Industrial image real-time analysis system and method

Abstract

The present invention relates to the field of image analysis technology, specifically to a real-time analysis system and method for industrial images, wherein the system includes an image capture and difference calculation module, a spectrum analysis and feature extraction, a defect recognition and classification module, and a real-time monitoring and alarm module. In the present invention, by capturing industrial image sequences in real time and performing pixel-level comparison, the efficiency and response speed of production line monitoring are effectively improved, the pixel difference value is calculated, and the detection accuracy is greatly improved, which is particularly suitable for rapid changes in dynamic environments. By performing in-depth analysis of image data through spectral analysis, it is possible to identify tiny changes in material composition, classify defect types, and compare them with defect databases, greatly enhancing the accuracy and reliability of identification. Real-time monitoring and continuous data comparison ensure that the system can capture and respond to defect evolution trends in a timely manner, thereby providing early warnings, enhancing the ability of equipment maintenance and safety monitoring, and effectively optimizing production processes and improving industrial safety.

Description

Technical Field

[0001] The present invention relates to the technical field of image analysis, and in particular to a real-time industrial image analysis system and method. Background Art

[0002] The field of image analysis technology involves the use of algorithms and statistical methods to analyze data obtained from images, which can automatically identify the content of the image and extract valuable information and features. In industrial applications, image analysis technology is often used in quality control, automated inspection, machine vision, and real-time monitoring. This field combines multiple technologies such as image processing, pattern recognition, artificial intelligence, and machine learning to improve processing speed and analysis accuracy, and achieve instant response and decision support for dynamic scenes.

[0003] Among them, the industrial image real-time analysis system is a system that can instantly analyze and process image data collected in an industrial environment. The main uses of the system include but are not limited to monitoring production lines, automated defect detection, equipment maintenance warnings, and safety monitoring. Through real-time analysis, the system can quickly identify problems, optimize production processes, and improve industrial safety, thereby reducing production costs and improving efficiency.

[0004] Although existing technologies integrate methods such as image processing and pattern recognition, they are still limited in processing speed and real-time response capabilities, especially when processing large-scale data sets. They are slow to respond and cannot meet the needs of high-speed production environments. Existing technologies are also significantly insufficient in accuracy and adaptability to complex environments, especially when identifying and classifying small or complex defects, resulting in poor results in immediate response and decision support in dynamic scenarios. This technical limitation not only affects production efficiency, but may also increase safety risks. For example, in the process of automated quality control, key defects may be overlooked due to insufficient recognition functions, which in turn affects product quality and safety. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose an industrial image real-time analysis system and method.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: The industrial image real-time analysis system includes:

[0007] The image capture and difference calculation module continuously captures industrial image sequences through the camera, performs pixel-level comparison on every two frames of images, calculates pixel difference values, identifies significant change areas, and outputs surface difference data;

[0008] The spectrum analysis and feature extraction module uses a spectrum scanner to analyze the key frames in the industrial image sequence, calculates the reflectance spectrum response of each pixel point, and extracts the spectrum features of each area in the image based on the calculation results combined with the surface difference data to obtain the spectrum feature information;

[0009] The defect recognition and classification module compares defect features according to the spectral feature information, scores each feature, compares the scoring result with the known defect database, determines the surface state, identifies the defect type and determines the defect area, and generates a defect type analysis record;

[0010] Based on the defect type analysis record, the real-time monitoring and alarm module continuously monitors the defect area, performs continuous data comparison, captures the image changes of the defect area in real time, identifies the defect evolution trend, and issues alarm information in real time to obtain industrial image dynamic monitoring alarm records.

