Low-power acquisition and transmission system for industrial visual inspection

Abstract

The application relates to the technical field of visual detection, in particular to a low-power-consumption acquisition and transmission system for industrial visual detection, which comprises the following steps: clustering pixel points in a detection image, analyzing the similarity between different rows in the detection image to divide a plurality of target regions, dividing the target regions into a plurality of subregions, dividing the detection image into a plurality of block regions according to the distribution of gradient values and gray values of the pixel points in the subregions, obtaining a box plot of each block region and obtaining noise points in each block region according to the box plot, filtering all noise points in the detection image and transmitting the detection image. The application improves the adaptive capacity of corresponding data quantity when analyzing noise points in each block region through the box plot, thereby improving the accuracy of noise detection and filtering, and effectively ensuring the low-power-consumption degree when transmitting the detection image.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and more specifically to a low-power acquisition and transmission system for industrial visual inspection. Background Technology

[0002] Industrial machine vision inspection is an automated technology that uses cameras and computer algorithms to inspect and measure the quality of products in industrial production. It helps factories complete quality control in production, improve production efficiency, and reduce defect rates. However, the image acquisition and processing process requires a lot of power and computing power to support high-speed, high-precision image acquisition and processing. High-power devices can overheat during long-term operation, affecting the stability and reliability of the system. This can also cause noise in the image, affecting the efficiency of image compression and processing, and further increasing the power consumption of the device.

[0003] To reduce the impact of noise, box plots are often used to detect noise in images. However, the upper and lower limits of the box plots obtained by constructing all the pixels of the image into a box plot are not accurate enough, which makes it impossible to obtain accurate noise detection results. Furthermore, it makes it impossible to effectively guarantee low-power acquisition and transmission during industrial vision inspection. Summary of the Invention

[0004] This invention provides a low-power acquisition and transmission system for industrial visual inspection to solve existing problems.

[0005] The low-power acquisition and transmission system for industrial visual inspection of the present invention adopts the following technical solution:

[0006] One embodiment of the present invention provides a low-power acquisition and transmission system for industrial visual inspection, the system comprising the following modules:

[0007] The image acquisition module is used to acquire the detection image;

[0008] The image analysis module is used to cluster the pixels in each row of the detection image to obtain several clusters; it performs similarity analysis based on the distance between different rows, the gray value and gradient value of the pixels in different rows, and the clustering results to obtain the weighted similarity of each row; and it divides all rows in the detection image into several target regions based on the magnitude of the weighted similarity.

[0009] The noise detection module is used to divide the target area into several sub-regions with a preset number of columns. Based on the distribution of gradient values ​​and gray values ​​of pixels in the sub-regions, it obtains the degree of preference of the sub-regions under different number of columns. Based on the degree of preference, it divides the detection image into several block regions, obtains the box plot of each block region, and obtains the noise points in each block region based on the box plot.

[0010] The image processing and transmission module is used to filter all noise points in the detected image and then transmit it.

[0011] Optionally, the step of performing similarity analysis based on the distance between different rows, the grayscale values ​​and gradient values ​​of pixels within different rows, and the clustering results to obtain the weighted similarity of each row includes the following specific methods:

[0012] The gradient values ​​of all pixels in the detection image are obtained using the Sobel operator; the cluster center of each cluster in the detection image is obtained; the attribute parameters between any two rows in the detection image are obtained based on the gray values, gradient values, and clustering results of pixels in different rows in the detection image; the row similarity between any two rows in the detection image is obtained based on the attribute parameters, and the attribute parameters are inversely proportional to the row similarity.

[0013] The weighted similarity of any row in the detected image is obtained by weighting the row similarity between any two rows using the distance between them.

