A method, system, device and medium for automatic detection of stripe noise

Abstract

The application discloses a kind of stripe noise automatic detection method, system, equipment and medium, method includes the following steps: to the remote sensing image data to be detected with any band is preprocessed, obtain preprocessed result chart, and calculate the main stripe direction of the preprocessed result chart, according to main stripe direction is carried out coordinate rotation transformation, obtain direction correction chart;After two-level screening, the comprehensive noise detection result is carried out coordinate reverse rotation, restores to original image coordinate system, obtains the noise mask of remote sensing image data to be detected.This application is processed by block to symbiotic image information chart, carries out local stripe detection in each sub-region, and verifies the cross-regional continuity of stripe in global area, utilize the feature that real stripe noise has strong continuity, usually throughout image most or all row, to distinguish stripe noise and ground feature texture, reduce the false detection rate of ground feature texture complex area.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image inspection technology, and in particular to an automatic stripe noise detection method, system, device and medium. Background Technology

[0002] With the rapid development of remote sensing satellite technology, pushbroom imaging sensors have become the mainstream configuration for optical remote sensing satellites. Pushbroom sensors acquire image data by scanning line by line using a linear array detector during satellite flight. However, due to issues such as response differences between individual detector units in the array, inconsistent circuit noise, and drift in on-orbit radiometric calibration parameters, periodic or non-periodic stripe noise along the flight direction is commonly found in remote sensing images. This stripe noise severely affects the visual quality of remote sensing images and subsequent applications; therefore, it is necessary to detect and label stripe noise during image preprocessing.

[0003] Existing stripe noise detection methods primarily employ statistical methods based on inter-column gray-level differences. This involves calculating the gray-level difference between adjacent columns, and determining the presence of stripe noise in a column when the difference exceeds a preset threshold. However, this method cannot distinguish stripe noise from the linear texture features of the ground features themselves. In remote sensing imagery, many ground features exhibit regular linear textures, such as furrows in farmland, well-planned road networks, building arrangements, and river courses. These textures also produce gray-level differences along the column direction, leading to numerous false detections in detection methods based on simple gray-level thresholds. Even more challenging is that in areas with complex ground textures, genuine stripe noise is often masked by the texture itself, its gray-level differences being buried by texture variations, resulting in missed detections.

[0004] True stripe noise is caused by abnormal response of detector units, and therefore has strong continuity in the column direction, usually spanning most or all rows of the image. Linear textures of ground features, on the other hand, are often localized and fragmented, exhibiting spatial discontinuity. Existing detection methods based on local grayscale statistics only make judgments within a single or a few sub-regions, failing to verify whether detected suspected stripes possess cross-regional continuity characteristics, thus making it difficult to eliminate interference from local ground feature textures. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides an automatic stripe noise detection method, system, device and medium to solve the problem that existing detection methods cannot distinguish stripe noise from linear textures of ground features, and thus cannot meet the needs of stripe detection at different scales.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an automatic stripe noise detection method, comprising the following steps: preprocessing an arbitrary band of the remote sensing image data to be detected to obtain a preprocessed result image, calculating the main stripe direction of the preprocessed result image, and performing coordinate rotation transformation based on the main stripe direction to obtain a direction correction image; resampling the direction correction image to obtain multi-scale images, and extracting texture information and grayscale gradient information from the multi-scale images to construct co-occurrence image information maps corresponding to each scale image; performing image block processing on the co-occurrence image information maps, performing local stripe detection in each sub-region, and verifying the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale; performing first-level verification detection on the preliminary noise detection results at each scale to obtain a comprehensive noise detection result, and performing second-level screening on the comprehensive noise detection result; rotating the comprehensive noise detection result after second-level screening in the opposite direction to restore it to the original image coordinate system to obtain a noise mask of the remote sensing image data to be detected.

[0008] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of obtaining the orientation correction map includes: performing orientation gradient statistics on the preprocessed result map, calculating the gradient response intensity at multiple preset orientation angles; selecting the orientation angle with the largest gradient response intensity as the main stripe direction; and performing coordinate rotation transformation on the preprocessed result map according to the angle of the main stripe direction until the main stripe direction is aligned with the image column direction to obtain the orientation correction map.

[0009] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of constructing co-occurrence image information maps corresponding to images at each scale includes: performing cloud and snow detection on each scale image in the multi-scale images to obtain a cloud and snow region mask image; performing grayscale processing on each scale image based on the cloud and snow region mask image to increase the grayscale difference between adjacent image points in the column direction to obtain a preprocessed difference map; performing secondary grayscale processing on the preprocessed difference map to further increase the grayscale difference between adjacent image points in the column direction to obtain a secondary processed difference map; and performing secondary processing on the difference map... Texture features are extracted to obtain a texture feature map. The gradient value of the secondary processing difference map is calculated to obtain a gray-level gradient co-occurrence map. The gray-level gradient co-occurrence map is used to transform the secondary processing difference map, and the signal-to-noise ratio (SNR) information map is calculated in combination with the texture feature map. The texture feature map, gray-level gradient co-occurrence map, and SNR information map are used to transform the secondary processing difference map to obtain a tertiary processing difference map. Land cover classification is performed on the image at each scale to obtain a water body mask. The cloud and snow area mask image and the water body mask are used to remove interference areas from the tertiary processing difference map to obtain the co-occurrence image information map corresponding to that scale.

[0010] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of performing image block processing on the co-occurring image information map includes: dividing the co-occurring image information map at each scale into M×N sub-regions; for each sub-region, dividing the co-occurring image information map into positive line maps and negative line maps according to grayscale values; limiting the line lengths of the positive line maps and negative line maps respectively, retaining lines with line lengths greater than a preset threshold, to obtain a first-processed line map; counting the number of noise pixels and the number of non-interference region pixels in the first-processed line map, and limiting them through a concentration characterization parameter, to obtain a second-processed line map; merging the second-processed results of the positive line map and the negative line map to obtain the local noise detection result of the sub-region.

