A monitoring image defogging enhancement method in a high-dust environment of a beneficiation site

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a kind of ore dressing field high dust environment under monitoring image defogging enhancement method, its method includes: using weighted guided filter to carry out smoothing denoising to original image, global atmosphere light is obtained by quadtree subdivision method;Heterogeneous discrimination index of combining spectral similarity factor and texture confidence factor is constructed, which is used to distinguish dust and ore with similar color;Based on the discrimination index, the dynamic defogging coefficient at pixel level is calculated, and the adaptive transmittance is obtained combined with the dark channel prior;Finally, the image is restored using the atmospheric scattering model and the limited contrast adaptive histogram equalization processing is carried out.The present application effectively solves the misjudgment problem caused by the similar color of ore and dust in the ore dressing field, realizes strong defogging of dust area and detail preservation of ore area, and improves the clarity and contrast of monitoring image.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site. Background Technology

[0002] In modern mineral processing, in order to monitor the production status of key processes such as crushing, screening and transportation in real time, industrial video monitoring systems are usually deployed at the work site to remotely monitor equipment operation, ore transportation and personnel safety.

[0003] However, the working environment at the ore processing site is quite complex, with a large amount of associated dust generated during operations. These high concentrations of suspended particles and aerosols permeate the area around the equipment and ore, causing severe fogging in the images captured by the monitoring probes. Specifically, this manifests as a significant reduction in image contrast, color distortion, and blurred details. This not only hinders the control room personnel from clearly observing the on-site conditions but also renders subsequent machine vision-based intelligent analysis algorithms ineffective due to the low image quality.

[0004] To address these visual obstacles, existing technologies typically employ general image dehazing algorithms based on theories such as dark channel priors. However, in the specific scenario of a mineral processing site, the problem of color interference often exists, meaning the color of the ore itself is quite similar to the color of the dust. When processing such images using conventional dehazing techniques, the algorithm struggles to distinguish between ore entities of the same color and suspended dust, easily misinterpreting bright ore surfaces as high-concentration fog layers. This misinterpretation leads the algorithm to incorrectly dehaze ore areas during actual processing, resulting in abnormal phenomena such as decreased ore surface brightness, large-area blackening, or color oversaturation. Ultimately, this leads to the loss of crucial texture information such as cracks and grain size on the ore surface, rendering the enhanced image ineffective in monitoring production details. Summary of the Invention

[0005] To address the technical problem of poor image enhancement in high-dust environments at mineral processing sites, this invention provides a method for dehazing and enhancing monitoring images in such environments, comprising:

[0006] The process involves acquiring raw images of the ore dressing site, smoothing and denoising them using weighted guided filtering to obtain a clean image, iteratively filtering the clean image using quadtree subdivision, and calculating the global atmospheric light vector. Based on the similarity between the RGB color vectors of local pixels in the clean image and the global atmospheric light vector, the spectral similarity factor of any pixel in the clean image is calculated. Based on the dispersion of local gray levels in the clean image, the texture confidence factor of any pixel in the clean image is calculated. Combining the spectral similarity factor and the texture confidence factor, a heterogeneous color discrimination index for any pixel in the clean image is constructed. The heterogeneous color discrimination index is linearly weighted onto the baseline dehazing value to obtain the dynamic dehazing coefficient for any pixel in the clean image. Using the dark channel prior principle, the dynamic dehazing coefficient is applied to the clean image to calculate the adaptive transmittance of any pixel in the clean image. Based on an atmospheric scattering physics model, the adaptive transmittance and the global atmospheric light vector are used to invert the clean image to obtain a physically restored image. Finally, the physically restored image is enhanced to obtain the final enhanced image.

[0007] This invention constructs a heterogeneous discrimination index that includes spectral similarity factors and texture confidence factors, which can distinguish the regional differences between suspended dust and solid ore. Based on this, a dynamic dehazing coefficient is generated, enabling the algorithm to adaptively reduce the dehazing intensity on the ore area while penetrating thick dust interference. This preserves the color and texture details of the ore surface, avoids image blackening or color distortion caused by excessive dehazing, and improves the visibility of monitoring in harsh industrial environments.