[0014] Optionally, the attribute parameters include: grayscale parameters, gradient parameters, cluster distance parameters, and cluster size parameters;

[0015] Wherein, the grayscale parameter is the first grayscale value in the detected image. Line and number The absolute value of the difference between the average gray values ​​of all pixels in a row; the gradient parameter represents the value of the first pixel in the detected image. Line and number The absolute value of the difference between the average gradient values ​​of all pixels in a row; the specific method for obtaining the clustering distance parameter is: obtain the th in the detection image Line and number The Euclidean distance between the cluster centers of any clusters with the same ordinal number within a row is denoted as the cluster distance factor for clusters with the same ordinal number. Line and number The mean of the clustering distance factors for all clusters with the same ordinal number within a row, denoted as the i-th Line and number The clustering distance parameter of the row; the specific method for obtaining the cluster size parameter is: obtain the clustering distance parameter of the row in the detection image. Line and number The absolute value of the difference in the number of pixels in any cluster with the same ordinal number within a row is denoted as the cluster size factor for clusters with the same ordinal number. This factor is used to measure the number of pixels in the detected image. Line and number The mean of the cluster size factors of all clusters with the same ordinal number in a row, denoted as the i-th Line and number The cluster size parameter for rows.

[0016] Optionally, the method for weighting the row similarity of any two rows in the detection image by using the distance between them to obtain the weighted similarity of any row in the detection image includes the following specific methods:

[0017] The process iterates through each row in the detected image from top to bottom, constructing an empty set denoted as the row set. During the traversal, any row in the detected image is designated as the target row and placed into the row set. Rows in the row set are designated as determined rows, and rows in the detected image that do not belong to the row set are designated as undetermined rows. Based on the distance and row similarity between the undetermined row and all determined rows in the row set, the weighted similarity of any undetermined row is obtained.

[0018] Optionally, the specific method for calculating the weighted similarity of the rows to be determined is as follows:

[0019] By detecting the first in the image The undetermined rows and the row set of the first row. The distance of the determined row in the detected image. The undetermined rows and the row set of the first row. We weight the row similarity of the identified rows to obtain the first row in the detected image. The weighted similarity of a row to be determined, wherein the distance is inversely proportional to the weighted similarity, and the row similarity is directly proportional to the weighted similarity.

[0020] Optionally, the specific method for dividing all rows in the detected image into several target regions based on the weighted similarity is as follows:

[0021] When the weighted similarity of the row to be determined is greater than or equal to the preset similarity threshold, the row to be determined is added to the row set as a determined row; when the weighted similarity of the row to be determined is less than the preset similarity threshold, the row to be determined is not merged into the row set, and the iteration stops.

[0022] The region formed by all rows in the row set is denoted as the target region.

[0023] Optionally, the specific method for dividing the target region into several sub-regions with a preset number of columns, and obtaining the optimality of the sub-regions under different numbers of columns based on the distribution of gradient values ​​and grayscale values ​​of pixels in the sub-regions, includes:

[0024] Traverse each target region from left to right, and during the traversal, divide any target region into several regions, which are denoted as sub-regions of the target region;

[0025] When any sub-region contains the number of columns When retrieving the number of columns in the sub-region, the number of columns is... The degree of preference at that time is calculated as follows:

[0026]

[0027] in , and These are the preset first hyperparameter and second hyperparameter, respectively. The number of columns representing the sub-region is The degree of preference at that time; The number of columns representing the sub-region is At that time, the first sub-region One gradient-grayscale factor; This indicates the number of gradient-grayscale factors in the sub-region; This represents the linear normalization function.

[0028] Optionally, the gradient-grayscale factor includes: the mean gradient value of all pixels in the sub-region, the variance of the grayscale values ​​of all pixels in the sub-region, and the absolute value of the skewness of the grayscale values ​​of all pixels in the sub-region.

[0029] Optionally, the specific method for dividing the detection image into several block regions according to the degree of preference, obtaining a box plot for each block region, and obtaining noise points in each block region based on the box plot is as follows:

[0030] Get the number of columns in the sub-region within the range Within the image, the sub-region corresponding to the maximum optimization degree under all column numbers is denoted as a block region. Several block regions are obtained in the detection image. The box plot of the gray values ​​of all pixels in each block region is obtained. The upper and lower limits of the box plot corresponding to each block region are obtained. Pixels in the block region that exceed the upper and lower limits of the box plot are regarded as noise points.