[0011] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of verifying the cross-regional continuity of stripes in the global region includes: counting the number of sub-regions crossed by each noise line segment in the local noise detection results of each sub-region; retaining noise line segments that cross more than a preset continuity threshold, and removing isolated noise line segments that appear only in a single or a few sub-regions; and obtaining preliminary noise detection results at each scale.

[0012] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of obtaining the comprehensive noise detection result includes: spatially registering the preliminary noise detection results at each scale and mapping the detection results at different scales to a unified coordinate system; counting the number of scales that are marked as noise at each pixel position in the columns of the mapped detection results at each scale; retaining the pixel positions that are detected as noise at at least K scales simultaneously to obtain the comprehensive noise detection result.

[0013] As a preferred embodiment of the automatic stripe noise detection method of the present invention, the step of performing secondary screening on the comprehensive noise detection results includes: based on the preprocessed difference map, calculating the statistical characteristics of the gray-level difference between each noise line segment and its adjacent columns in the comprehensive noise detection results; retaining the noise line segment when the mean square error of the gray-level difference is greater than a preset difference threshold, and removing the noise line segment when the mean square error of the gray-level difference is less than or equal to the preset difference threshold; counting the number of non-interference area pixels and the number of pixels detected as noise in each column; calculating the proportion of noise pixels to non-interference area pixels, as well as the starting and ending row numbers of the noise pixels; retaining the noise detection results of the column when the proportion and row number range meet the concentration condition, otherwise removing the noise detection results of the column; and obtaining the comprehensive noise detection results after secondary screening.

[0014] Secondly, the present invention provides an automatic stripe noise detection system based on single-band texture features, comprising: a correction map acquisition module, used to calculate the main stripe direction of the preprocessed result map, and to perform coordinate rotation transformation based on the main stripe direction to obtain a direction correction map; The co-occurrence image information map construction module is used to extract texture information and grayscale gradient information from multi-scale images and construct co-occurrence image information maps corresponding to each scale image. The preliminary noise detection module performs local stripe detection in each sub-region and verifies the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale. The integrated noise detection module performs a first-level verification test on the preliminary noise detection results at each scale to obtain the integrated noise detection results, and then performs a second-level screening on the integrated noise detection results.

[0015] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the automatic stripe noise detection method.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the automatic stripe noise detection method.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: By constructing multi-scale images and extracting texture and gray-level gradient information, a co-occurrence image information map is formed. This utilizes the differences in texture features between stripe noise and ground feature textures, thereby enhancing the detectability of stripe noise. By segmenting the co-occurrence image information map into blocks, local stripe detection is performed within each sub-region, and the cross-regional continuity of stripes is verified in the global region. This leverages the strong continuity of real stripe noise, which typically runs through most or all rows of the image, while ground feature linear textures are often localized, fragmented, and spatially discontinuous. This distinguishes stripe noise from ground feature textures, reducing the false detection rate in areas with complex ground feature textures. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the overall process of an automatic stripe noise detection method according to an embodiment of the present invention. Detailed Implementation

[0020] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0021] Example 1, referring to Figure 1 As an embodiment of the present invention, an automatic stripe noise detection method is provided, comprising the following steps: S1. Take any band of the remote sensing image data to be detected for preprocessing to obtain a preprocessed result map, calculate the main fringe direction of the preprocessed result map, and perform coordinate rotation transformation based on the main fringe direction to obtain a direction correction map.

[0022] S2. Resample the orientation correction map to obtain multi-scale images, and extract texture information and grayscale gradient information from the multi-scale images to construct co-occurrence image information maps corresponding to each scale image.

[0023] S3. Perform image block processing on the symbiotic image information map, perform local stripe detection in each sub-region, and verify the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale.

[0024] S4. Perform first-level verification testing on the preliminary noise detection results at each scale to obtain the comprehensive noise detection results, and then perform second-level screening on the comprehensive noise detection results.

[0025] S5. Rotate the coordinates of the comprehensive noise detection results after secondary screening in reverse to restore them to the original image coordinate system, and obtain the noise mask of the remote sensing image data to be detected.

[0026] It should be noted that during the imaging process of pushbroom remote sensing sensors, issues such as response differences among detector units in the detector array, inconsistent circuit noise, and drift in on-orbit radiometric calibration parameters lead to the prevalence of stripe noise along the flight direction in remote sensing images. However, existing detection methods based on statistical analysis of inter-column grayscale differences cannot distinguish stripe noise from the linear texture features of ground objects, easily resulting in numerous false detections in areas with regular linear textures, such as farmland furrows, road networks, and building arrangements. Furthermore, in areas with complex ground textures, the grayscale differences of the actual stripe noise are often masked by variations in ground texture, leading to missed detections. In addition, existing methods only make judgments in local areas and cannot verify whether detected suspected stripes have cross-regional continuity characteristics, making it difficult to eliminate interference from local ground textures.

[0027] Therefore, to address the aforementioned issues of accuracy and robustness in stripe noise detection, steps S1-S5 are employed. First, multi-scale images are constructed, and texture and gray-level gradient information are extracted to form a co-occurrence image information map. This leverages the differences in texture features between stripe noise and ground feature textures to enhance the detectability of stripe noise. Second, by segmenting the image into blocks and verifying the cross-regional continuity of stripes globally, the strong continuity and widespread presence of true stripe noise across most rows of the image, compared to the local fragmentation and spatial discontinuity of linear ground feature textures, effectively distinguishes stripe noise from ground feature textures. Third, through multi-scale primary verification and secondary screening, stripe noise in remote sensing images is detected, reducing the false positive and false negative rates in areas with complex ground feature textures.

[0028] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, an automatic stripe noise detection method is provided.

[0029] In the embodiments of this application, S1, an arbitrary band of the remote sensing image data to be detected is preprocessed to obtain a preprocessed result map, and the main fringe direction of the preprocessed result map is calculated. Based on the main fringe direction, a coordinate rotation transformation is performed to obtain a direction correction map.

[0030] Choose any band from the remote sensing image data to be tested, such as the visible light band or the near-infrared band. Preprocess the image data in this band. The preprocessing steps include, but are not limited to, radiometric calibration, atmospheric correction, and geometric correction, to eliminate the effects of sensor response, atmospheric scattering, and geometric distortion, and obtain the preprocessed image.