[0008] Preferably, the step of iteratively filtering the clean image using the quadtree subdivision method and calculating the global atmospheric light vector includes:

[0009] The clean image is divided into four rectangular sub-blocks; the difference between the mean and standard deviation of the brightness of each rectangular sub-block is calculated and recorded as the brightness stability index of the corresponding rectangular sub-block; the rectangular sub-block with the largest brightness stability index is selected as the target region for the next iteration; the target region is iterated until the number of pixels of the selected largest rectangular sub-block is less than the size threshold for stopping the iteration, the iteration is stopped, and the largest rectangular sub-block selected in the last iteration is recorded as the final selected sub-block;

[0010] The average RGB three-channel brightness of all pixels within the final selected sub-block is calculated to obtain the global atmospheric light vector; the size threshold for stopping iteration is a preset value.

[0011] Preferably, the spectral similarity factor satisfies the expression:

[0012] ;

[0013] In the formula, The spectral similarity factor of the x-th pixel in a clean image; This represents the color vector of the x-th pixel in the clean image; Represents the global atmospheric light vector; This represents the dot product operator. The sign indicating the magnitude of a vector; This represents the first tiny positive value.

[0014] This invention can accurately identify potential dust-covered areas by calculating the cosine similarity between local pixel color vectors and global atmospheric light vectors. Especially in complex situations where the dust color is extremely similar to the background color, it provides accurate spectral feature basis for subsequent differentiation of dust and ore of the same color.

[0015] Preferably, the texture confidence factor satisfies the expression:

[0016] ;

[0017] In the formula, This represents the texture confidence factor of the x-th pixel in a clean image. This represents the neighborhood grayscale variance of the x-th pixel in the clean image; This represents the mean gray value of the neighborhood of the x-th pixel in the clean image; This represents the second smallest positive value.

[0018] This invention captures the rough texture features of the ore surface and the smooth features of the dust area by calculating the ratio of the variance to the mean of the neighborhood gray level. Thus, when colors are difficult to distinguish, the richness of texture is used as a supplementary criterion, reducing the probability of misjudging textured ore as smooth dust.

[0019] Preferably, the step of combining the spectral similarity factor and the texture confidence factor to construct the isochromatic heterogeneity discrimination index for any pixel in a clean image includes:

[0020] Preset spectral similarity weights and texture confidence weights;

[0021] The product of the spectral similarity weight and the spectral similarity factor of any pixel in the clean image is subtracted from the product of the texture confidence weight and the texture confidence factor of the pixel in the clean image to obtain the homochromatic heterogeneity discrimination index of the pixel in the clean image.

[0022] Preferably, the acquisition of the dynamic defogging coefficient includes:

[0023] A preset base dehazing reference value and sensitivity adjustment coefficient are established; the product of the sensitivity adjustment coefficient and the heterogeneity discrimination index of any pixel in the clean image is added to the base dehazing reference value to obtain the heterogeneity discrimination index of the pixel.

[0024] Preferably, the adaptive transmittance satisfies the expression:

[0025] ;

[0026] In the formula, This represents the adaptive transmittance of the x-th pixel in the clean image; This represents the dynamic dehazing coefficient of the x-th pixel in the clean image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; Describes the minimum value function; This indicates that the minimum value among the three channels R, G, and B is selected during the operation. This represents the third smallest positive value.

[0027] This invention constrains transmittance by introducing a dynamic dehazing coefficient, preventing errors in transmittance estimation caused by the high reflectivity of the ore surface in specific channels. This ensures that the physical model can correctly distinguish between foreground objects and background scattering during inversion, and eliminates common object edge artifacts and halo phenomena in dehazed images.

[0028] Preferably, the acquisition of the physically restored image includes:

[0029] Calculate the channel values ​​of any pixel in the physically restored image;

[0030] The physical reconstruction image is composed of all channel values ​​of all pixels in the physical reconstruction image.

[0031] Preferably, the channel values ​​of any pixel in the physically restored image satisfy the expression:

[0032] ;

[0033] In the formula, This represents the c-channel value of the x-th pixel in the physically restored image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; This represents the adaptive transmittance of the x-th pixel in the clean image; This indicates the lower limit threshold of transmittance; This represents the maximum value function.