[0031] Optionally, the specific method for filtering and transmitting all noise points in the detected image includes:

[0032] The mean filtering algorithm is used to filter all noise points in the detection image to obtain the filtered image; the filtered image is then processed by run-length encoding to obtain the compressed detection image; the compressed detection image is transmitted via wireless network or wired connection.

[0033] The beneficial effects of the technical solution of the present invention are as follows: by clustering the pixels in the detection image, different clusters can be identified, and similarity analysis can be performed using the distance between different rows, the gray value of the pixels, the gradient value, and the clustering results to accurately divide the target region. Furthermore, the degree of optimization is determined by the distribution of the gradient value and gray value of the pixels in the sub-regions, and the target region is divided into several block regions. This improves the adaptive capability of the corresponding data volume when analyzing the noise points in each block region through box plots, thereby improving the accuracy of noise detection and filtering, and effectively ensuring low power consumption when transmitting the detection image. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. 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.

[0035] Figure 1 This is a structural block diagram of a low-power acquisition and transmission system for industrial vision inspection according to the present invention;

[0036] Figure 2 The flowchart is provided by the present invention for a low-power acquisition and transmission system for industrial visual inspection. Detailed Implementation

[0037] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a low-power acquisition and transmission system for industrial visual inspection proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0039] The following description, in conjunction with the accompanying drawings, details a specific solution for a low-power acquisition and transmission system for industrial visual inspection provided by this invention.

[0040] Please see Figure 1 The diagram illustrates a structural block diagram of a low-power acquisition and transmission system for industrial visual inspection according to an embodiment of the present invention. The system includes the following modules:

[0041] Image acquisition module 101 is used to acquire detection images.

[0042] It should be noted that industrial vision inspection is a process of inspecting and controlling the quality of industrial products using computer vision and image processing technologies. During the vision inspection process in industrial production, a large amount of image data is generated. In the industrial vision inspection process, the images captured by industrial cameras are usually transmitted to computers or servers for image processing and analysis. However, when there is noise in the image, it increases the complexity of the image data, resulting in the need for more bandwidth and resources for image transmission. Therefore, in order to reduce the power consumption of image acquisition and transmission during industrial vision inspection, this embodiment proposes a low-power acquisition and transmission system for industrial vision inspection.

[0043] To implement the low-power acquisition and transmission system for industrial vision inspection proposed in this embodiment, it is first necessary to acquire the inspection images during the industrial vision inspection process. The specific process is as follows:

[0044] In the industrial production process, images are acquired by CCD (Charge Coupled Device) industrial cameras during industrial vision inspection and recorded as the original inspection image. The image after grayscale processing of the original inspection image is also recorded as the inspection image.

[0045] Thus, the detection image is obtained through the above method.

[0046] The image analysis module 102 is used to cluster the pixels in each row of the detection image, perform similarity analysis based on the distance between different rows and the gray values, gradient values ​​and clustering results of the pixels in different rows, obtain the weighted similarity of each row, and divide all rows in the detection image into several target regions according to the magnitude of the weighted similarity.

[0047] It should be noted that the increase in power consumption of image data transmission due to noise in image data is specifically manifested in the following ways: noisy pixels in the image increase the detail and complexity of the image, requiring more information to describe the additional details during encoding and transmission. Furthermore, since existing image compression algorithms rely on the correlation between image regions to predict and remove redundancy, noise disrupts this correlation, making the encoding and transmission of inspection images using existing image compression algorithms less effective. This reduces the efficiency of encoding and transmitting acquired images during industrial vision inspection and increases the power consumption of image processing.