[0031] Specifically, the steps for obtaining the orientation correction map include S1.1 to S1.3: S1.1 Perform directional gradient statistics on the preprocessed result image and calculate the gradient response intensity under multiple preset directional angles.

[0032] Specifically, multiple preset directional angles are set, for example, from 0° to 180°, and sampling is performed at certain angular intervals, such as 5°. For each preset directional angle θ, the image gradient in that direction is calculated. The gradient values ​​of the entire image in direction θ are accumulated or squared to obtain the gradient response intensity G(θ) in that direction. In this embodiment, the formula for calculating the gradient response intensity is: ; in, This represents the gradient value of the image at pixel position (x, y) along direction θ, summed over all pixels of the entire image. Stripe noise has a weak gradient response in its main extension direction and a strong gradient response perpendicular to the stripe direction; therefore, the direction with the strongest gradient response is the perpendicular direction of the stripes.

[0033] For example, directional gradient statistics are performed on the preprocessed visible light band image of a remote sensing image. Gradient response intensities are calculated at 37 directional angles (0°, 5°, 10°, ..., 175°, 180°) at 5° intervals. For the directional angle θ = 115°, the gradient value of each pixel in that direction is calculated, and then the squares of the gradient values ​​of all pixels are summed to obtain G(115°) = 1250. The gradient response intensities of the other 36 directional angles are calculated using the same method.

[0034] S1.2 Select the direction angle with the largest gradient response intensity as the main fringe direction.

[0035] Iterate through all preset direction angles to find the direction angle corresponding to the maximum value of the gradient response intensity G(θ). This directional angle is perpendicular to the stripe noise; therefore, the main stripe direction is... According to the coordinate system definition, this method allows us to identify the main direction of extension of stripe noise in the image.

[0036] For example, comparing the gradient response intensities in the above 37 directional angles, it was found that the maximum value was found in the direction θ=115°, where the gradient response intensity G(115°)=1250, and the gradient response intensities in other directions were all less than 1250. Therefore, it was determined that... The vertical direction of the stripe noise. According to the coordinate system definition, the main stripe direction is 115°-90°=25°, that is, the stripe noise has a 25° tilt angle relative to the column direction of the image.

[0037] S1.3. Based on the angle of the main stripe direction, perform coordinate rotation transformation on the preprocessed result image until the main stripe direction is aligned with the image column direction to obtain the orientation correction image.

[0038] Specifically, the required rotation angle α is calculated based on the main fringe direction, ensuring that the fringe direction is aligned with the column direction (vertical direction) of the image after rotation. The preprocessed image is rotated using an image rotation transformation matrix with a rotation angle of α. The rotation transformation matrix is: ; ; Where (x,y) are the original image coordinates, and (x',y') are the rotated image coordinates. All pixels in the preprocessed image are mapped according to the rotation transformation matrix, and resampling is performed using methods such as bilinear interpolation to obtain the orientation correction image.

[0039] For example, based on the determined main stripe direction of 25°, the required counterclockwise rotation angle α = -25° is calculated. In this embodiment, the negative sign indicates counterclockwise rotation, aligning the stripe direction with the image column direction. A rotation transformation matrix is ​​applied to the preprocessed image, rotating all pixel coordinates by the rotation angle α = -25°. The rotated pixel values ​​are then resampled to obtain the orientation correction image. In the orientation correction image, the stripe noise that was originally tilted at 25° is now vertically distributed and perfectly aligned with the image column direction.

[0040] S2. Resample the orientation correction map to obtain multi-scale images, and extract texture information and grayscale gradient information from the multi-scale images to construct co-occurrence image information maps corresponding to each scale image.

[0041] Multi-scale resampling is performed on the orientation correction map. The orientation correction map is resampled according to different sampling ratios to obtain images at multiple scales with different resolutions. For example, setting sampling ratios of 1:1, 1:2, 1:4, and 1:8 yields images at the original scale, half-scale, 1 / 4-scale, and 1 / 8-scale, respectively. Multi-scale images can capture the features of stripe noise at different spatial resolutions, enhancing the robustness of detection. Stripe noise exhibits different response characteristics at different scales, while ground texture is often smoothed or weakened after downsampling.

[0042] The steps for constructing co-occurrence image information maps corresponding to images at each scale include S2.1 to S2.8: S2.1 Perform cloud and snow detection on each scale image in the multi-scale image to obtain a cloud and snow region mask image.

[0043] In this embodiment, a threshold segmentation method is used for cloud and snow region detection. Because clouds and snow have high reflectivity, their grayscale values ​​in the visible light band are typically high. A cloud and snow detection threshold is set, and pixels with grayscale values ​​greater than this threshold are marked as cloud and snow regions. Morphological processing, such as dilation and erosion operations, is performed on the detection results to remove isolated noise and smooth cloud and snow boundaries. A cloud and snow region mask image is obtained, where pixel values ​​in cloud and snow regions are 1, and pixel values ​​in non-cloud and snow regions are 0. Due to their high brightness and uniformity, cloud and snow regions are easily confused with stripe noise, therefore they need to be excluded in subsequent processing.

[0044] For example, cloud and snow detection is performed on a 1 / 2 scale image, with a detection threshold of 200. All pixels in the image are traversed, and pixels with a grayscale value greater than 200 are marked as cloud and snow regions. The detection results show that the upper right corner of the image, accounting for approximately 15% of the total area, is marked as cloud and snow. A 3×3 structuring element dilation operation is performed on the detection results, expanding the cloud and snow boundary by 2 pixels to ensure that uncertain areas at the cloud and snow edges are also excluded. This yields a mask image of the cloud and snow regions at this scale, where the pixel value of the cloud and snow regions is 1, and the pixel value of the remaining regions is 0.

[0045] S2.2. Based on the mask image of the cloud and snow region, perform grayscale processing on the image at each scale to increase the grayscale difference between adjacent image points in the column direction, and obtain the preprocessed difference map.