[0034] This invention prevents the amplification of white spot noise caused by division operations in areas with low transmittance by setting a lower limit threshold for transmittance. At the same time, by retaining the atmospheric light component as the brightness reference in the inversion formula, the restored image not only removes the obstructed scattered light but also maintains a suitable overall brightness, avoiding the problem that physically restored images are usually quite dark.

[0035] Preferably, the step of enhancing the details of the physically restored image to obtain the final enhanced image includes:

[0036] The restored image is processed using a contrast-limited adaptive histogram equalization algorithm to obtain the final enhanced image.

[0037] The beneficial effects of this invention are as follows:

[0038] (1) In view of the problem of visual confusion caused by the high similarity between the color of dust and the color of ore surface at the mineral processing site, this invention introduces texture frequency domain features. By calculating the spectral similarity factor that measures the spectral distance between the pixel and the atmospheric light / dust color, and the texture confidence factor based on the texture complexity of the local gradient amplitude and entropy value, a same-color heterogeneity discrimination index is constructed in the multi-dimensional feature space, which can capture the essential difference between smooth dust mask and rough ore texture.

[0039] (2) The present invention achieves a precise balance between defogging and fidelity through a dynamic defogging mechanism. While effectively removing interference from high-concentration suspended dust, it retains the key features of the ore surface, avoids image oversaturation and loss of dark details, and provides high-quality image data for subsequent ore identification, particle size analysis and foreign object detection. Attached Figure Description

[0040] Figure 1 This is a flowchart illustrating a method for defogging and enhancing monitoring images in a high-dust environment at a mineral processing site, according to the present invention.

[0041] Figure 2 This is a schematic diagram illustrating a clean image;

[0042] Figure 3 This is a schematic diagram illustrating the final enhanced image. Detailed Implementation

[0043] This invention discloses a method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site, referring to... Figure 1 This includes steps S1-S4:

[0044] S1: Acquire the original image of the ore dressing site, use weighted guided filtering to smooth and denoise the original image to obtain a clean image; use quadtree subdivision method to iteratively filter the clean image and calculate the global atmospheric light vector.

[0045] It should be noted that the environment at the ore processing site is filled with fine dust aerosols and suspended larger ore particles. During imaging, these large suspended particles generate high-frequency speckle noise. If the original image is used directly for subsequent atmospheric light estimation, these high-brightness noise points will be misidentified as ambient light sources, leading to an overestimation of atmospheric light values. This, in turn, results in an overall darker image and color distortion after dehazing. Furthermore, while ordinary mean filtering or Gaussian filtering can remove noise, it blurs the edge details of the ore, affecting subsequent texture feature extraction. Therefore, this invention first utilizes a weighted guided filter with edge-preserving properties to clean the data, providing a high-quality signal source for subsequent steps.

[0046] Specifically, the process involves acquiring raw images of the ore dressing site, using weighted guided filtering to smooth and denoise the raw images, and obtaining clean images, including:

[0047] Industrial cameras are used to capture real-time video streams from the mineral processing site, and the current frame is extracted as the original image.

[0048] Set the filter radius, denoted as... Exemplary Set the regularization parameter, denoted as . Exemplary .

[0049] It's important to note that the filter radius determines the smoothing range. If the filter radius is set too small, it won't cover the diameter of large dust particles, resulting in incomplete noise reduction. If the filter radius is set too large, the image will become too blurry. The regularization parameter controls the smoothing degree. If the value is too large, edge gradient information will be lost. If the value is too small, the threshold for the guide filter to determine non-edge regions will be too low, causing high-frequency speckle noise to be misjudged as valid edge features and preserved, resulting in ineffective noise reduction and failure to effectively filter out visual interference from suspended dust particles.

[0050] A weighted guided filter is used to perform convolutional smoothing on the original image to filter out high-frequency speckle noise and output a clean image.

[0051] It should be noted that, as Figure 2 This is a schematic diagram of a clean image, showing a clean image of a mineral processing site.

[0052] Preferably, the clean image is iteratively filtered using a quadtree subdivision method, and the global atmospheric light vector is calculated, including:

[0053] It is important to note that accurate atmospheric light values ​​are crucial for physical model reconstruction. Traditional dark channel algorithms often directly select the pixels with the highest brightness in the image as references, for example, selecting the pixels with the highest total number of pixels. The average value of the pixels is taken as atmospheric light. However, in mineral processing sites, the surface reflection of metal ore or the direct light from lighting fixtures is often brighter than atmospheric light. Although the number of pixels in these interference points is small, their brightness is high, and direct selection will lead to misjudgment. Considering that the quadtree subdivision method can effectively avoid local bright noise by finding areas with high brightness and gradual brightness changes, this invention uses the quadtree subdivision method to iteratively filter clean images and calculate the global atmospheric light value.