[0048] Therefore, this embodiment uses box plots to analyze the detection image and identify noise points in the detection image, thereby further achieving the effect of image denoising. However, since the gray values ​​of noise points in the detection image are different, when using box plots on the entire detection image, the upper and lower limits of the box plots will be affected by noise points with different gray values, making it impossible to effectively and accurately obtain noise points in the detection image. In order to better detect noise pixels in the image, this embodiment constructs multiple box plots for different ranges of the detection image, that is, each box plot corresponds to a part of the detection image. By analyzing the box plots corresponding to each part of the detection image, the noise points in the detection image are determined.

[0049] Specifically, in step 2.1, the pixels in each row of the detection image are clustered. The specific process is as follows: the K value of the K-means clustering algorithm is preset, and all pixels in each row of the detection image are clustered according to the gray value and coordinates of the pixels, so as to obtain K clusters in each row of the detection image and the cluster center of each cluster.

[0050] It should be noted that the K value of the K-means clustering algorithm is preset to 5 based on experience, and can be adjusted according to the actual situation. This embodiment does not impose a specific limitation.

[0051] Step 2.2: Perform similarity analysis based on the distance between different rows, the grayscale values ​​and gradient values ​​of pixels within different rows, and the clustering results to obtain the weighted similarity of each row. As an example, the specific method for obtaining this similarity is as follows:

[0052] First, the gradient values ​​of all pixels in the detection image are obtained using the Sobel operator; then, the row similarity between any two rows in the detection image is obtained. As an example, the specific calculation method is as follows:

[0053]

[0054] in, Indicates the first element in the detected image Line and number Row similarity; Indicates the first element in the detected image Line and number The first line One attribute parameter; Indicates the first element in the detected image Line and number The number of attribute parameters for a row; This represents the linear normalization function.

[0055] The attribute parameters include: grayscale parameter, gradient parameter, cluster distance parameter, and cluster size parameter.

[0056] Wherein, the grayscale parameter represents the number of pixels in the detected image. Line and number The absolute value of the difference in average grayscale values ​​among all pixels in a row.

[0057] The gradient parameter represents the first gradient in the detected image. Line and number The absolute value of the difference in average gradient values ​​among all pixels in a row.

[0058] The specific method for obtaining the clustering distance parameter is as follows: obtain the first clustering distance parameter in the detected image. Line and number The Euclidean distance between the cluster centers of any clusters with the same ordinal number within a row is denoted as the cluster distance factor for clusters with the same ordinal number. Line and number The mean of the clustering distance factors for all clusters with the same ordinal number within a row, denoted as the i-th Line and number The clustering distance parameter for rows.

[0059] The specific method for obtaining the cluster size parameter is as follows: obtain the cluster size parameter in the detection image. Line and number The absolute value of the difference in the number of pixels in any cluster with the same ordinal number within a row is denoted as the cluster size factor for clusters with the same ordinal number. This factor is used to measure the number of pixels in the detected image. Line and number The mean of the cluster size factors of all clusters with the same ordinal number in a row, denoted as the i-th Line and number The cluster size parameter for rows.

[0060] It should be noted that the Sobel operator is an existing operator for obtaining the gradient value of a pixel, so it will not be described in detail in this embodiment.

[0061] It should be noted that the row similarity is used to describe the degree of similarity of image features between different rows in the detection image. Specifically, it reflects the similarity between any two rows in the detection image in terms of gray level, gradient, and clustering results. The smaller the gray level parameter and gradient parameter, the greater the overall similarity between the two rows in terms of gray level and gradient values, and the greater the row similarity. Since all the pixels contained in each row of the detection image are clustered to obtain the clustering result, which reflects the distribution of the gray level of the pixels in the corresponding row in the spatial domain, the smaller the clustering distance parameter and cluster size parameter, which reflect the similarity of the clustering results between different rows, the more similar the distribution of the gray level of the pixels in the spatial domain in the corresponding two rows, and the greater the row similarity.