[0011] As a further solution of the present invention, the step of acquiring the surface difference data is:

[0012] Capture continuous industrial images through the camera, select two frames of images, and use the formula:

[0013] , calculate the pixel-by-pixel difference value D of the two frames of images and obtain preliminary difference data, where, Represents the first frame image, Represents the second frame image, and C are the adjustment coefficient and correction constant used to adjust the accurate control of the difference sensitivity;

[0014] Based on the preliminary difference data, the formula is used,

[0015] ,

[0016] and

[0017] ,

[0018] Calculate the average difference and standard deviation , get the average difference and standard deviation data, where m is the total number of pixels, and It is a weight parameter adjusted according to actual measurement conditions to increase the flexibility and accuracy of calculation;

[0019] Based on the mean difference and standard deviation data, the formula is used,

[0020] ,

[0021] Calculate the screening threshold T, retain pixels with difference values ​​greater than T, identify areas of significant change, and generate surface difference data, where k is a coefficient selected based on image characteristics and is used to identify substantial changes.

[0022] As a further solution of the present invention, the calculation steps of the reflectance spectrum response of each pixel point are:

[0023] Use spectral scanners to capture raw light intensity data for each pixel in industrial image sequences ,in, Represents the light intensity of pixel (i, j) at wavelength x, generating raw light intensity data;

[0024] Based on the raw light intensity data, wavelength integration is performed using the formula,

[0025] ,

[0026] Calculate the total spectral response value of pixel (i, j) in the target wavelength range , generates a spectral response matrix, where represents the wavelength interval, and They are the minimum and maximum wavelengths, respectively, used to determine the wavelength range;

[0027] The spectral response matrix is ​​integrated and statistically analyzed using the formula:

[0028] ,

[0029] and

[0030] , calculate the spectral average and mutation value , extract key features and generate a spectral response overview, where K is the number of wavelength samples, indicating the number of samples in the entire wavelength range.

[0031] As a further solution of the present invention, the step of acquiring the spectral characteristic information is:

[0032] The total spectral response value is fused with the surface difference data, using the formula,

[0033] ,

[0034] Calculate data fusion value , and the fused spectral data is obtained, where g is the weight coefficient, which is used to adjust the influence of surface differences. is the surface difference value;

[0035] The fused spectral data is deeply analyzed using the formula:

[0036] ,

[0037] Calculate spectral eigenvalues , extract the significant spectral features of each region and generate spectral feature data;

[0038] Statistical analysis is performed based on the spectral feature data to evaluate the contribution and importance of the features in image analysis and integrate to generate spectral feature information.

[0039] As a further solution of the present invention, the specific steps of scoring each feature are:

[0040] Based on the spectral feature information, feature vectors are extracted, and feature vector data sets are obtained through data conversion and normalization. ,in, represents the feature vector of the i-th detection area, Indicates the characteristic intensity of the difference spectrum interval;

[0041] Based on the feature vector data set, for each feature in the feature vector To score, use the formula,

[0042] ,

[0043] Calculate the score of the nth feature in the i-th region , we get the feature scoring dataset, where The weight factor is determined based on historical data analysis, where N represents the number of features;

[0044] Based on the feature score dataset, the feature scores of each region are summarized using the formula,

[0045] ,

[0046] Calculate the total score of region i , output the score summary data.

[0047] As a further solution of the present invention, the step of obtaining the defect type result is:

[0048] Based on the score summary data, it is compared with the standard score in the defect database, using the formula,

[0049] ,

[0050] Calculate the similarity quantification between the total score and the database defect standard , get the matching score data, where is the standard score for each defect in the database;

[0051] Based on the matching score data, the defect type of each area is determined using the formula,

[0052] ,

[0053] The defect type confirmation result Y is obtained, where: is the reliability weight corresponding to defect q, which is used to adjust the trust. is a small constant used for mathematical stability, Q represents the defect type;

[0054] Based on the defect type confirmation result, the identified defect type is spatially located, the defect area is marked, and a defect type analysis record is generated.