[0054] Set a size threshold to stop iteration; for example, the size threshold to stop iteration is the total number of pixels in the original image. .

[0055] Divide the clean image into four rectangular sub-blocks; calculate the difference between the mean and standard deviation of the brightness of each rectangular sub-block, and record it as the brightness stability index of the corresponding rectangular sub-block; select the rectangular sub-block with the largest brightness stability index as the target region for the next iteration; iterate over the target region until the number of pixels of the selected largest rectangular sub-block is less than the size threshold for stopping iteration, then stop the iteration, and record the largest rectangular sub-block selected in the last iteration as the final selected sub-block.

[0056] It should be noted that the maximum brightness stability index means that the corresponding rectangular sub-block not only has a large average brightness, but also a very uniform internal brightness distribution and a small standard deviation, which is more consistent with the atmospheric photophysical characteristics of the sky or distant dense fog.

[0057] Calculate the average RGB three-channel brightness of all pixels within the final selected sub-block to obtain the global atmospheric light vector, denoted as . , ,in , , These are the global atmospheric light values ​​for the R, G, and B channels, respectively. The transpose symbol for a vector.

[0058] At this point, a clean image and global atmospheric light vectors have been obtained.

[0059] S2: Calculate the spectral similarity factor of any pixel in the clean image based on the similarity between the RGB color vector of local pixels and the global atmospheric light vector; calculate the texture confidence factor of any pixel in the clean image based on the dispersion of local gray levels; and construct the same-color heterogeneity discrimination index of any pixel in the clean image by combining the spectral similarity factor and the texture confidence factor.

[0060] It should be noted that in the beneficiation sites of hematite and copper mines, the color of the dust often closely matches the color of the ore itself; for example, hematite produces red dust. Traditional dehazing algorithms are usually based on the assumption that non-sky areas have extremely low grayscale values ​​in a certain channel. However, in high-dust environments of the same color, the red components on the ore surface are difficult to distinguish from the red components of the dust, causing the algorithm to easily misjudge the ore as dense fog and over-remove it, resulting in the ore surface turning black and losing details. Therefore, this invention introduces two dimensions, spectral similarity and texture confidence, to construct a heterogeneous color discrimination index.

[0061] Specifically, based on the similarity between the RGB color vectors of local pixels in the clean image and the global atmospheric light vector, the spectral similarity factor of any pixel in the clean image is calculated, including:

[0062] A local window is set centered on any pixel of the clean image; for example, the size of the local window is [size missing]. .

[0063] It should be noted that relying solely on pixel color to determine whether a pixel is dust or mineral is insufficient; it is also necessary to consider its spectral and textural characteristics. Spectrally, dust colors closely resemble atmospheric light; texturally, dust areas are smooth, while mineral surfaces are rough.

[0064] The color vector of any pixel is constructed by taking the RGB three-channel values ​​of that pixel. Taking the x-th pixel as an example, the color vector of the x-th pixel is denoted as... , ,in, , , These are the R, G, and B channel values ​​for the x-th pixel. The transpose sign of the vector is used; the cosine similarity between the pixel color vector within the local window of the pixel and the global atmospheric light vector is calculated and denoted as the spectral similarity factor of the pixel.

[0065] The spectral similarity factor of any pixel in a clean image satisfies the expression:

[0066] ;

[0067] In the formula, The spectral similarity factor of the x-th pixel in a clean image; This represents the color vector of the x-th pixel in the clean image; Represents the global atmospheric light vector; This represents the dot product operator. The sign indicating the magnitude of a vector; This represents a first small positive value, used to prevent the denominator from being zero due to the color vector being 0. For example, .

[0068] In the formula, The cosine of the angle between the x-th pixel in the clean image and the global atmospheric light vector was calculated. The closer this value is to the x-th pixel, the better. The closer the neighborhood color of the x-th pixel in a clean image is to atmospheric light, i.e., dust color, the more likely it is to be dust or an area completely covered by dust.