[0046] For each scale of the image, calculate the gray-level difference between adjacent columns. For pixel I(i,j) in the j-th column and i-th row of the image, calculate its gray-level difference with adjacent column pixels I(i,j-1) or I(i,j+1). To enhance the response to stripe noise, the grayscale difference is enhanced, for example, by using absolute value operations. Based on the mask image of the cloud and snow regions, the pixel differences in the cloud and snow regions are set to 0 to avoid interference from the cloud and snow regions on the difference image. This yields the preprocessed difference image.

[0047] Specifically, for a pixel in the 100th column of the image, calculate its grayscale difference with the corresponding pixel in the 99th column. For example, if the grayscale value of the pixel in the 50th row of the 100th column is 120, and the grayscale value of the pixel in the 50th row of the 99th column is 115, then the difference is... Take the absolute value of the difference and multiply it by the enhancement factor of 2 to obtain the enhanced difference. Based on the masked image of the cloud and snow region, it was found that column 100, row 50 is located in the non-cloud and snow region, so this difference is preserved. The same processing is applied to all pixels in the image to obtain a preprocessed difference map. In the preprocessed difference map, columns with stripe noise show larger difference responses.

[0048] S2.3. Perform secondary grayscale processing on the preprocessed difference map to further increase the grayscale difference between adjacent image points in the column direction, and obtain the secondary processed difference map.

[0049] For the pixel in the j-th column and i-th row of the preprocessed difference map Calculate its relationship with adjacent column pixels The difference This step is equivalent to performing a second-order difference operation on the original image, which can further highlight the edge features of stripe noise. The absolute value of the difference is then calculated to obtain the secondary processed difference map. In the secondary processing difference map, the boundary response of stripe noise is more obvious, while the response of gently changing ground texture is further suppressed.

[0050] For example, the difference in column 100, row 50 of the preprocessed difference graph. and the difference between column 99 and row 50. Calculate the quadratic difference Taking the absolute value, we get D2(50,100)=7. We then perform the same second-order difference processing on all pixels in the preprocessed difference map to obtain the second-order difference map. In the second-order difference map, a stronger response is observed at the boundary between the stripe noise column and the normal column, with the difference value increasing from the original 5-10 to 7-15.

[0051] S2.4 Extract texture features from the secondary processing difference map to obtain a texture feature map, calculate the gradient value of the secondary processing difference map, and obtain a gray-level gradient co-occurrence map.

[0052] Specifically, texture feature extraction employs the gray-level co-occurrence matrix (GLCM) method to calculate texture feature parameters. A sliding window size is set, for example, 7×7. Within the window, the GLCM is calculated, and texture feature values ​​are extracted. These texture feature values ​​are assigned to the center pixel of the window, and the entire image is traversed to obtain the texture feature map. Striped noise regions, due to their regular linear structure, exhibit texture features that differ from those of random ground features. Simultaneously, the image gradient is calculated from the secondary processing difference map. The horizontal gradient Gx and vertical gradient Gy are calculated separately, and then the gradient magnitudes are calculated. By jointly statistically analyzing the gradient magnitude and grayscale value, a grayscale gradient co-occurrence matrix is ​​constructed, resulting in a grayscale gradient co-occurrence map.

[0053] For example, a 7×7 sliding window is set up, and the gray-level co-occurrence matrix is ​​calculated within the window. For a 7×7 window centered at the 50th row of the 100th column, the gray-level co-occurrence matrices are calculated in four directions: 0°, 45°, 90°, and 135°, and then the contrast feature value is extracted. The calculation result shows that the contrast of this window is 12.5. This value is assigned to the pixel in the 50th row of the 100th column to obtain the contrast value of that pixel in the texture feature map. The entire image is traversed to obtain the complete texture feature map. At the same time, for the pixel in the 50th row of the 100th column, the horizontal gradient Gx=8, the vertical gradient Gy=6, and the gradient magnitude... The grayscale value of this pixel Pairing with gradient magnitude G=10, statistically analyzing the gray-level-gradient pair distribution of the entire image, constructing a gray-level gradient co-occurrence matrix, and obtaining a gray-level gradient co-occurrence map.

[0054] S2.5. Transform the secondary processing difference map using the gray-level gradient co-occurrence map, and calculate the signal-to-noise ratio information map by combining it with the texture feature map.

[0055] Specifically, based on the gray-level gradient co-occurrence map, the signal-to-noise ratio (SNR) at each pixel location is calculated. SNR is defined as the ratio of signal strength to noise intensity. For stripe noise detection, signal strength can be represented by the gray-level values ​​of pixels in the secondary processing difference map, and noise intensity can be represented by the standard deviation or variance of gray-level values ​​within a local window. The calculation formula is: ; in, For secondary processing of pixels in the difference image grayscale value, This represents the standard deviation of grayscale values ​​within a local window centered on that pixel. The signal-to-noise ratio is then weighted and adjusted based on the texture feature map.

[0056] In practice, the signal-to-noise ratio (SNR) is calculated for the second-order difference image at a 1 / 2 scale. For the pixel in the 50th row of the 100th column, its grayscale value D2(50,100) = 7. A 5×5 window is set centered on this pixel, and the standard deviation of the grayscale values ​​of the 25 pixels within the window is calculated, resulting in σ(50,100) = 2.1. The SNR is then calculated. Based on the contrast value of 12.5 for this pixel in the texture feature map, a contrast threshold of 10 is set. Since 12.5 > 10, it is considered that there may be stripe noise in this area. Therefore, the signal-to-noise ratio is adjusted by a factor of 1.2 to obtain the adjusted signal-to-noise ratio. The same processing is applied to the entire image to obtain a signal-to-noise ratio (SNR) map. In the SNR map, areas with striped noise show higher SNR values, while areas with normal ground features show lower SNR values.

[0057] S2.6. Transform the secondary processing difference map using the texture feature map, gray-level gradient co-occurrence map, and signal-to-noise ratio information map to obtain the tertiary processing difference map.

[0058] The formula for transforming the secondary processing difference map is expressed as follows: ; in, For pixels in the difference image processed three times The value, To perform secondary processing on the values ​​of corresponding pixels in the difference image, This represents the texture feature value of the corresponding pixel in the texture feature map. This represents the gradient magnitude of the corresponding pixel in the grayscale gradient co-occurrence map. w1, w2, and w3 are the signal-to-noise ratios of the corresponding pixels in the signal-to-noise ratio information map, and w1, w2, and w3 are weighting coefficients.