[0055] As a further solution of the present invention, the steps for acquiring the industrial image dynamic monitoring record are:

[0056] According to the defect type analysis record, configure the monitoring system parameters, using the formula,

[0057] ,

[0058] Calculate the configuration sensitivity of the monitoring system to the i-th region , get the monitoring system configuration sensitivity data, where, represents the defect type score of the ith region, a and b are parameters for adjusting sensitivity;

[0059] Based on the sensitivity data of the monitoring system configuration, real-time data comparison is performed to identify key changes and adopt the formula,

[0060] ,

[0061] Calculate the difference in images captured during real-time monitoring , determine the defect evolution trend and obtain real-time image change analysis data, where P represents the threshold parameter, which is used to control the error tolerance. and Represents real-time image data at two consecutive moments;

[0062] According to the real-time image change analysis data, it is compared with the preset alarm threshold W. If it is greater than W, an alarm is triggered, otherwise no alarm is triggered, and the industrial image dynamic monitoring alarm record is obtained.

[0063] The industrial image real-time analysis method includes the following steps:

[0064] S1: Capture continuous industrial images through the camera, select two frames of images, analyze the pixel-by-pixel difference between the two frames of images, calculate the average difference value and standard deviation, determine the screening threshold and perform pixel point screening, identify the significant change area, and generate surface difference data;

[0065] S2: Use a spectral scanner to capture the raw light intensity data of each pixel in the industrial image sequence, perform wavelength integration, calculate the total spectral response value of the pixel in the target wavelength range, perform statistical analysis, extract key features, and generate a spectral response overview;

[0066] S3: fusing the spectral response overview with the surface difference data, and performing in-depth analysis, extracting significant spectral features of each region by calculating spectral feature values, evaluating the contribution and importance of the features in image analysis, and generating spectral feature information;

[0067] S4: extracting a feature vector based on the spectral feature information, scoring each feature in the feature vector through data conversion and normalization, and summarizing the scores to output score summary data;

[0068] S5: Based on the score summary data, compare with the standard score in the defect database, calculate the similarity between the total score and the database defect standard, determine the defect type of each area, spatially locate the defect type, synchronously mark the defect area, and generate a defect type analysis record;

[0069] S6: According to the defect type analysis record, perform real-time monitoring, conduct real-time data comparison, identify key changes, calculate the difference in images captured during real-time monitoring, determine the defect evolution trend, compare with the preset alarm threshold, determine the alarm information, and obtain the industrial image dynamic monitoring alarm record.

[0070] Compared with the prior art, the advantages and positive effects of the present invention are:

[0071] In the present invention, by real-time capture of industrial image sequences and pixel-level comparison, the efficiency and response speed of production line monitoring are effectively improved, the pixel difference value is calculated, and the detection accuracy is greatly improved. It is especially suitable for rapid changes in dynamic environments. Through spectral analysis, the image data is deeply analyzed, and slight changes in material composition can be identified. At the same time, the defect type is classified and compared with the defect database, which greatly enhances the accuracy and reliability of recognition. Real-time monitoring and continuous data comparison ensure that the system can capture and respond to defect evolution trends in a timely manner, thereby providing early warning, enhancing equipment maintenance and safety monitoring capabilities, and effectively optimizing production processes and improving industrial safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 is a system flow chart of the present invention;

[0073] Figure 2 is a flow chart for obtaining surface difference data of the present invention;

[0074] Figure 3 is a calculation flow chart of the reflectance spectrum response of each pixel point of the present invention;

[0075] Figure 4 It is a flow chart for obtaining spectral characteristic information of the present invention;

[0076] Figure 5 A specific flow chart for scoring each feature for the present invention;

[0077] Figure 6 A flowchart for obtaining defect type results of the present invention;

[0078] Figure 7 This is a flow chart for obtaining dynamic monitoring records of industrial images according to the present invention. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0080] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0081] See also Figure 1 , the industrial image real-time analysis system includes:

[0082] The image capture and difference calculation module continuously captures industrial image sequences through the camera, performs pixel-level comparison on every two frames of images, calculates pixel difference values, identifies significant change areas, and outputs surface difference data;

[0083] The spectrum analysis and feature extraction module uses a spectrum scanner to analyze the key frames in the industrial image sequence, calculates the reflectance spectrum response of each pixel point, and extracts the spectrum features of each area in the image based on the calculation results combined with the surface difference data to obtain the spectrum feature information;

[0084] The defect recognition and classification module compares defect features based on spectral feature information, scores each feature, compares the scoring results with the known defect database, determines the surface state, identifies the defect type and determines the defect area, and generates defect type analysis records;

[0085] The real-time monitoring and alarm module continuously monitors the defective area based on the defect type analysis records, performs continuous data comparison, captures the image changes of the defective area in real time, identifies the defect evolution trend, and issues alarm information in real time to obtain industrial image dynamic monitoring alarm records.

