic being detected from any topic channel at the current moment, acquires the image at the current moment corresponding to the image dehazing topic; The dehazing module determines the scene radiance of the current image based on the current image, selects a dehazing model that matches the current image from a pre-built set of dehazing models based on the scene radiance, uses the selected dehazing model to dehaze the current image, obtains the dehazed current image, and publishes the dehazed current image as a topic. In the dehazing model, the current image is processed with multi-resolution to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into grids. An appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed and used as the dehazed current image.

[0012] According to another aspect of this application, a computer-readable storage medium is provided that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform an image dehazing method for scenes with varying visibility.

[0013] According to another aspect of this application, a computer device is provided, including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing an image dehazing method for a scene with varying visibility.

[0014] The beneficial effects achieved by this invention are as follows: This invention selects an appropriate dehazing model based on the scene radiance of the image at the current moment, and performs overall dehazing on the low-resolution image layer and mesh-based dehazing on the high-resolution image layer in the dehazing model, reconstructing the dehazed image at the current moment. Compared with the traditional fixed model, it can adapt to scenes with drastic changes in visibility. Attached Figure Description

[0015] Figure 1 This is a flowchart of an image dehazing method for scenes with varying visibility; Figure 2 A block diagram of an image dehazing device for scenarios with varying visibility; Figure 3 This is a detailed block diagram of an image dehazing device for scenarios with varying visibility. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0017] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application.

[0018] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0019] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0020] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0021] It should be noted that similar symbols and letters in the accompanying drawings represent similar items; therefore, once an item is defined in one accompanying drawing, it does not need to be discussed further in subsequent accompanying drawings.

[0022] See Figure 1 , Figure 1This is a flowchart illustrating an image dehazing method for scenes with varying visibility, provided in an embodiment of this application. The image dehazing method can be executed by an image dehazing device, which can be a terminal device or a server. The terminal device can include, but is not limited to, mobile phones, computers, etc., as described in this embodiment. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, big data, and artificial intelligence platforms, etc., as described in this embodiment. Optionally, the image dehazing method can also be executed collaboratively by multiple electronic devices with computing power. For ease of explanation, subsequent embodiments will be described as being executed by an image dehazing device.

[0023] Figure 1 In this context, an image dehazing method for scenes with varying visibility may include at least the following steps: Step 1: In response to the image dehazing topic being detected from any topic channel at the current moment, obtain the image at the current moment corresponding to the image dehazing topic.

[0024] It should be noted that, taking large power plants or industrial parks as examples, front-end robots or cameras will be deployed in different scenarios to collect scene data in real time. To facilitate multi-channel image processing, a dynamic topic pipeline approach can be adopted, that is, processing corresponding images based on image dehazing topics, and then publishing the processed images as topics.

[0025] Step 2: Based on the current image, determine the scene radiance of the current image. Based on the scene radiance, select a dehazing model that matches the current image from a pre-built dehazing model set. Use the selected dehazing model to dehaze the current image to obtain the dehazed current image, and publish the dehazed current image as a topic. In the dehazing model, the current image is processed by multi-resolution to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into grids. The appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed, and the new current image is used as the dehazed current image.

[0026] In some embodiments, determining the scene radiance of the image acquisition environment at the current moment may include: 21) Estimate the atmospheric light and transmittance based on the image at the current moment.

[0027] Atmospheric light needs to be estimated through the dark channel. The dark channel D(x) is used to estimate the dust blocking intensity, and the formula can be expressed as: ; In the formula, This represents finding the minimum value in the neighborhood Ω(x) surrounding pixel x, where Ω(x) is the set of pixels within a certain range centered at pixel x. For each pixel x, the minimum value among all neighboring pixels is found by searching within this neighborhood. R, G, and B represent the red, green, and blue channels of the image, respectively. c (y) represents the pixel value of pixel y in the c-th channel (i.e., R, G, or B) of the image. This means that for each pixel y, the values ​​I of that pixel in the red, green, and blue channels are considered separately. c (y), and then take the minimum value among these values. In short, it selects the minimum color value in the RGB channels of each pixel.