[0062] Then, the row similarity of any two rows in the detected image is weighted by the distance between them to obtain the weighted similarity of any row in the detected image. As an example, the specific method for obtaining this weighted similarity is as follows:

[0063] The process iterates through each row of the detected image from top to bottom, constructing an empty set denoted as the row set. During the traversal, any row in the detected image is designated as the target row and added to the row set. Rows in the row set are designated as determined rows, and rows in the detected image that do not belong to the row set are designated as undetermined rows. Based on the distance and row similarity between the undetermined row and all determined rows in the row set, the weighted similarity of any undetermined row is obtained. The specific calculation method for the weighted similarity of the undetermined row is as follows:

[0064]

[0065] in, Indicates the first element in the detected image Weighted similarity of the rows to be determined; Indicates the first element in the detected image The undetermined rows and the row set of the first row. The distance of a determined row; Indicates the first element in the detected image The undetermined rows and the row set of the first row. Row similarity of a given row.

[0066] Step 2.3: Divide all rows in the detected image into several target regions according to the weighted similarity. The specific steps are as follows: First, when the weighted similarity of the row to be determined is greater than or equal to the preset similarity threshold, the row to be determined is placed into the row set as a determined row; when the weighted similarity is less than the preset similarity threshold, the row to be determined is not merged into the row set, and the iteration stops.

[0067] Then, the region formed by all rows in the row set is denoted as the target region.

[0068] Repeat steps 2.2 and 2.3 to obtain several target regions in the detected image.

[0069] It should be noted that, based on experience, this embodiment presets a similarity threshold of 0.6. The specific value of the similarity threshold can be adjusted according to the actual situation. This embodiment does not impose a specific limitation. In this embodiment, it is considered that when the weighted similarity of the row to be determined is greater than or equal to 0.6, the weighted similarity between the row to be determined and all the determined rows in the corresponding row set is high enough. That is, the row to be determined and all the determined rows in the row set are sufficiently similar to form a corresponding region. Therefore, the row to be determined is divided into a set and the iteration continues downward. Conversely, when the weighted similarity is less than the preset similarity threshold, it is considered that the weighted similarity between the row to be determined and all the determined rows in the row set is low. Therefore, it cannot be divided into the same row set to form a corresponding region. At the same time, the iteration stops and the target region formed by the number of rows contained in the row set is determined.

[0070] It should be noted that, since determining the number of rows in the target region is an iterative process, in order to ensure that the target region is accurately obtained, this embodiment obtains the weighted similarity between the row to be determined and all the determined rows in the row set, and then determines whether to continue adding rows in the row set to obtain the target region.

[0071] It should be noted that, since there are quantity requirements for the data to be calculated and analyzed when obtaining box plots, that is, the data in a box plot cannot be too few, otherwise there will be a large detection error. Therefore, in order to avoid the above problem, in this embodiment, when the number of rows contained in the target region is less than A, even if the weighted similarity between the row to be determined and all the determined rows in the row set is less than the preset similarity threshold during the iteration process, it is still placed into the row set and forms the same target region with the determined rows in the row set. Here, A is a preset quantity parameter. Based on experience, A is preset to 8, which can be adjusted according to the actual situation. This embodiment does not specifically limit it.

[0072] It should be noted that, in this embodiment, when determining the number of rows in the target region, in order to avoid statistical errors caused by excessive differences and dispersion of pixels within the same target region, this embodiment divides rows with high similarity and adjacent rows into the same region. During the division process, the row similarity between different rows is obtained, and the weighted similarity of each row is obtained by combining the distance between different rows. This allows rows with similar pixel gradient values ​​and grayscale value distributions to be grouped together to form a region, namely the target region.

[0073] Thus, several target areas have been obtained through the above methods.

[0074] The noise detection module 103 is used to divide the target area into several sub-regions with a preset number of columns, obtain the degree of preference of the sub-regions under different number of columns based on the distribution of gradient values ​​and gray values ​​of pixels in the sub-regions, divide the detection image into several block regions according to the degree of preference, obtain the box plot of each block region, and obtain the noise points in each block region based on the box plot.