[0086] Surface difference data includes pixel difference values ​​and significant change area identification results; spectral feature information includes reflection spectrum response results and spectrum feature analysis results; defect type analysis records include defect feature scores, defect types and defect areas; industrial image dynamic monitoring alarm records include defect evolution trend analysis results and alarm information.

[0087] See also Figure 2 , the steps to obtain surface difference data are:

[0088] Capture continuous industrial images through the camera, select two frames of images, and use the formula:

[0089] ,

[0090] Calculate the pixel-by-pixel difference value D of the two frames of images to obtain preliminary difference data, where: Represents the first frame image, Represents the second frame image, and C are the adjustment coefficient and correction constant used to adjust the accurate control of the difference sensitivity;

[0091] Based on preliminary difference data, the formula is used,

[0092] ,

[0093] and

[0094] ,

[0095] Calculate the average difference and standard deviation , get the average difference and standard deviation data, where m is the total number of pixels, and It is a weight parameter adjusted according to actual measurement conditions to increase the flexibility and accuracy of calculation;

[0096] Based on the mean difference and standard deviation data, the formula is used,

[0097] ,

[0098] Calculate the screening threshold T, retain pixels with difference values ​​greater than T, identify areas of significant change, and generate surface difference data, where k is a coefficient selected based on image characteristics and is used to identify substantial changes.

[0099] and : Represents the pixel values ​​of two frames of images captured continuously, collecting and They are 120 and 100 respectively (the grayscale value of a certain pixel).

[0100] : Adjustment coefficient, used to control the influence of correction constant C on calculation, set to 0.5.

[0101] C: Correction constant, set to 10, to adjust the sensitivity to noise.

[0102] Substitute into the formula and calculate:

[0103] ,

[0104] The result 20.12 indicates the difference between the two frames of images at this pixel point. The larger the difference value, the greater the change at this point.

[0105] The above calculation results D are collected as [20.12, 15.5, 18.3, 12.2].

[0106] and : Adjustment coefficient, used to increase the flexibility of calculation, setting .

[0107] m: total number of pixels, which is 4.

[0108] calculate :

[0109] , calculate the average difference and standard deviation :

[0110]

[0111] k: weight coefficient, used to adjust the threshold sensitivity, set to 1.5.

[0112] Continue to calculate the screening threshold:

[0113] ,

[0114] The final threshold of 21.23 is used to filter out important pixel differences, that is, only pixels with a difference value greater than 21.23 are considered to have significant changes, which helps to identify key change areas in the image.

[0115] See also Figure 3 , the calculation steps of the reflectance spectrum response of each pixel are:

[0116] Use spectral scanners to capture raw light intensity data for each pixel in industrial image sequences ,in, Represents the light intensity of pixel (i, j) at wavelength x, generating raw light intensity data;

[0117] Based on the original light intensity data, wavelength integration is performed using the formula,

[0118] ,

[0119] Calculate the total spectral response value of pixel (i, j) in the target wavelength range , generates a spectral response matrix, where represents the wavelength interval, and They are the minimum and maximum wavelengths, respectively, used to determine the wavelength range;

[0120] The spectral response matrix is ​​integrated and statistically analyzed using the formula,

[0121] ,

[0122] and

[0123] ,

[0124] Calculate the spectral average and mutation value , extract key features and generate a spectral response overview, where K is the number of wavelength samples, indicating the number of samples in the entire wavelength range.