[0028] The atmospheric light A, which reflects the offset of the background brightness, can then be expressed as: ; ; In the formula, argmax x D(x) refers to finding the x that maximizes D(x). In dark channel dehazing, D(x) typically represents the dark channel value at a certain location in the image; therefore, this step involves finding the x corresponding to the maximum dark channel value. Indicates will As atmospheric light It is in the image The pixel value at the location (i.e., the location corresponding to the maximum dark channel value).

[0029] It should be noted that, In some cases (e.g., when there are multiple strong light sources in the image or the image quality is poor), only use Using this to estimate atmospheric light may not be accurate enough. Therefore, the maximum value of the dark channel (i.e., For several pixels near the light source, calculate their brightness and take a weighted average of these brightness values ​​to obtain a more robust atmospheric light, thereby improving robustness.

[0030] The transmittance t(x) can be estimated using the normalized dark channel representation, and the formula is as follows: ; In the formula, ω is a weighting factor that controls the defogging intensity, and A c It is the atmospheric light value of a certain channel in the image, which is obtained through the aforementioned atmospheric light estimation process, and represents the intensity of the background light.

[0031] It should be noted that, if necessary, edge-preserving operators such as guided filtering and bilateral filtering can be introduced to further optimize t(x) to improve the image restoration quality.

[0032] 22) Determine the scene radiance of the image acquisition environment at the current moment based on air light and transmittance.

[0033] The formula for determining the scene radiance of the image acquisition environment at the current moment can be: ; In the formula, J(x) is the scene radiance, I(x) is the image at the current time, A is the air light, t(x) is the transmittance, t0 is the lower limit threshold to prevent noise amplification caused by excessively low transmittance, and max(t(x),t0) means taking the maximum value of t(x) and t0.

[0034] It should be noted that the dehazing models in the dehazing model set can include dark channel prior models, atmospheric scattering models, Retinex models, and deep learning dehazing models. Among them, dark channel prior models are suitable for general low visibility environments, especially those affected by fog, dust, etc.; atmospheric scattering models are more suitable for complex lighting conditions; Retinex models are suitable for enhancing image brightness and contrast; and deep learning dehazing models are based on convolutional neural networks (CNN) and are suitable for scenarios requiring high-precision dehazing.

[0035] It should be noted that in real-world industrial scenarios, the distribution of dust obstruction exhibits distinct local characteristics. For example, areas near light sources are prone to overexposure and highlights, while areas farther from the light source suffer from strong shadows. Furthermore, the concentration and size distribution of dust particles in the air also show spatial variations. Traditional globally uniform processing strategies struggle to simultaneously address detail restoration, brightness balance, and noise suppression. Therefore, the dehazing model in this embodiment employs a differentiated processing strategy to improve overall image quality and detail representation.

[0036] In the dehazing model, the current image is first processed with multi-resolution to obtain an image pyramid. The resolution increases sequentially from top to bottom in the image pyramid. Then, for the first image layer (i.e., the low-resolution image layer) with a resolution less than a threshold (e.g., 32×32) in the image pyramid, brightness correction is performed on the entire first image layer. Finally, for the second image layer (i.e., the high-resolution image layer) with a resolution not less than the threshold in the image pyramid, detail enhancement is performed. Specifically, the second image layer is divided into grids, and an appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed, and the new current image is used as the dehazed current image.

[0037] In some embodiments, dividing the second image layer into a grid and applying an adapted dehazing strategy to dehaze the sub-image layers within the grid may include: The second image layer is divided into grids. For each sub-image layer within a grid, atmospheric light, transmittance, and contrast are calculated. Based on these parameters, the grid state is determined, classifying it as a bright area, dark area, high-dust area, or low-dust area. For bright areas, the pre-built dehazing strategy selected is to reduce overexposure and enhance details; for dark areas, it's to increase brightness and improve local contrast; for high-dust areas, a strong dehazing mode is used; and for low-dust areas, a weak dehazing mode is used.