[0069] Preferably, the texture confidence factor of any pixel in the clean image is calculated based on the degree of dispersion of local gray levels, including:

[0070] Calculate the variance and mean of the grayscale values ​​of any pixel in a local window of a clean image, and denot them as the neighborhood grayscale variance and the neighborhood grayscale mean of the pixel, respectively.

[0071] The texture confidence factor of any pixel in a clean image satisfies the expression:

[0072] ;

[0073] In the formula, This represents the texture confidence factor of the x-th pixel in a clean image. This represents the neighborhood grayscale variance of the x-th pixel in the clean image; This represents the mean gray value of the neighborhood of the x-th pixel in the clean image; This represents the second smallest positive value, used to prevent the denominator from being zero. For example, ; This represents the natural exponential function.

[0074] In the formula, This represents the ratio of the neighborhood gray-level variance to the neighborhood gray-level mean of the x-th pixel in a clean image. It is used to measure the richness of texture; the larger the neighborhood gray-level variance, the richer the texture, such as cracks on the surface of a mineral. The closer to This makes the texture confidence factor more and more similar to Conversely, if the region is smooth, then Approaching .

[0075] Preferably, by combining the spectral similarity factor and the texture confidence factor, a homogeneity / heterogeneity discrimination index for any pixel in a clean image is constructed, including:

[0076] It should be noted that relying on a single feature can easily lead to misclassification. For example, smooth red metal might be misidentified as dust, and rough gray walls might be misidentified as ore. Therefore, this invention uses a weighted fusion of spectral and texture features to construct a comprehensive index that characterizes the degree to which each pixel in a clean image needs to be dehazed.

[0077] Set the spectral similarity weight, denoted as . Exemplary Set the texture confidence weight, denoted as . Exemplary It should be noted that, and The value of has been experimentally calibrated. If Setting it too high can cause the algorithm to be overly sensitive to color, misclassifying slightly reddish, non-dust areas as dust; if... Setting the value too high can cause the algorithm to be too lenient with textures, misjudging thin dust as minerals and resulting in incomplete dehazing.

[0078] The heterogeneity discrimination index of any pixel in a clean image satisfies the expression:

[0079] ;

[0080] In the formula, The index representing the heterogeneity of the same color at the x-th pixel in a clean image; Indicates the weight of spectral similarity; The spectral similarity factor of the x-th pixel in a clean image; Indicates the texture confidence weight; This represents the texture confidence factor of the x-th pixel in the clean image.

[0081] In the formula, A positive contribution indicates that the more it resembles dust color, the greater the heterogeneity discrimination index. The negative penalty term indicates that the more it resembles a mineral texture, the smaller the discrimination index for heterogeneous materials of the same color. This means that when the color of a region is close to that of dust and the texture is smooth, the same-color heterogeneity discrimination index increases significantly, and it is judged as high-concentration dust; conversely, if a region is similar in color to dust but has a rich texture, the same-color heterogeneity discrimination index decreases, and it is judged as a region of the same color ore, thus realizing the discrimination of heterogeneous materials in the same-color environment.

[0082] Thus, the same-color heterogeneity discrimination index of each pixel in the clean image has been obtained.

[0083] S3: Based on the same color heterogeneity discrimination index, the basic dehazing benchmark value is linearly weighted to obtain the dynamic dehazing coefficient of any pixel in the clean image; combined with the dark channel prior principle, the clean image is processed using the dynamic dehazing coefficient to calculate the adaptive transmittance of any pixel in the clean image.

[0084] It should be noted that in traditional dark channel prior algorithms, the dehazing coefficient is usually a fixed value, meaning that the same level of dehazing is applied to the entire image. However, in a mineral processing site with a homogeneous environment, applying strong dehazing to the ore area will remove the color components of the ore itself, resulting in oversaturation or blackening. Therefore, this invention utilizes a homogeneous heterogeneity discrimination index to achieve dehazing according to the required degree. Areas identified as dust are dehazed more forcefully to penetrate the obstruction, while areas identified as ore are dehazed less forcefully to preserve the original reflected light information.