[0125] A spectral scanner is used to scan the wavelength range from 500nm to 700nm, and the light intensity is measured at 50nm intervals. The light intensity of pixel (1, 1) at a wavelength of 500nm is collected as 120 units (read directly from the scanner), providing intuitive information about the reflectivity of the pixel point to light of different wavelengths.

[0126] Wavelength interval .

[0127] For pixel (1, 1), the light intensity at 500nm, 550nm, 600nm, 650nm and 700nm is 120, 130, 125, 115 and 110 units respectively, then:

[0128] ,

[0129] The value is the total spectral response of the pixel in the specified wavelength range, which helps to identify the material characteristics or state of the pixel.

[0130] Determine the wavelength sample K=5, and combine the above calculation results to get:

[0131] , ,

[0132] average value and mutation value Provides detailed information about the consistency of pixel spectral response, helping to identify heterogeneity in industrial imagery. The mean response shows the average behavior of a pixel across the entire wavelength range, while the variability reveals differences in response between pixels or under different conditions for the same pixel.

[0133] See also Figure 4 , the steps to obtain spectral feature information are:

[0134] The total spectral response value is fused with the surface difference data, using the formula,

[0135] ,

[0136] Calculate data fusion value , and the fused spectral data is obtained, where g is the weight coefficient, which is used to adjust the influence of surface differences. is the surface difference value;

[0137] The fused spectral data is deeply analyzed using the formula,

[0138] ,

[0139] Calculate spectral eigenvalues , extract the significant spectral features of each region and generate spectral feature data;

[0140] Statistical analysis is performed based on the spectral feature data to evaluate the contribution and importance of the features in image analysis and integrate to generate spectral feature information.

[0141] : Weight coefficient, which determines the degree of influence of surface differences on spectral data.

[0142] : For pixel (1, 1), it can be directly obtained using image analysis software and is 0.05.

[0143] Combined with the above calculation results Unit, then:

[0144]

[0145] It represents the adjustment result of spectral data after considering surface differences, which helps to understand how surface changes affect the spectral characteristics of materials and can more accurately reflect the impact of surface treatment or environmental changes on the spectral characteristics of materials.

[0146] For pixel (1, 1), we know .

[0147] Substituting into the formula, we get:

[0148] ,

[0149] The value of represents the spectral feature obtained after a specific transformation. The extracted spectral features reveal the unique spectral behavior of each region, which is crucial for understanding the complex properties of materials. For example, a specific spectral peak can indicate the presence of a specific chemical component, while the width and shape of the spectral feature can provide information about the material structure.

[0150] See also Figure 5 , the specific steps for scoring each feature are:

[0151] Based on the spectral feature information, the feature vector is extracted, and the feature vector data set is obtained through data conversion and normalization. ,in, represents the feature vector of the i-th detection area, Indicates the characteristic intensity of the difference spectrum interval;

[0152] Based on the feature vector dataset, for each feature in the feature vector To score, use the formula,

[0153] ,

[0154] Calculate the score of the nth feature in the i-th region , we get the feature scoring dataset, where The weight factor is determined based on historical data analysis, where N represents the number of features;

[0155] Based on the feature score dataset, the feature scores of each region are summarized using the formula,

[0156] ,

[0157] Calculate the total score of region i , output the score summary data.

[0158] For a certain area, the spectrometer measured light intensities of 150 units, 160 units, and 170 units at wavelengths of 500 nm, 600 nm, and 700 nm, respectively. The eigenvector .

[0159] , represents the weight factor of the feature.

[0160] calculate Rating:

[0161]

[0162] calculate Rating:

[0163]

[0164] calculate Rating:

[0165]

[0166] The results show that within this region, the characteristics The scores are 2250, 2560 and 2890 respectively, which shows the contribution of each spectral feature to defect identification and reflects the importance of the feature.

[0167] Combining the above calculation results, substituting into the formula:

[0168]

[0169] It represents a comprehensive score of all features, indicating the overall status of the area and providing a basis for subsequent defect identification.