[0038] It should be noted that the image reconstruction here has two parts: reconstruction of the second image layer based on the grid division, and reconstruction of the entire image at the current moment. The specific process is as follows: A1) Based on the dehazed sub-image layers and their weights within the same layer, reconstruct the dehazed second image layer of the corresponding layer; wherein, the weights of the sub-image layers are determined based on the average brightness, average dark channel, and average transmittance of the sub-image layers.

[0039] Assuming the second image layer is divided into N grids, i.e., containing N sub-image layers Ri, the average brightness, average dark channel, and average transmittance of each sub-image layer can be calculated using the following formula: L i =mean(I(Ri)); D i =mean(D(Ri)); T i =mean(t(Ri)); In the formula, mean represents the average value of all pixels within a certain region, and D i T is the mean of the dark channel of the i-th sub-image layer. i Let L be the average transmittance of the i-th sub-image layer. i Let be the average brightness value of the i-th sub-image layer, I(Ri) be the original pixel value of region Ri in the image, D(Ri) be the dark channel value of region Ri in the image, and t(Ri) be the transmittance value of region Ri in the image.

[0040] Therefore, the dust index can be expressed as: ; In the formula, P i Let α be the dust index of the i-th sub-image layer, and β be the empirical weights.

[0041] The weights of the sub-image layer can be expressed as: ; In the formula, w i Let f be the weight of the i-th sub-image layer, and f be a monotonically increasing function that enhances the grid with denser dust and lower brightness.

[0042] The final reconstructed second image layer can be represented as: ; In the formula, J out (x) The reconstructed second image layer, where N is the number of sub-image layers, J i (x) represents the enhancement result of the sub-image, i.e., the result after dehazing.

[0043] A2) Based on the second image layer after dehazing and the first image layer after dehazing, reconstruct the new image at the current moment.

[0044] It should be noted that the reconstruction of the current image here is a known reconstruction method, namely, reconstructing the image from the image pyramid, which will not be described in detail here.

[0045] It should be noted that the current moment image after defogging is published as a topic, and the topic output party can obtain the published topic from the dynamic topic pipeline.

[0046] The above method selects an appropriate dehazing model based on the scene radiance of the image at the current moment. In the dehazing model, the low-resolution image layer is dehazed as a whole, and the high-resolution image layer is dehazed by mesh. The dehazed image at the current moment is then reconstructed. Compared with the traditional fixed model, it can adapt to scenes with drastic changes in visibility.

[0047] See Figure 2 , Figure 2 This is a block diagram of an image dehazing device for scenes with varying visibility, provided in an embodiment of this application. It is a virtual device that can be loaded and executed by a computer device, which may include the aforementioned image dehazing device. Figure 2 The apparatus may include an image acquisition module and a dehazing module, which, when performing the above-described image dehazing method, can: The image acquisition module, in response to the current moment when an image dehazing topic is detected from any topic channel, acquires the current moment's image corresponding to the image dehazing topic.

[0048] The dehazing module determines the scene radiance of the current image based on the current moment image. Based on the scene radiance, it selects a suitable dehazing model from a pre-built set of dehazing models. The selected dehazing model is then used to dehaze the current image, resulting in a dehazed current image, which is then published as a topic. Specifically, within the dehazing model, the current image undergoes multi-resolution processing to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into a grid, and an appropriate dehazing strategy is used to dehaze the sub-image layers within the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed and used as the dehazed current image.

[0049] The aforementioned device selects an appropriate dehazing model based on the scene radiance of the image at the current moment. In the dehazing model, it performs overall dehazing on low-resolution image layers and performs grid-based dehazing on high-resolution image layers, reconstructing the dehazed image at the current moment. Compared with the traditional fixed model, it can adapt to scenes with drastic changes in visibility.

[0050] See Figure 3 , Figure 3 For based on Figure 2 A more detailed device block diagram further divides the dehazing module into an adaptive model scheduling module and a unified dehazing engine module. The adaptive model scheduling module mainly determines the scene radiance of the current image based on the current image, and selects a dehazing model that is suitable for the current image from a pre-built set of dehazing models based on the scene radiance. The unified dehazing engine module mainly uses the selected dehazing model to dehaze the current image, obtains the dehazed current image, and publishes the dehazed current image as a topic.