[0075] Specifically, first, each target region is traversed from left to right. During the traversal, any target region is divided into several regions, denoted as sub-regions of the target region. When any sub-region contains a certain number of columns... When retrieving the number of columns in the sub-region, the number of columns is... The optimality of the time, as an example, is specifically calculated as follows:

[0076]

[0077] in , and These are the preset first hyperparameter and second hyperparameter, respectively. The number of columns representing the sub-region is The degree of preference at that time; The number of columns representing the sub-region is At that time, the first sub-region One gradient-grayscale factor; This indicates the number of gradient-grayscale factors in the sub-region; This represents the linear normalization function.

[0078] The gradient-grayscale factor includes: the mean gradient value of all pixels in the sub-region, the variance of the grayscale value of all pixels in the sub-region, and the absolute value of the skewness of the grayscale value of all pixels in the sub-region.

[0079] It should be noted that, according to the statistical principles of box plots for anomaly detection, the effectiveness of box plots in detecting anomalies is easily affected by the sample size. If the sample size is too small, the quartiles and outliers displayed by the box plot may be meaningless. If the sample size is too large, the box plot may fail to effectively display the data distribution and outliers, thus affecting the detection effectiveness. Furthermore, the distribution of the sample data also affects the effectiveness of box plots in detecting anomalies; the closer the distribution of the sample data is to a normal distribution, the better the detection effect of the box plot on anomalies. Therefore, in the detection image, to improve the accuracy of box plots in detecting noise points in sub-regions, this embodiment obtains the distribution of all pixels in the sub-region in terms of gradient and grayscale values, and obtains the degree of preference of the sub-region at a corresponding number of columns, using the number of columns describing the sub-region as... At that time, the accuracy of noise points in a sub-region can be detected by box plots.

[0080] It should be noted that in this embodiment, the gradient-grayscale factor is used to describe the distribution of gradient values ​​and grayscale values ​​of all pixels in the sub-region. The mean gradient values ​​and variance of grayscale values ​​of all pixels in the sub-region reflect the stability of the distribution of grayscale values ​​in the sub-region. The smaller the mean gradient values ​​and variance of grayscale values, the more stable the distribution of grayscale values ​​in the sub-region. The absolute value of the skewness of the grayscale values ​​of all pixels in the sub-region reflects how closely the distribution curve formed by the grayscale values ​​of all pixels in the sub-region resembles a normal distribution. The smaller the absolute value of the skewness, the closer the distribution curve formed by the grayscale values ​​of all pixels in the sub-region resembles a normal distribution, and the smaller the number of columns in the sub-region. The higher the degree of preference, the better.

[0081] It should be noted that, in order to avoid the algorithm being too complex and to take into account the impact of the amount of data on the box plot, the first hyperparameter and the second hyperparameter are preset to 10 and 30 respectively in this embodiment. They can be adjusted according to the actual situation, and this embodiment does not impose any specific limitations.

[0082] Then, obtain the column number of the sub-region within the range. Within the image, the sub-region corresponding to the maximum optimization degree under all column numbers is denoted as a block region. Several block regions are obtained in the detection image. The box plot of the gray values ​​of all pixels in each block region is obtained. The noise points in the corresponding block region are obtained based on the box plot of each block region.

[0083] The specific method for obtaining noise points in the corresponding block region based on the box plot of each block region is as follows: obtain the upper and lower limits of the box plot corresponding to each block region, and take the pixels in the block region that exceed the upper and lower limits of the box plot as noise points.

[0084] Thus, all noise points in the detected image are obtained using the above method.

[0085] The image processing and transmission module 104 is used to filter and transmit all noise points in the detected image.

[0086] Specifically, a filtering algorithm is used to filter all noise points in the detection image to obtain a filtered result image. A compression algorithm is then used to encode and compress the filtered result image to obtain a compressed detection image. The compressed detection image is then transmitted via a wireless network or a wired connection.