[0085] Specifically, the dynamic dehazing coefficient of any pixel in the clean image is obtained by linearly weighting the basic dehazing benchmark value according to the same-color heterogeneity discrimination index, including:

[0086] Set a baseline dehazing value, denoted as... Exemplary Set the sensitivity adjustment coefficient, denoted as . Exemplary .

[0087] It should be noted that, A global minimum level of defogging is defined to ensure that even minor atmospheric light scattering is removed in all areas, including those with mineral deposits. The sensitivity of defogging intensity to changes in dust concentration is determined by the same-color heterogeneity discrimination index. If... Setting it too high, such as 0.9, will result in excessively high base dehazing intensity for mineral entities, causing a sharp drop in overall image brightness, oversaturation or blackening of the mineral surface, and loss of original color details; if... Setting 's' too low, such as 0.3, will fail to effectively remove the slight atmospheric scattering commonly found in scenes, resulting in a hazy image that still appears as if veiled, with little improvement in visual clarity. Setting 's' too high, such as 0.8, will cause the system to be overly sensitive to changes in the same-color heterogeneity discrimination index, leading to drastic fluctuations in the dehazing coefficient across different regions. This can easily cause artifacts or amplified noise at the transition edges between minerals and dust. Setting 's' too low, such as 0.05, will result in insufficient response to high-concentration dust, failing to significantly increase dehazing power based on dust concentration and losing the ability to adaptively penetrate thick dust.

[0088] The dynamic dehazing coefficient of any pixel in a clean image satisfies the expression:

[0089] ;

[0090] In the formula, This represents the dynamic dehazing coefficient of the x-th pixel in the clean image; Indicates the baseline defogging value; This represents the sensitivity adjustment coefficient; It represents the same-color heterogeneity discrimination index of the x-th pixel in a clean image.

[0091] In the formula, This constitutes the gain term. When the same color heterogeneity discrimination index is large and it is located in the high-concentration dust area, the dynamic defogging coefficient increases and the defogging intensity is enhanced. When the same color heterogeneity discrimination index is small and it is located in the ore solid area, the gain term is small or negative, and the dynamic defogging coefficient is maintained at the basic level, thereby protecting the ore surface from being over-treated.

[0092] Preferably, combining the dark channel prior principle, the clean image is processed using a dynamic dehazing coefficient to calculate the adaptive transmittance of any pixel in the clean image, including:

[0093] It's important to note that the dark channel prior theory states that the dark channel value of a haze-free image approaches zero in non-sky regions. However, in mineral processing sites, because the color of dust is highly consistent with the color of ore (e.g., in a red ore dust scene), the reflected light from the ore contains high-intensity specific spectral components, causing the dark channel value in the ore region to be significantly higher, exhibiting characteristics similar to dense fog, thus invalidating the prior hypothesis. If a conventional globally fixed high dehazing coefficient, such as 0.95, is directly used, the algorithm will misjudge the bright, same-colored ore as a thick dust layer and forcibly remove it, resulting in severe color loss or even blackening of the ore surface. Introducing a dynamic dehazing coefficient essentially constructs a confidence weight field for the prior theory. That is, in areas identified as ore, the dehazing coefficient is reduced to decrease the dependence on the dark channel prior, forcibly preserving the original pixel information; in areas identified as dust, the dehazing coefficient is increased to strengthen the prior hypothesis, thereby achieving local adaptive correction of the prior theory in specific same-color scenes and avoiding misprocessing of entity targets.

[0094] The adaptive transmittance of any pixel in a clean image satisfies the expression:

[0095] ;

[0096] In the formula, This represents the adaptive transmittance of the x-th pixel in the clean image; This represents the dynamic dehazing coefficient of the x-th pixel in the clean image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; Describes the minimum value function; This indicates that the minimum value among the three channels R, G, and B is selected during the operation. This represents the third smallest positive value, used to prevent the denominator from being zero. For example, .

[0097] In the formula, This means dividing the RGB three-channel values ​​of the image by the corresponding atmospheric light value of the channel, and taking the minimum of the three results to characterize the fog concentration distribution. This indicates that the concentration distribution of fog was finely adjusted using a dynamic defogging coefficient, ensuring that the true color and details of the ore were preserved while removing dust of the same color.

[0098] At this point, the adaptive transmittance of each pixel in the clean image has been obtained.