[0051] The adaptive model scheduling module automatically switches between multiple preset defogging models or parameter files by estimating air light and transmittance. For example, around the biomass combustion grate, due to the high smoke concentration and large brightness fluctuations, the system will switch to the enhanced model; in the silo or conveyor belt area, the balanced model is used to ensure the naturalness of the image.

[0052] To enhance adaptability, an environmental self-learning mechanism can be introduced into the adaptive model scheduling module. By recording daily air light change trends and transmittance distribution characteristics, an environmental profile is automatically generated for the next cycle of model parameter recommendation and automatic switching, ultimately achieving long-term performance optimization capability that becomes more accurate with use.

[0053] The unified dehazing engine module can be configured to support concurrent input of multi-channel images and achieve real-time image dehazing through a dehazing model. This module can adopt multi-threaded control, vectorization optimization, and block-level task scheduling strategies to improve the speed and stability of image processing in high-dust scenes. It can adaptively adjust to environments with different dust intensities, automatically increasing the dehazing intensity and brightness enhancement ratio in high-dust, low-brightness scenes, while reducing the processing intensity in well-lit scenes to maintain a natural visual effect.

[0054] Addressing the characteristics of large, densely distributed dust particles and strong light variations in biomass power plants, the unified dehazing engine module employs multi-scale filtering, block adaptive strategies, and edge-preserving enhancement techniques to restore image details more naturally and avoid over-enhancement or noise amplification. The unified dehazing engine module supports both compressed and original image inputs, automatically performing format conversion and memory optimization during processing.

[0055] To accommodate the input and output of topics, a dynamic topic pipeline module and a visual control panel module can also be configured in the device.

[0056] The dynamic topic pipeline module supports flexible combinations of any ROS input and output topics, and can dynamically add or delete image processing channels. This module can automatically perform compressed image decoding, data buffering, processing triggering, status feedback, and defogging result publishing, making it suitable for scenarios involving simultaneous processing of multiple areas of video feed, such as biomass silos, conveyor chains, and incinerator front-end monitoring. It automatically allocates resources during multi-channel operation; if a channel malfunctions or the frame rate drops, a warning will be displayed on the control panel, and the channel can be automatically restarted or a backup source can be switched.

[0057] The implementation principle of the dynamic topic pipeline module can be summarized as follows: Input images are passed sequentially between different modules through an event-driven mechanism, multi-threaded queues are used to reduce blocking, and the independence between multiple channels is guaranteed. The content and steps are: Listen to the topic channel → Decode the image → Send it to the dehazing engine → Obtain the processing result → Publish the topic → Status feedback → Interface display.

[0058] The visual control panel module, based on the Qt framework, provides a graphical interface for configuring, monitoring, and debugging the device. Users can directly input or select ROS topics, adjust model parameters, enable or disable the defogging channel, and view images, transmittance curves, and atmospheric light trends before and after defogging in real time. The interface employs a multi-level color-coded indicator: green indicates normal operation, yellow indicates processing delays or frame rate drops, and red indicates channel failures or model anomalies, helping biomass power plant personnel quickly identify problems and take corrective action.

[0059] To accommodate ROS and Qt, a collaborative ROS and Qt module can be configured in the device. This module uses a dual-loop structure to enable parallel data processing and UI updates, ensuring stable performance even under high concurrency. To prevent the GUI from blocking ROS callbacks, a message queue isolation mechanism can be used, ensuring that defogging tasks and UI updates do not interfere with each other. This is particularly suitable for scenarios involving simultaneous monitoring of multiple screens, such as biomass fuel delivery systems.

[0060] To enable environmental monitoring, an environmental health monitoring module can also be configured in the device. Based on transmittance and air light curves, it automatically generates a visibility index. When a sharp increase in dust concentration, lens contamination, or light source obstruction is detected, an alarm will automatically pop up and suggest that on-site personnel perform cleaning or inspection. Simultaneously, it can record daily visibility trends, providing a basis for dust control and equipment maintenance in power plants.