[0059] In the example, the weighting coefficients are set to w1=0.1, w2=0.05, and w3=0.8. The difference between the three treatments is calculated, as follows: .

[0060] In the three-processing difference map, the response value of the striped noise area increased from 7 in the second processing to 14.0, while the response value of the normal ground feature area remained at a low level (2-4), and the distinction between striped noise and ground feature texture was improved.

[0061] S2.7. Classify land features for each scale of image to obtain water body masks.

[0062] For each image scale, spectral features or classification algorithms are used to classify land cover and identify water bodies. Water bodies have low reflectivity in the near-infrared band and can be detected using the Normalized Water Index (NDWI). The NDWI is calculated as follows: ; in, For green light band reflectivity, The reflectance is in the near-infrared band. An NDWI threshold is set, and pixels with NDWI values ​​greater than this threshold are marked as water regions. Morphological processing is performed on the detection results to remove isolated noise points and smooth water boundaries, resulting in a water mask.

[0063] S2.8. Using the cloud and snow area mask image and the water body mask, the interference area is removed from the three-stage processing difference map to obtain the co-occurrence image information map corresponding to this scale.

[0064] In the composite interference area mask, the pixel value of cloud, snow, or water areas is 1, while the pixel value of other areas is 0. For the three-stage processing difference map, the difference values ​​corresponding to the positions with pixel values ​​of 1 in the composite interference area mask are set to 0 or invalid values ​​to remove the influence of these interference areas. This yields the co-occurrence image information map corresponding to this scale, which retains the stripe noise response information of the non-interference areas.

[0065] For example, a logical OR operation is performed on the cloud / snow area mask image (cloud / snow area occupies 15%) and the water body mask (water body area occupies 8%) to obtain a comprehensive interference area mask. In the comprehensive interference area mask, the pixel value of the cloud / snow or water body areas is 1, accounting for approximately 23% of the total area, while the pixel value of the remaining 77% of the area is 0. For the pixel in column 100, row 50 of the three-processing difference map, its difference D3(50, 100) = 14.0. Querying the comprehensive interference area mask, the pixel value at this position is found to be 0, indicating a non-interference area, so this difference value is retained. For the pixel in the cloud / snow area in the upper right corner of the image, such as column 500, row 100, querying the comprehensive interference area mask, its pixel value is found to be 1, indicating an interference area, so the value at this position in the three-processing difference map is set to 0. The same masking process is performed on the entire three-processing difference map to obtain the co-occurrence image information map corresponding to this scale. In the co-occurrence image information map, the cloud / snow and water body areas are removed, and only the stripe noise response information of the normal ground feature areas is retained.

[0066] S3. Perform image block processing on the symbiotic image information map, perform local stripe detection in each sub-region, and verify the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale.

[0067] The steps of performing image block processing on the co-occurring image information map include S3.1 to S3.5: S3.1 Divide the co-occurrence image information map at each scale into M×N sub-regions.

[0068] Assume the size of the co-occurrence image infographic is 2000×2000 pixels. Set M=8 and N=8, divide the image into 8×8=64 sub-regions. Each sub-region is 250×250 pixels in size. After division, each sub-region has a unique number.

[0069] S3.2 For each sub-region, the co-occurrence image information map is divided into positive line map and negative line map according to the gray value.

[0070] For each sub-region, analyze the grayscale characteristics of each column in the co-occurring image infographic. Stripe noise manifests as a systematic deviation in the grayscale values ​​of certain columns relative to surrounding columns. This deviation could be positive (higher grayscale values ​​than surrounding columns) or negative (lower grayscale values ​​than surrounding columns). Calculate the average grayscale value of each column and compare it with the average grayscale values ​​of all columns within that sub-region or with the average grayscale values ​​of adjacent columns. When the average grayscale value of a column is higher than a reference value, the column is marked as a positive line and included in the positive line graph; when the average grayscale value of a column is significantly lower than the reference value, the column is marked as a negative line and included in the negative line graph.

[0071] The positive and negative line graphs are two binary images, where the labeled column pixels have a value of 1, and the remaining column pixels have a value of 0. This separation process can detect bright and dark stripes separately, improving the comprehensiveness of the detection.

[0072] For example, positive and negative lines are separated in the sub-region (3,4), which covers the row range [501,750] and column range [751,1000], containing 250 columns. The average gray value of all 250 columns in this sub-region is calculated, resulting in an average gray value of 8.5. For column 850, the average gray value of the 250 rows of pixels in this column within the sub-region is calculated, resulting in 12.3. Comparing 12.3 with the average gray value of 8.5, the difference is 3.8. The positive line detection threshold is set to 2.5. Since 3.8 > 2.5, column 850 is marked as a positive line. In the positive line image, all rows of column 850, i.e., rows 501 to 750, are set to a value of 1. For column 920, its average gray value is calculated to be 4.2, with a difference of -4.3. The negative line detection threshold is set to -2.5. Since -4.3 < -2.5, column 920 is marked as a negative line. In the negative line image, all rows in column 920 (rows 501 to 750) have a pixel value of 1. The same process is applied to all 250 columns within the sub-region, resulting in both positive and negative line images for that sub-region.

[0073] S3.3. Limit the line length of both the positive and negative line graphs, and retain lines whose length is greater than a preset threshold to obtain a processed line graph.

[0074] For both positive and negative line graphs, the continuous length of each line along the row direction is calculated. Line length is defined as the number of consecutive pixels marked as 1 in that column. True stripe noise typically exhibits long continuity along the row direction. In contrast, feature textures or random noise often appear only within local row ranges, resulting in shorter line lengths. A line length threshold is set, retaining lines with lengths greater than the threshold and discarding short lines with lengths less than or equal to the threshold.

[0075] S3.4. Count the number of noise pixels and non-interference area pixels in the first-processed line graph, and limit them by the concentration characterization parameter to obtain the second-processed line graph.