[0087] The method for filtering all noise points in the detected image using a filtering algorithm to obtain the filtered result image includes:

[0088] Optionally, the mean filtering algorithm is used to filter all noise points in the detection image to obtain the filtered result image;

[0089] Optionally, median filtering is performed on all noise points in the detected image to obtain the filtered result image;

[0090] Optionally, Gaussian filtering can be applied to all noise points in the detected image to obtain the filtered result image.

[0091] The method for encoding and compressing the filtered image using a compression algorithm to obtain a compressed detection image includes:

[0092] Optionally, the filtered image is subjected to run-length encoding to obtain a compressed detection image;

[0093] Optionally, the filtered image can be encoded and compressed using a Huffman coding algorithm to obtain a compressed detection image.

[0094] It should be noted that the above filtering method is merely an example to clearly illustrate the acquisition of the filtered image, and is not a limitation on the specific implementation. For those skilled in the art, other variations or modifications can be made based on the above description, and it is impossible to exhaustively list all implementation methods here.

[0095] Please see Figure 2 The document presents a flowchart of a low-power acquisition and transmission system for industrial vision inspection provided in one embodiment of the present invention.

[0096] This concludes the embodiment.

[0097] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A low-power acquisition and transmission system for industrial visual inspection, characterized in that, The system includes the following modules: The image acquisition module is used to acquire the detection image; The image analysis module is used to cluster the pixels in each row of the detection image to obtain several clusters; it performs similarity analysis based on the distance between different rows, the gray value and gradient value of the pixels in different rows, and the clustering results to obtain the weighted similarity of each row; and it divides all rows in the detection image into several target regions based on the magnitude of the weighted similarity. The noise detection module is used to divide the target area into several sub-regions with a preset number of columns. Based on the distribution of gradient values ​​and gray values ​​of pixels in the sub-regions, it obtains the degree of preference of the sub-regions under different number of columns. Based on the degree of preference, it divides the detection image into several block regions, obtains the box plot of each block region, and obtains the noise points in each block region based on the box plot. The image processing and transmission module is used to filter all noise points in the detected image and then transmit it. The specific method for dividing the target region into several sub-regions with a preset number of columns, and obtaining the optimality of the sub-regions under different numbers of columns based on the distribution of gradient values ​​and grayscale values ​​of pixels in the sub-regions, includes: Traverse each target region from left to right, and during the traversal, divide any target region into several regions, which are denoted as sub-regions of the target region; When any sub-region contains the number of columns When retrieving the number of columns in the sub-region, the number of columns is... The degree of preference at that time is calculated as follows: ; in , and These are the preset first hyperparameter and second hyperparameter, respectively. The number of columns representing the sub-region is The degree of preference at that time; The number of columns representing the sub-region is At that time, the first sub-region One gradient-grayscale factor; This indicates the number of gradient-grayscale factors in the sub-region; This represents the linear normalization function.

2. The low-power acquisition and transmission system for industrial visual inspection according to claim 1, characterized in that, The method for performing similarity analysis based on the distance between different rows, the grayscale values ​​and gradient values ​​of pixels within different rows, and clustering results to obtain the weighted similarity of each row includes the following specific methods: The Sobel operator is used to obtain the gradient values ​​of all pixels in the detection image; the cluster center of each cluster in the detection image is obtained; and the attribute parameters between any two rows in the detection image are obtained based on the gray values, gradient values ​​and clustering results of pixels in different rows in the detection image. The row similarity between any two rows in the detected image is obtained based on the attribute parameters, wherein the attribute parameters are inversely proportional to the row similarity. The weighted similarity of any row in the detected image is obtained by weighting the row similarity between any two rows using the distance between them.