[0061] To adapt to multi-node collaborative management, a multi-node collaborative management module can also be configured in the device to support cross-device collaboration. It can uniformly manage cameras in different locations (such as the top of the silo, in front of the furnace, along the conveyor chain, etc.), remotely send model updates and synchronize parameter configurations, and automatically switch to backup nodes when a node is abnormally offline, thereby improving the continuity and reliability of monitoring.

[0062] It should be noted that in large power plants or industrial parks, there may be dozens of cameras distributed in different locations. If all images are directly transmitted to the central node for processing, it will create enormous network bandwidth pressure. The multi-node collaborative management module in the device can adopt a "distributed preprocessing + central fusion" architecture, which can complete the basic dehazing at the front-end robot or camera side, and only send the enhanced keyframes or thumbnails to the center, significantly reducing the data transmission pressure.

[0063] It should be noted that in high-dust environments, cameras may experience issues such as image jitter, short-term exposure anomalies, temporary device disconnections, and intermittent frame drops in ROS. Without an automatic detection mechanism, this can easily lead to drastic fluctuations in dehazing results or image distortion. The device can be designed with intelligent anomaly detection and self-recovery mechanisms, proactively identifying faults and automatically recovering through deep statistical detection, temporal consistency detection, and anomaly pattern recognition technologies, thereby improving long-term stability.

[0064] Specifically, the device can detect the following anomalies: (1) black screen, white screen, or frozen image; (2) sudden drop in frame rate or no data for a long time; (3) brightness jump exceeding the threshold; (4) abnormal transmittance estimation (such as negative value or explosive jump); (5) abnormal increase in ambient light (such as extreme cases caused by direct light source). When an anomaly occurs, the device will: (1) automatically restart the channel or subscriber; (2) switch to the backup model or backup input source; (3) fall back to the low enhancement mode to protect the image quality; (4) synchronously send an anomaly prompt to the GUI to assist in manual confirmation of the cause.

[0065] The above-mentioned device is applicable to the following scenarios: 1) Loader feeding operations in biomass power plants, where materials such as branches, planks, and bamboo strips generate extremely high concentrations of dust during shoveling, dropping, and crushing, often causing "instantaneous white screens" in the camera's field of view. The above-mentioned device can effectively suppress large-particle dust obstruction, improving operational safety. 2) Long-term monitoring of dust suspension time in silos and stockpiles, which easily leads to continuous degradation of the field of view; the output of the above-mentioned device can be used to guide the scheduling of dust suppression spraying in the plant area. 3) Monitoring of periodic dust bursts at the crusher inlet and conveyor chain; the above-mentioned device can suppress the brightness jumps caused by these bursts, improving stability. 4) High-temperature smoke areas in front of the furnace; the above-mentioned device can maintain the stability of the edge structure under severe thermal disturbance. 5) It can be extended to smoke environments in mines, tunnels, and fire-fighting robots; the above-mentioned device can achieve cross-industry migration and application, improving its versatility in multiple scenarios.

[0066] This application also relates to a computer-readable storage medium that stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform an image dehazing method for scenes with varying visibility.

[0067] This application also relates to a computer device including one or more processors and one or more memories, wherein one or more programs are stored in the one or more memories and configured to be executed by the one or more processors, and the one or more programs include instructions for performing an image dehazing method for scenes with varying visibility.

[0068] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0069] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A method for image dehazing in scenes with varying visibility, characterized in that, include: In response to the current moment when an image dehazing topic is detected from any topic channel, obtain the current moment image corresponding to the image dehazing topic; Based on the current image, determine the scene radiance of the current image. Based on the scene radiance, select a dehazing model that matches the current image from a pre-built set of dehazing models. Use the selected dehazing model to dehaze the current image to obtain the dehazed current image. Then publish the dehazed current image as a topic. In the dehazing model, the current image is processed with multi-resolution to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into grids. An appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed and used as the dehazed current image.