[0099] S4: Based on the atmospheric scattering physical model, the clean image is inverted using the adaptive transmittance and global atmospheric light vector to obtain the physically restored image; the physically restored image is enhanced with a limited contrast adaptive histogram equalization algorithm to obtain the final enhanced image.

[0100] It should be noted that after calculating the adaptive transmittance and global atmospheric light vector, the radiance of the scene can be restored through the inverse transformation of the atmospheric scattering model. However, although the dehazed image obtained through physical model inversion removes occlusion, the overall brightness is often too dark due to the removal of atmospheric light superimposed on objects, and the histogram distribution is relatively concentrated, resulting in insufficient visual contrast. Mineral processing production monitoring focuses more on the texture features of ore such as cracks and grain size, thus requiring post-processing enhancement. Ordinary histogram equalization tends to amplify background noise; therefore, this invention uses contrast-limited adaptive histogram equalization for detail enhancement.

[0101] Specifically, based on the atmospheric scattering physics model, the clean image is inverted using the adaptive transmittance and global atmospheric light vector to obtain a physically restored image, including:

[0102] Set a lower limit threshold for transmittance, denoted as . Exemplary It should be noted that when the adaptive transmittance approaches... Direct division can lead to unstable values ​​and produce white spot noise, so a lower limit threshold for transmittance is set.

[0103] The channel values ​​of any pixel in the physically restored image satisfy the expression:

[0104] ;

[0105] In the formula, This represents the c-channel value of the x-th pixel in the physically restored image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; This represents the adaptive transmittance of the x-th pixel in the clean image; This indicates the lower limit threshold of transmittance; This represents the maximum value function.

[0106] In the formula, It is to measure the x-th pixel of the observed clean image. The true light intensity reflected from the object's surface is restored by subtracting the corresponding channel value from the channel value and then dividing by the adaptive transmittance. This means that adding the actual light intensity to the global atmospheric light value of the corresponding channel is to maintain the overall brightness baseline of the physically restored image.

[0107] The physical reconstruction image is composed of all channel values ​​of all pixels in the physical reconstruction image.

[0108] Preferably, a contrast-limited adaptive histogram equalization algorithm is used to enhance the details of the physically restored image to obtain the final enhanced image, including:

[0109] It should be noted that in order to enhance local contrast while suppressing noise amplification, the histogram needs to be cropped and limited.

[0110] The physically restored image is divided into several non-overlapping small blocks. For example, the size of each small block is [size missing]. .

[0111] The frequency of each gray level within any small block is statistically analyzed to obtain a frequency histogram. A truncation threshold is set, and the sum of the frequencies of gray levels exceeding the threshold is obtained. This sum of frequencies is then evenly distributed across all gray level frequencies in the frequency histogram to obtain the processed histogram of the small block. It should be noted that adjusting the gray values ​​of pixels makes the overall frequency histogram smoother, avoiding local noise amplification caused by excessively high frequencies of a single gray level.

[0112] Bilinear interpolation is performed on the processed histograms of all small blocks to merge the boundaries between blocks and eliminate block artifacts.

[0113] The final enhanced image's arbitrary channel values ​​for any pixel satisfy the expression:

[0114] ;

[0115] In the formula, This represents the c-channel value of the x-th pixel in the final enhanced image; This represents a contrast-limited adaptive histogram equalization function. This represents the c-channel value of the x-th pixel in the physically restored image.

[0116] In the formula, the grayscale distribution of the physical restoration image is redistributed through the CLAHE operation, which solves the problem of the physical restoration image being too dark and improves the contrast between the surface texture of the ore and the background, enabling monitoring personnel to clearly identify the grain size and crack features of the ore.

[0117] It should be noted that, as Figure 3 This is a schematic diagram of the final enhanced image, showing the adjusted and enhanced final enhanced image corresponding to the clean image.

[0118] At this point, the final enhanced image was obtained, and the defogging enhancement of the monitoring image in the high-dust environment of the mineral processing site was completed.