[0170] See also Figure 6 , the steps to obtain the defect type results are:

[0171] Based on the score summary data, it is compared with the standard score in the defect database and the formula is used.

[0172]

[0173] Calculate the similarity quantification between the total score and the database defect standard , get the matching score data, where is the standard score for each defect in the database;

[0174] Based on the matching score data, the defect type of each area is determined using the formula,

[0175]

[0176] The defect type confirmation result Y is obtained, where: is the reliability weight corresponding to defect q, which is used to adjust the trust. is a small constant used for mathematical stability, Q represents the defect type;

[0177] Based on the defect type confirmation results, the identified defect type is spatially located, the defect area is marked, and a defect type analysis record is generated.

[0178] , which represents the standard score of a defect type in the database.

[0179] Combined with the above calculation results , substitute into the formula to calculate the matching degree:

[0180]

[0181] It reflects the similarity between the total score of each region and the database defect standard, which is very close to 1, indicating a strong correlation.

[0182] set up , making the calculation more sensitive to the matching score, , ensuring the stability of the division operation, Indicates three possible defect types.

[0183] The calculated values ​​are {0.9993, 0.85, 0.75}, corresponding to three different defect types. Substitute them into the formula and calculate them respectively:

[0184]

[0185]

[0186]

[0187] Then select the defect type corresponding to the highest normalized matching score:

[0188]

[0189] Final calculation results Indicates that the most likely defect type for region i is type 1. This is based on the highest matching score with the known defect database, that is, the characteristics of this region are closest to the typical characteristics of type 1 defects.

[0190] Based on the confirmation results, spatial positioning is performed and the defect area is marked to obtain a defect type analysis record.

[0191] See also Figure 7 ,The steps for obtaining industrial image dynamic monitoring records are:

[0192] According to the defect type analysis record, configure the monitoring system parameters, using the formula,

[0193]

[0194] Calculate the configuration sensitivity of the monitoring system to the i-th region , get the monitoring system configuration sensitivity data, where, represents the defect type score of the ith region, a and b are parameters for adjusting sensitivity;

[0195] Based on the sensitivity data of the monitoring system configuration, real-time data comparison is performed to identify key changes and adopt formulas.

[0196]

[0197] Calculate the difference in images captured during real-time monitoring , determine the defect evolution trend and obtain real-time image change analysis data, where P represents the threshold parameter, which is used to control the error tolerance. and Represents real-time image data at two consecutive moments;

[0198] According to the real-time image change analysis data, it is compared with the preset alarm threshold W. If it is greater than W, an alarm is triggered, otherwise no alarm is triggered, and the industrial image dynamic monitoring alarm record is obtained.

[0199] Set a=0.5, b=10 to adjust the sensitivity.

[0200] Combined with the above calculation results, the defect type score of the area , substitute into the formula to calculate :

[0201] ,

[0202] Sensitivity value It indicates that the monitoring system for this area is adjusted to a very high response sensitivity, which means that the system is highly sensitive to such defects and is suitable for rapid response monitoring scenarios.

[0203] The image value captured in two consecutive monitorings is Units and unit.

[0204] Set the error tolerance threshold to 5 and substitute it into the formula to get:

[0205]

[0206] Difference It indicates that there is a significant change within the area between two consecutive monitorings, exceeding the set error tolerance threshold, indicating the development or change of defects.

[0207] The preset alarm threshold W is set to 12, and it is obvious that 15>12, which directly triggers the alarm. This means that according to the real-time image analysis, it has been determined that the defect will directly affect production safety or product quality and requires manual intervention and correction.

[0208] The industrial image real-time analysis method includes the following steps:

[0209] S1: Capture continuous industrial images through the camera, select two frames of images, analyze the pixel-by-pixel difference between the two frames of images, calculate the average difference value and standard deviation, determine the screening threshold and perform pixel point screening, identify the significant change area, and generate surface difference data;

[0210] S2: Use a spectral scanner to capture the raw light intensity data of each pixel in the industrial image sequence, perform wavelength integration, calculate the total spectral response value of the pixel in the target wavelength range, perform statistical analysis, extract key features, and generate a spectral response overview;

[0211] S3: Fusion of the spectral response overview with the surface difference data, and in-depth analysis. By calculating the spectral eigenvalues, the significant spectral features of each region are extracted, the contribution and importance of the features in image analysis are evaluated, and spectral feature information is generated.