2. The method according to claim 1, characterized in that, Based on the image at the current moment, determine the scene radiometer of the image acquisition environment at the current moment, including: Estimate atmospheric light and transmittance based on the image at the current moment; The scene radiance of the image acquisition environment at the current moment is determined based on air light and transmittance.

3. The method according to claim 2, characterized in that, The formula for determining the scene radiometer of the image acquisition environment at the current moment is: ； In the formula, J(x) is the scene radiance, I(x) is the image at the current time, A is the air light, t(x) is the transmittance, t0 is the lower limit threshold to prevent noise amplification caused by excessively low transmittance, and max(t(x),t0) means taking the maximum value of t(x) and t0.

4. The method according to claim 1, characterized in that, For the first image layer in the image pyramid with a resolution less than a threshold, dehazing is performed on the entire first image layer, including: For the first image layer in the image pyramid with a resolution less than a threshold, brightness correction is performed on the entire first image layer.

5. The method according to claim 1, characterized in that, The second image layer is divided into a grid, and an adapted dehazing strategy is used to dehaze the sub-image layers within the grid, including: The second image layer is divided into grids. For each sub-image layer in the grid, the atmospheric light, transmittance, and contrast are calculated. Based on the atmospheric light, transmittance, and contrast, the grid state is determined. Based on the grid state, a suitable dehazing strategy is selected from a pre-built set of dehazing strategies. The suitable dehazing strategy is then used to dehaze the sub-image layers in the grid.

6. The method according to claim 5, characterized in that, The grid status includes bright areas, dark areas, high dust areas, or low dust areas; For bright areas, the dehazing strategy selected from the pre-built dehazing strategy set is to reduce overexposure and enhance details; for dark areas, the dehazing strategy selected from the pre-built dehazing strategy set is to increase brightness and improve local contrast; for high-dust areas, the dehazing strategy selected from the pre-built dehazing strategy set is to use a strong dehazing mode; for low-dust areas, the dehazing strategy selected from the pre-built dehazing strategy set is to use a weak dehazing mode.

7. The method according to claim 1, characterized in that, Based on all the dehazed sub-image layers and the first image layer, a new image at the current moment is reconstructed, including: Based on the dehazed sub-image layers and their weights within the same layer, the dehazed second image layer of the corresponding layer is reconstructed. Based on the dehazed second image layer and the dehazed first image layer, a new image at the current moment is reconstructed; The weights of the sub-image layers are determined based on the average brightness, average dark channel, and average transmittance of the sub-image layers, using the following formula: ； In the formula, w i Let f be the weight of the i-th sub-image layer, and f be a monotonically increasing function. Let D be the dust index of the i-th sub-image layer, α and β be empirical weights, and D be the dust index of the i-th sub-image layer. i T is the mean of the dark channel of the i-th sub-image layer. i Let L be the average transmittance of the i-th sub-image layer. i Let be the average brightness value of the i-th sub-image layer.

8. An image dehazing device for scenes with varying visibility, characterized in that, include: The image acquisition module, in response to the image dehazing topic being detected from any topic channel at the current moment, acquires the image at the current moment corresponding to the image dehazing topic; The dehazing module determines the scene radiance of the current image based on the current image, selects a dehazing model that matches the current image from a pre-built set of dehazing models based on the scene radiance, uses the selected dehazing model to dehaze the current image, obtains the dehazed current image, and publishes the dehazed current image as a topic. In the dehazing model, the current image is processed with multi-resolution to obtain an image pyramid. For the first image layer in the image pyramid with a resolution less than a threshold, the entire first image layer is dehazed. For the second image layer in the image pyramid with a resolution not less than the threshold, the second image layer is divided into grids. An appropriate dehazing strategy is used to dehaze the sub-image layers in the grid. Based on all the dehazed sub-image layers and the first image layer, a new current image is reconstructed and used as the dehazed current image.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores one or more programs, the one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of any one of claims 1 to 7.

10. A computer device, characterized in that, include: One or more processors and one or more memories, one or more programs stored in one or more memories and configured to be executed by one or more processors, the one or more programs including instructions for performing the method of any one of claims 1 to 7.

Images