[0119] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site, characterized in that, include: The original images of the mineral processing site are acquired, and weighted guided filtering is used to smooth and denoise the original images to obtain clean images. The clean images are iteratively filtered using the quadtree subdivision method, and the global atmospheric light vector is calculated. Calculate the spectral similarity factor of any pixel in the clean image based on the similarity between the RGB color vectors of local pixels and the global atmospheric light vector. In the formula, The spectral similarity factor of the x-th pixel in a clean image; This represents the color vector of the x-th pixel in the clean image; Represents the global atmospheric light vector; This represents the dot product operator. The sign indicating the magnitude of a vector; Represents the first tiny positive value; calculate the texture confidence factor for any pixel in the clean image based on the degree of dispersion of local gray levels in the clean image; By combining the spectral similarity factor and the texture confidence factor, a heterogeneity discrimination index for any pixel in the clean image is constructed; the basic dehazing benchmark value is linearly weighted according to the heterogeneity discrimination index to obtain the dynamic dehazing coefficient for any pixel in the clean image; and the clean image is processed using the dynamic dehazing coefficient based on the dark channel prior principle to calculate the adaptive transmittance of any pixel in the clean image. Based on the atmospheric scattering physics model, the clean image is inverted using the adaptive transmittance and global atmospheric light vector to obtain a physically restored image; the physically restored image is then enhanced to obtain the final enhanced image.

2. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The iterative screening of clean images using the quadtree subdivision method and the calculation of the global atmospheric light vector include: The clean image is divided into four rectangular sub-blocks; the difference between the mean and standard deviation of the brightness of each rectangular sub-block is calculated and recorded as the brightness stability index of the corresponding rectangular sub-block; the rectangular sub-block with the largest brightness stability index is selected as the target region for the next iteration; the target region is iterated until the number of pixels of the selected largest rectangular sub-block is less than the size threshold for stopping the iteration, the iteration is stopped, and the largest rectangular sub-block selected in the last iteration is recorded as the final selected sub-block; The average RGB three-channel brightness of all pixels within the final selected sub-block is calculated to obtain the global atmospheric light vector; the size threshold for stopping iteration is a preset value.

3. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The texture confidence factor satisfies the expression: ； In the formula, This represents the texture confidence factor of the x-th pixel in a clean image. This represents the neighborhood grayscale variance of the x-th pixel in the clean image; This represents the mean gray value of the neighborhood of the x-th pixel in the clean image; This represents the second smallest positive value.

4. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The step of combining the spectral similarity factor and the texture confidence factor to construct a homochromatic heterogeneity discrimination index for any pixel in a clean image includes: Preset spectral similarity weights and texture confidence weights; The product of the spectral similarity weight and the spectral similarity factor of any pixel in the clean image is subtracted from the product of the texture confidence weight and the texture confidence factor of the pixel in the clean image to obtain the homochromatic heterogeneity discrimination index of the pixel in the clean image.

5. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The acquisition of the dynamic defogging coefficient includes: A preset base dehazing reference value and sensitivity adjustment coefficient are established; the product of the sensitivity adjustment coefficient and the heterogeneity discrimination index of any pixel in the clean image is added to the base dehazing reference value to obtain the heterogeneity discrimination index of the pixel.

6. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The adaptive transmittance satisfies the expression: ； In the formula, This represents the adaptive transmittance of the x-th pixel in the clean image; This represents the dynamic dehazing coefficient of the x-th pixel in the clean image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; Describes the minimum value function; This indicates that the minimum value among the three channels R, G, and B is selected during the operation. This represents the third smallest positive value.

7. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The acquisition of the physically restored image includes: Calculate the channel values ​​of any pixel in the physically restored image; The physical reconstruction image is composed of all channel values ​​of all pixels in the physical reconstruction image.

8. The method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 7, characterized in that, The arbitrary channel values ​​of any pixel in the physically restored image satisfy the expression: ； In the formula, This represents the c-channel value of the x-th pixel in the physically restored image; This indicates that the x-th pixel in the clean image is located at... The channel value; Represents the global atmospheric light vector Channel values; This represents the adaptive transmittance of the x-th pixel in the clean image; This indicates the lower limit threshold of transmittance; This represents the maximum value function.

9. A method for dehazing and enhancing monitoring images in a high-dust environment at a mineral processing site according to claim 1, characterized in that, The process of enhancing the details of the physically restored image to obtain the final enhanced image includes: The restored image is processed using a contrast-limited adaptive histogram equalization algorithm to obtain the final enhanced image.

Images