[0212] S4: Extract feature vectors based on spectral feature information, score each feature in the feature vector through data conversion and normalization, perform summary processing, and output score summary data;

[0213] S5: Based on the score summary data, compare it with the standard score in the defect database, calculate the similarity between the total score and the database defect standard, determine the defect type in each area, spatially locate the defect type, mark the defect area synchronously, and generate a defect type analysis record;

[0214] S6: Analyze the records based on the defect type, perform real-time monitoring, conduct real-time data comparison, identify key changes, calculate the difference in images captured during real-time monitoring, determine the defect evolution trend, compare with the preset alarm threshold, determine the alarm information, and obtain the industrial image dynamic monitoring alarm record.

[0215] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. Industrial image real-time analysis system, characterized by: The system comprises: The image capture and difference calculation module continuously captures industrial image sequences through the camera, performs pixel-level comparison on every two frames of images, calculates pixel difference values, identifies significant change areas, and outputs surface difference data; The spectrum analysis and feature extraction module uses a spectrum scanner to analyze the key frames in the industrial image sequence, calculates the reflectance spectrum response of each pixel point, and extracts the spectrum features of each area in the image based on the calculation results combined with the surface difference data to obtain the spectrum feature information; The defect recognition and classification module compares defect features according to the spectral feature information, scores each feature, compares the scoring result with the known defect database, determines the surface state, identifies the defect type and determines the defect area, and generates a defect type analysis record; The real-time monitoring and alarm module continuously monitors the defect area based on the defect type analysis record, performs continuous data comparison, captures image changes in the defect area in real time, identifies defect evolution trends, and issues alarm information in real time to obtain industrial image dynamic monitoring alarm records; The calculation steps of the reflectance spectrum response of each pixel are as follows: Use spectral scanners to capture raw light intensity data for each pixel in industrial image sequences ,in, Represents the light intensity of pixel (i, j) at wavelength x, generating raw light intensity data; Based on the raw light intensity data, wavelength integration is performed using the formula, , calculate the total spectral response value of pixel (i, j) in the target wavelength range , generates a spectral response matrix, where represents the wavelength interval, and They are the minimum and maximum wavelengths, respectively, used to determine the wavelength range; The spectral response matrix is ​​integrated and statistically analyzed using the formula: ,and , calculate the spectral average and mutation value , extract key features and generate a spectral response overview, where K is the number of wavelength samples, indicating the number of samples in the entire wavelength range; The steps for obtaining the spectral characteristic information are: The total spectral response value is fused with the surface difference data, using the formula, , calculate the data fusion value , and the fused spectral data is obtained, where g is the weight coefficient, which is used to adjust the influence of surface differences. is the surface difference value; The fused spectral data is deeply analyzed using the formula: ， Calculate spectral eigenvalues , extract the significant spectral features of each region and generate spectral feature data; Statistical analysis is performed based on the spectral feature data to evaluate the contribution and importance of the features in image analysis and integrate to generate spectral feature information.

2. The industrial image real-time analysis system according to claim 1, characterized in that: The steps for obtaining the surface difference data are as follows: Capture continuous industrial images through the camera, select two frames of images, and use the formula: , calculate the pixel-by-pixel difference value D of the two frames of images and obtain preliminary difference data, where, Represents the first frame image, Represents the second frame image, and C are the adjustment coefficient and correction constant used to adjust the accurate control of the difference sensitivity; Based on the preliminary difference data, the formula is used, ， and , calculate the average difference and standard deviation , get the average difference and standard deviation data, where m is the total number of pixels, and It is a weight parameter adjusted according to actual measurement conditions to increase the flexibility and accuracy of calculation; Based on the mean difference and standard deviation data, the formula is used, , calculate the screening threshold T, retain the pixels with difference values ​​greater than T, identify the significant change areas, and generate surface difference data, where k is a coefficient selected based on image characteristics and is used to identify substantial changes.