[0076] Specifically, the number of pixels marked as noise and the number of pixels in the non-noise area of ​​each line are counted. A concentration characteristic parameter is calculated, defined as the ratio of noise pixels to non-noise area pixels: ; Where Nnoise represents the number of noise pixels, and Nvalid represents the number of pixels in the non-interference area. Concentration C reflects the degree of concentration of stripe noise in this column. True stripe noise usually appears throughout the entire column or most rows, thus having a high concentration; while ground texture or isolated noise often appears only in a few rows, resulting in a low concentration. A concentration threshold is set, retaining lines with a concentration greater than the threshold and discarding lines with a concentration less than or equal to the threshold. Concentration filtering is applied to the first-processed line map to obtain a second-processed line map. The second-processed line map further eliminates suspected stripes with uneven noise distribution, improving detection accuracy.

[0077] S3.5. Combine the secondary processing results of the positive and negative line graphs to obtain the local noise detection results for this sub-region.

[0078] The secondary processed line graphs of positive and negative lines are logically ORed and merged into a unified local noise detection result image. In the merged image, the column pixel value marked by positive or negative lines is 1, and the pixel value of the remaining columns is 0. The merged local noise detection result includes all filtered suspected stripe noise columns in the sub-region, including both bright and dark stripes. The start row number, end row number, and column number of each noise line segment are recorded to provide a data foundation for subsequent cross-regional continuity verification.

[0079] The steps for verifying the cross-regional continuity of fringes in the global region include S3.6~S3.8: S3.6 Count the number of sub-regions crossed by each noise line segment in the local noise detection results of each sub-region.

[0080] Specifically, the local noise detection results of all sub-regions are summarized, and the noise line segments are statistically analyzed globally according to their column numbers. For each column, the number of sub-regions in which that column is marked as noise is counted. Since the image is divided into M×N sub-regions, the same column may appear in sub-regions in multiple row directions. Traversing all M×N sub-regions, for each column j (1≤j≤number of image columns), the number of sub-regions in which that column is marked as noise is counted, denoted as the cross-region number S(j). True stripe noise, due to its strong continuity, is usually detected simultaneously in multiple sub-regions, resulting in a large cross-region number; while local ground texture often appears only in a single or a few sub-regions, resulting in a smaller cross-region number. By counting the cross-region number, stripe noise and ground texture can be distinguished.

[0081] S3.7. Retain noise line segments that span more than the preset continuity threshold, and remove isolated noise line segments that appear only in a single or a few sub-regions.

[0082] A continuity threshold is set, representing the minimum number of sub-regions that a noise line segment needs to cross. For each column j, its number of crossed regions S(j) is compared with the continuity threshold. When S(j) is greater than the continuity threshold, the column is considered to have cross-region continuity and is retained as true stripe noise. Otherwise, the column is considered to be only local feature texture or isolated noise and is discarded. Through this global continuity verification, the interference of local feature textures can be eliminated, and only stripe noise that continuously appears over a large area of ​​the image can be retained.

[0083] S3.8 Obtain preliminary noise detection results at each scale.

[0084] S4. Perform first-level verification testing on the preliminary noise detection results at each scale to obtain the comprehensive noise detection results, and then perform second-level screening on the comprehensive noise detection results.

[0085] The steps for obtaining the comprehensive noise detection result include S4.1 to S4.3: S4.1 Perform spatial registration on the preliminary noise detection results at each scale, mapping the detection results at different scales to a unified coordinate system.

[0086] Specifically, since multi-scale images have different resolutions and sizes, it is necessary to map the preliminary noise detection results at each scale to a unified coordinate system for comparison and fusion. The original scale is chosen as the reference coordinate system. For the preliminary noise detection results at other scales, an upsampling operation is performed to map them to the resolution of the original scale. For each pixel location in the detection results... Calculate its corresponding position in the original scale coordinate system. The mapping relationship is as follows: ; ; For example, for the detection results of the 1 / 2 scale, the sampling ratio is 2, so mapped to . After spatially registering the preliminary noise detection results of all scales, a set of multi-scale detection results in a unified coordinate system is obtained.

[0087] S4.2. For the detection results of each scale after mapping, count the number of scales in which each pixel position is marked as noise column by column.

[0088] Specifically, traverse the detection results of all scales. For each column j, check whether it is marked as noise in the detection results of each scale such as the original scale, 1 / 2 scale, 1 / 4 scale, 1 / 8 scale, etc.

[0089] Accumulate the number of scales marked as noise, denoted as V(j). The value range of V(j) is from 0 to the total number of scales. For example, if 4 scales are used, the value range of V(j) is from 0 to 4. The larger V(j) is, the more consistently the column is detected as stripe noise at multiple scales, and the higher the credibility. The smaller V(j) is, the fewer scales the column is detected in.

[0090] S4.3. Retain the pixel positions that are detected as noise by at least K scales to obtain the comprehensive noise detection result.

[0091] Specifically, set the multi-scale verification threshold K, which represents that a column position needs to be detected as noise in at least K scales to be retained in the comprehensive detection result. For each column j, compare its scale marking number V(j) with the verification threshold K. When V(j) ≥ K, it is considered that the column is consistently detected at multiple scales and is retained as real stripe noise. When V(j) < K, it is considered that the detection result of the column is not stable enough and may be a false detection, so it is excluded.

[0092] The steps for secondary screening of the comprehensive noise detection result include S4.4~S4.9: S4.4. Based on the preprocessed difference map, calculate the gray-scale difference statistical features between each noise line segment in the comprehensive noise detection result and its adjacent columns before and after.

[0093] Specifically, for each noise line segment in the comprehensive noise detection result, extract the gray-scale values of this noise column and its adjacent columns before and after through the original-scale preprocessed difference map generated in step S2.2. For the j-th column, extract the gray-scale value sequences of the j - 1-th column, the j-th column, and the j + 1-th column in the preprocessed difference map.

[0094] Calculate the grayscale difference Δj-1 between column j and column j-1, and the grayscale difference Δj+1 between column j and column j+1. Perform statistical analysis on these grayscale differences and calculate statistical characteristic parameters, including but not limited to mean, standard deviation, maximum value, and minimum value.