3. The low-power acquisition and transmission system for industrial visual inspection according to claim 2, characterized in that, The attribute parameters include: grayscale parameter, gradient parameter, cluster distance parameter, and cluster size parameter; Wherein, the grayscale parameter is the first grayscale value in the detected image. Line and number The absolute value of the difference between the average gray values ​​of all pixels in a row; the gradient parameter represents the value of the first pixel in the detected image. Line and number The absolute value of the difference between the average gradient values ​​of all pixels in a row; the specific method for obtaining the clustering distance parameter is: obtain the th in the detection image Line and number The Euclidean distance between the cluster centers of any clusters with the same ordinal number within a row is denoted as the cluster distance factor for clusters with the same ordinal number. Line and number The mean of the clustering distance factors for all clusters with the same ordinal number within a row, denoted as the i-th Line and number The clustering distance parameter of the row; the specific method for obtaining the cluster size parameter is: obtain the clustering distance parameter of the row in the detection image. Line and number The absolute value of the difference in the number of pixels in any cluster with the same ordinal number within a row is denoted as the cluster size factor for clusters with the same ordinal number. This factor is used to measure the number of pixels in the detected image. Line and number The mean of the cluster size factors of all clusters with the same ordinal number in a row, denoted as the i-th Line and number The cluster size parameter for rows.

4. The low-power acquisition and transmission system for industrial visual inspection according to claim 2, characterized in that, The method for weighting the row similarity of any two rows in the detection image by using the distance between them to obtain the weighted similarity of any row in the detection image includes the following specific methods: The process iterates through each row in the detected image from top to bottom, constructing an empty set denoted as the row set. During the traversal, any row in the detected image is designated as the target row and placed into the row set. Rows in the row set are designated as determined rows, and rows in the detected image that do not belong to the row set are designated as undetermined rows. Based on the distance and row similarity between the undetermined row and all determined rows in the row set, the weighted similarity of any undetermined row is obtained.

5. The low-power acquisition and transmission system for industrial visual inspection according to claim 4, characterized in that, The specific method for calculating the weighted similarity of the rows to be determined is as follows: By detecting the first in the image The undetermined rows and the row set of the first row. The distance of the determined row in the detected image. The undetermined rows and the row set of the first row. We weight the row similarity of the identified rows to obtain the first row in the detected image. The weighted similarity of a row to be determined, wherein the distance is inversely proportional to the weighted similarity, and the row similarity is directly proportional to the weighted similarity.

6. The low-power acquisition and transmission system for industrial visual inspection according to claim 4, characterized in that, The specific method for dividing all rows in the detected image into several target regions based on the weighted similarity is as follows: When the weighted similarity of the row to be determined is greater than or equal to the preset similarity threshold, the row to be determined is added to the row set as a determined row; when the weighted similarity of the row to be determined is less than the preset similarity threshold, the row to be determined is not merged into the row set, and the iteration stops. The region formed by all rows in the row set is denoted as the target region.

7. The low-power acquisition and transmission system for industrial visual inspection according to claim 1, characterized in that, The gradient-grayscale factor includes: the mean gradient value of all pixels in the sub-region, the variance of the grayscale value of all pixels in the sub-region, and the absolute value of the skewness of the grayscale value of all pixels in the sub-region.

8. The low-power acquisition and transmission system for industrial vision inspection according to claim 1, characterized in that, The specific method for dividing the detection image into several block regions according to the degree of preference, obtaining the box plot of each block region, and obtaining the noise points in each block region based on the box plot is as follows: Get the number of columns in the sub-region within the range Within the image, the sub-region corresponding to the maximum optimization degree under all column numbers is denoted as a block region. Several block regions are obtained in the detection image. The box plot of the gray values ​​of all pixels in each block region is obtained. The upper and lower limits of the box plot corresponding to each block region are obtained. Pixels in the block region that exceed the upper and lower limits of the box plot are regarded as noise points.

9. The low-power acquisition and transmission system for industrial visual inspection according to claim 1, characterized in that, The specific method for filtering and transmitting all noise points in the detected image is as follows: The mean filtering algorithm is used to filter all noise points in the detection image to obtain the filtered image; the filtered image is then processed by run-length encoding to obtain the compressed detection image; the compressed detection image is transmitted via wireless network or wired connection.

Images