3. The industrial image real-time analysis system according to claim 1, characterized in that: The specific steps for scoring each feature are: Based on the spectral feature information, feature vectors are extracted, and feature vector data sets are obtained through data conversion and normalization. ,in, represents the feature vector of the i-th detection area, Indicates the characteristic intensity of the difference spectrum interval; Based on the feature vector data set, for each feature in the feature vector To score, use the formula, ， Calculate the score of the nth feature in the i-th region , we get the feature scoring dataset, where The weight factor is determined based on historical data analysis, where N represents the number of features; Based on the feature score dataset, the feature scores of each region are summarized using the formula, ， Calculate the total score of region i , output the score summary data.

4. The industrial image real-time analysis system according to claim 3, characterized in that: The steps for obtaining the defect type result are: Based on the score summary data, it is compared with the standard score in the defect database, using the formula, , calculate the similarity quantification value between the total score and the database defect standard , get the matching score data, where is the standard score for each defect in the database; Based on the matching score data, the defect type of each area is determined using the formula, , and obtain the defect type confirmation result Y, where: is the reliability weight corresponding to defect q, which is used to adjust the trust. is a small constant used for mathematical stability, Q represents the defect type; Based on the defect type confirmation result, the identified defect type is spatially located, the defect area is marked, and a defect type analysis record is generated.

5. The industrial image real-time analysis system according to claim 4, characterized in that: The steps for obtaining the industrial image dynamic monitoring record are as follows: According to the defect type analysis record, configure the monitoring system parameters, using the formula, ， Calculate the configuration sensitivity of the monitoring system to the i-th region , get the monitoring system configuration sensitivity data, where, represents the defect type score of the ith region, a and b are parameters for adjusting sensitivity; Based on the sensitivity data of the monitoring system configuration, real-time data comparison is performed to identify key changes and adopt the formula, ， Calculate the difference in images captured during real-time monitoring , determine the defect evolution trend and obtain real-time image change analysis data, where P represents the threshold parameter, which is used to control the error tolerance. and Represents real-time image data at two consecutive moments; According to the real-time image change analysis data, it is compared with the preset alarm threshold W. If it is greater than W, an alarm is triggered, otherwise no alarm is triggered, and the industrial image dynamic monitoring alarm record is obtained.

6. A method for real-time analysis of industrial images, characterized in that: The industrial image real-time analysis system according to any one of claims 1 to 5 is implemented, comprising the following steps: Capture continuous industrial images through the camera, select two frames of images, analyze the pixel-by-pixel difference between the two frames, calculate the average difference value and standard deviation, determine the screening threshold and perform pixel point screening, identify the significant change area, and generate surface difference data; Use a spectral scanner to capture the raw light intensity data of each pixel in the industrial image sequence, perform wavelength integration, calculate the total spectral response value of the pixel in the target wavelength range, perform statistical analysis, extract key features, and generate a spectral response overview; The spectral response overview is fused with the surface difference data, and a deep analysis is performed to extract the significant spectral features of each region by calculating the spectral feature values, evaluate the contribution and importance of the features in the image analysis, and generate spectral feature information; Based on the spectral feature information, a feature vector is extracted, each feature in the feature vector is scored through data conversion and normalization, and a summary process is performed to output score summary data; Based on the score summary data, the score is compared with the standard score in the defect database, the similarity between the total score and the database defect standard is calculated, the defect type of each area is determined, the defect type is spatially located, the defect area is synchronously marked, and a defect type analysis record is generated; According to the defect type analysis record, real-time monitoring is performed, real-time data comparison is carried out, key changes are identified, the difference in images captured during real-time monitoring is calculated, the defect evolution trend is determined, and compared with the preset alarm threshold, the alarm information is determined, and the industrial image dynamic monitoring alarm record is obtained.

Images