[0095] S4.5 When the mean square error of the grayscale difference is greater than the preset difference threshold, the noise segment is retained; when the mean square error of the grayscale difference is less than or equal to the preset difference threshold, the noise segment is removed.

[0096] Set a grayscale difference threshold, which represents the minimum requirement for the mean square error of the grayscale difference between the noise column and its adjacent columns.

[0097] For each noise line segment, compare its mean square error of grayscale difference σ with the difference threshold. When σ > grayscale difference threshold, the column is considered to have a significant grayscale difference from adjacent columns and is retained as true stripe noise; when σ ≤ grayscale difference threshold, the grayscale difference of the column is considered not significant enough, and may be a false detection or ground feature boundary, and is removed.

[0098] S4.6 Count the number of non-interference area pixels and the number of pixels detected as noise in each column.

[0099] For each noise line segment retained after grayscale difference verification, count the number of pixels in the non-interference area and the number of pixels detected as noise in that column. The number of pixels in the non-interference area refers to the number of effective pixels in that column after excluding clouds, snow, and water bodies, which can be obtained by querying the cloud and snow area mask images and water body masks generated in steps S2.1 and S2.7. The number of pixels detected as noise refers to the number of pixels in that column that are marked as noise in the overall noise detection results.

[0100] S4.7 Calculate the proportion of noise pixels to non-interference area pixels, as well as the start and end row numbers of the noise pixels.

[0101] For each noise line segment, calculate the proportion R of noise pixels to non-interference region pixels: ; Where Nnoise represents the number of pixels detected as noise, and Nvalid represents the number of pixels in the non-interference area. The ratio R reflects the coverage of stripe noise in this column. True stripe noise typically appears throughout the entire column or in most of the non-interference area, thus the ratio R is higher. Localized feature textures or incomplete stripes often appear only in certain areas, resulting in a lower ratio R. Simultaneously, the start and end row numbers of the noise pixels are calculated. Traversing all pixels marked as noise in the column, the row number of the first noise pixel is used as the start row number Rstart, and the row number of the last noise pixel is used as the end row number Rend. The row number range of the noise pixels is calculated as follows: ; The line number range ΔR reflects the continuous span of noise along the line direction. Realistic striped noise typically has a large line number range, close to or equal to the total number of lines in the image. In contrast, local textures have a smaller line number range.

[0102] S4.8 When the ratio and row number range meet the concentration condition, retain the noise detection results of that column; otherwise, discard the noise detection results of that column.

[0103] Concentration conditions include a proportion threshold and a row number range threshold. The proportion threshold represents the minimum required proportion of noise pixels to non-interference pixels, for example, set to 0.7. The row number range threshold represents the minimum required row number range for noise pixels, for example, set to 70% of the total number of rows in the image. For each noise line segment, check whether its proportion R and row number range ΔR simultaneously satisfy the following conditions: R ≥ proportional threshold and ΔR ≥ row number range threshold; When both conditions are met, the noise distribution in that column is considered concentrated and continuous, and it is retained as true stripe noise. When either condition is not met, the noise distribution in that column is considered insufficiently concentrated or discontinuous, possibly a false detection or local texture, and is therefore discarded. This concentration screening further eliminates false detection columns with incomplete or discontinuous noise distribution, ensuring high quality of the final detection results.

[0104] S4.9 Obtain the comprehensive noise detection results after secondary screening.

[0105] S5. Rotate the coordinates of the comprehensive noise detection results after secondary screening in reverse to restore them to the original image coordinate system, and obtain the noise mask of the remote sensing image data to be detected.

[0106] To apply the detection results to the original remote sensing image data, the results need to be rotated in reverse to restore the original image coordinate system. Based on the rotation angle α determined in step S1.3, a reverse rotation of -α is performed. The coordinates of the comprehensive noise detection results after secondary screening are rotated using an image rotation transformation matrix: ; ; in, These are the coordinates in the orientation correction diagram coordinate system. The coordinates are in the original image coordinate system. All pixels from the comprehensive noise detection results after secondary screening are mapped to their coordinates using an inverse rotation transformation matrix, and resampled using nearest neighbor interpolation to obtain a noise mask in the original image coordinate system. The noise mask is a binary image of the same size as the original remote sensing image data, where pixels at stripe noise locations have a value of 1, and pixels at normal locations have a value of 0.

[0107] In summary, by constructing multi-scale images and extracting texture and grayscale gradient information to form a co-occurrence image information map, the differences in texture features between stripe noise and ground feature textures are utilized, thereby enhancing the detectability of stripe noise. By segmenting the co-occurrence image information map into blocks, local stripe detection is performed within each sub-region, and the cross-regional continuity of stripes is verified in the global region. This leverages the strong continuity of real stripe noise, which typically runs through most or all rows of the image, while ground feature linear textures are often localized, fragmented, and spatially discontinuous, thus distinguishing stripe noise from ground feature textures and reducing the false detection rate in areas with complex ground feature textures.

[0108] Example 3 illustrates an automatic stripe noise detection method. It should be noted that the technical solution of this automatic stripe noise detection system based on single-band texture features is based on the same concept as the automatic stripe noise detection method described above. Details not described in detail in this embodiment can be found in the description of the automatic stripe noise detection method described above.

[0109] This embodiment also provides an automatic stripe noise detection system based on single-band texture features, including: The calibration map acquisition module is used to calculate the main fringe direction of the preprocessed result map, and perform coordinate rotation transformation based on the main fringe direction to obtain the orientation calibration map. The co-occurrence image information map construction module is used to extract texture information and grayscale gradient information from multi-scale images and construct co-occurrence image information maps corresponding to each scale image. The preliminary noise detection module performs local stripe detection in each sub-region and verifies the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale. The integrated noise detection module performs a first-level verification test on the preliminary noise detection results at each scale to obtain the integrated noise detection results, and then performs a second-level screening on the integrated noise detection results.

[0110] This embodiment also provides an electronic device suitable for automatic stripe noise detection based on single-band texture features, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the automatic stripe noise detection method proposed in the above embodiment.

[0111] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the automatic stripe noise detection method proposed in the above embodiments.

[0112] The storage medium proposed in this embodiment belongs to the same inventive concept as the automatic stripe noise detection method proposed in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0113] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0114] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automatic method for detecting stripe noise, characterized in that, Includes the following steps: A band of the remote sensing image data to be detected is randomly selected for preprocessing to obtain a preprocessed result image. The main fringe direction of the preprocessed result image is calculated, and a coordinate rotation transformation is performed based on the main fringe direction to obtain a direction correction image. The orientation correction map is resampled to obtain multi-scale images, and texture information and gray-level gradient information are extracted from the multi-scale images to construct co-occurrence image information maps corresponding to each scale image. The co-occurring image information map is processed by image block segmentation, local stripe detection is performed in each sub-region, and the cross-regional continuity of stripes is verified in the global region to obtain preliminary noise detection results at each scale; The preliminary noise detection results at each scale are subjected to a first-level verification test to obtain a comprehensive noise detection result, and the comprehensive noise detection result is then subjected to a second-level screening. The comprehensive noise detection results after secondary screening are rotated in reverse to restore the original image coordinate system, thus obtaining the noise mask of the remote sensing image data to be detected.

2. The automatic stripe noise detection method as described in claim 1, characterized in that, The steps for obtaining the orientation correction map include: Perform directional gradient statistics on the preprocessed result image and calculate the gradient response intensity under multiple preset directional angles; The direction angle with the largest gradient response intensity is selected as the main fringe direction; Based on the angle of the main fringe direction, the preprocessed image is subjected to coordinate rotation transformation until the main fringe direction is aligned with the image column direction, thus obtaining the orientation correction image.

3. The automatic stripe noise detection method as described in claim 2, characterized in that, The steps for constructing co-occurrence image information maps corresponding to images at various scales include: Perform cloud and snow detection on each scale image in the multi-scale image to obtain a mask image of the cloud and snow region; Based on the mask image of the cloud and snow region, grayscale processing is performed on the image at each scale to increase the grayscale difference between adjacent image points in the column direction, thus obtaining a preprocessed difference map. The preprocessed difference image is subjected to secondary grayscale processing to further increase the grayscale difference between adjacent image points in the column direction, resulting in a secondary processed difference image. Extract texture features from the secondary processing difference map to obtain a texture feature map, calculate the gradient value of the secondary processing difference map, and obtain a gray-level gradient co-occurrence map. The gray-level gradient co-occurrence map is used to transform the secondary processing difference map, and the signal-to-noise ratio information map is calculated by combining it with the texture feature map. The secondary processing difference map is transformed using texture feature map, gray-level gradient co-occurrence map and signal-to-noise ratio information map to obtain the tertiary processing difference map; Land cover classification is performed on images at each scale to obtain water body masks; By using cloud and snow area mask images and water body mask images to process the difference map three times to remove interference areas, the corresponding symbiotic image information map at this scale is obtained.

4. The automatic stripe noise detection method as described in claim 3, characterized in that, The steps of performing image block processing on the symbiotic image information map include: The co-occurrence imagery at each scale is divided into M×N sub-regions; For each sub-region, the co-occurrence image infographic is divided into positive line graphs and negative line graphs based on the grayscale values; Limit the line length of both positive and negative line graphs, and retain lines whose length is greater than a preset threshold to obtain a processed line graph; The number of noise pixels and the number of non-interference area pixels in the first-processed line graph are counted, and then the concentration characterization parameter is used to limit the second-processed line graph. The secondary processing results of the positive and negative line graphs are combined to obtain the local noise detection results for this sub-region.

5. The automatic stripe noise detection method as described in claim 4, characterized in that, The steps for verifying the cross-regional continuity of stripes in the global region include: In the local noise detection results of each sub-region, count the number of sub-regions that each noise line segment crosses; Noise segments that span more than a preset continuity threshold are retained, while isolated noise segments that appear only in a single or a few sub-regions are removed. Preliminary noise detection results at various scales were obtained.

6. The automatic stripe noise detection method as described in claim 5, characterized in that, The steps to obtain the comprehensive noise detection results include: Spatial registration is performed on the preliminary noise detection results at each scale to map the detection results at different scales to a unified coordinate system; For the mapped detection results at each scale, count the number of scales that are marked as noise at each pixel location by column; The pixel locations that are simultaneously detected as noise at at least K scales are retained to obtain the comprehensive noise detection result.

7. The automatic stripe noise detection method as described in claim 6, characterized in that, The steps for secondary screening of comprehensive noise detection results include: Based on the preprocessed difference map, the statistical characteristics of the gray-scale difference between each noise line segment and its adjacent columns in the comprehensive noise detection results are calculated. When the mean square error of the grayscale difference is greater than the preset difference threshold, the noise line segment is retained; when the mean square error of the grayscale difference is less than or equal to the preset difference threshold, the noise line segment is removed. Count the number of non-interference region pixels and the number of pixels detected as noise in each column; Calculate the proportion of noise pixels to non-interference area pixels, as well as the start and end row numbers of the noise pixels; When the ratio and row number range meet the concentration condition, the noise detection results of that column are retained; otherwise, the noise detection results of that column are discarded. The comprehensive noise detection results after secondary screening were obtained.

8. An automatic stripe noise detection system based on single-band texture features, employing the method described in any one of claims 1-7, characterized in that, include: The calibration map acquisition module is used to calculate the main fringe direction of the preprocessed result map, and perform coordinate rotation transformation based on the main fringe direction to obtain the orientation calibration map. The co-occurrence image information map construction module is used to extract texture information and grayscale gradient information from multi-scale images and construct co-occurrence image information maps corresponding to each scale image. The preliminary noise detection module performs local stripe detection in each sub-region and verifies the cross-regional continuity of stripes in the global region to obtain preliminary noise detection results at each scale. The integrated noise detection module performs a first-level verification test on the preliminary noise detection results at each scale to obtain the integrated noise detection results, and then performs a second-level screening on the integrated noise detection results.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the automatic stripe noise detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the automatic stripe noise detection method according to any one of claims 1 to 7.

Images