High-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology

Abstract

The invention discloses a high-precision instrument and equipment detection system based on hyperspectral high-dimensional vision technology, relates to the technical field of hyperspectral imaging optimization, and is used to improve the detection error problem caused by large differences between different detection targets when switching detection targets. The system comprises the following steps: collecting target object data and detection equipment data, collecting switching target object data, comprehensively checking switching target object data and target object data to confirm switching target differences, comprehensively building an elastic net regression model with switching target differences and detection equipment data to determine whether to adjust detection equipment parameters, and after adjusting detection equipment parameters, dividing the switching target object into regions, determining high-dimensional spectral information of each region, collecting target region data, using an absolute difference threshold method and a logistic regression method to determine detection errors of each region of the switching target object, and correcting and adjusting the detection equipment according to the detection errors, thereby reducing the detection errors caused when switching detection target objects.

Description

Technical Field

[0001] The present invention relates to the technical field of hyperspectral imaging optimization, and more specifically, to a high-precision instrument and equipment detection system based on hyperspectral high-dimensional vision technology. Background Art

[0002] Bandwidth management technology is a technology that users use to allocate and distribute network bandwidth resources to meet the needs of different applications and users. Bandwidth management technology applied to the switch power failure alarm of the railway video surveillance system can make the power failure alarm more timely and convenient, thereby reducing railway safety risks.

[0003] The prior art has the following deficiencies:

[0004] Hyperspectral and high-dimensional vision technology can obtain high-dimensional spectral information of target objects when inspecting high-precision instruments. However, detection targets of different materials have different effects on the detection results. When switching the target object for inspection, the original detection equipment parameters may cause huge detection errors. Summary of the invention

[0005] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology, which optimizes the detection equipment in a targeted manner by analyzing the combination of measurement errors and influencing errors to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology includes: data acquisition module, adjustment confirmation module, area division module and error correction module, and signal connection between each module;

[0008] The data acquisition module is used to collect target object data and detection equipment data, classify the target object according to the target object data and determine whether to enter the active correction mechanism. If the active correction mechanism is entered, the adjustment signal is transmitted to the error correction module. Otherwise, the switching target data is collected and transmitted together with the target object data and the detection equipment data to the adjustment confirmation module;

[0009] The adjustment confirmation module is used to confirm the switching target difference, build an elastic net regression model based on the switching target difference and the detection equipment data, and make an initial evaluation of the switching target detection accuracy through the elastic net regression model, and send the evaluation result to the area division module or the error correction module;

[0010] The region division module is used to divide the switching target object into regions, detect the high-dimensional spectral information of each region of the switching target object, collect target region data, use the absolute difference threshold method and the logistic regression method to determine the detection error of the switching target corresponding region, and transmit the detection error result and the adjustment signal to the error correction module;

[0011] The error correction module is used to make corresponding adjustments to the detection equipment according to the adjustment signals transmitted by different modules.

[0012] In a preferred embodiment, the target object data includes the surface smoothness and light guide data of the target object, the switching target data includes the surface smoothness and light guide data of the switching target object, the detection device data is the sensor sensitivity, and the light guide data is whether the object itself has luminous characteristics.

[0013] In a preferred embodiment, the data acquisition module classifies the target objects according to the target object data, that is, the target objects with luminous characteristics are classified as Class I objects according to the light guide data; and the target objects without luminous characteristics are classified as Class II objects.

[0014] In a preferred embodiment, the active correction mechanism reduces the accumulation of the self-luminescence of the target object on a single frame image by shortening the exposure time of the detection device, thereby avoiding detection errors caused by the self-luminescence of the target object.

[0015] In a preferred embodiment, the adjustment confirmation module compares the surface smoothness of the target object and the switching target object to confirm the switching target difference, and constructs an elastic net regression model by recording the switching target difference and sensor sensitivity at multiple temperatures and merging them into an object change data set and a sensitivity data set respectively.

[0016] In a preferred embodiment, the specific steps of adjusting the confirmation module to construct the elastic net regression model are as follows:

[0017] Data preparation: Combine the object change dataset and the sensor sensitivity dataset into the input matrix X, mark the target as y, and set the ratio to divide the input matrix into a training set and a test set;

[0018] Model construction: Construct the objective function of the elastic net regression model;

[0019] Model evaluation: Use the training set and test set to calculate the objective function and set the parameter weight of the objective function according to the calculation results;

[0020] Evaluate detection accuracy: Select the minimum value in the calculated objective function and compare it with the preset evaluation threshold to evaluate the detection accuracy.

[0021] In a preferred embodiment, when the region division module divides the switching target object into regions, a sliding window is set to first crop the image. When the cropped image only contains the image features of the switching target object, N sliding windows are set to divide the image features of the switching target object into N regions.

[0022] In a preferred embodiment, the target area data in the area division module includes regional heat and regional area, the influence error is analyzed based on the target area data, the measurement error is analyzed based on the high-dimensional spectral information of each area, and the detection error of the switching target corresponding area is determined by combining the influence error and the measurement error.

[0023] In a preferred embodiment, the error correction module performs an active correction mechanism when it receives an adjustment signal from the data acquisition module; adjusts the detection equipment parameters when it receives an adjustment signal from the adjustment confirmation module; and detects error correction parameters according to the switching target corresponding area when it receives an adjustment signal from the area division module.

[0024] The technical effects and advantages of the high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology of the present invention are as follows:

[0025] The present invention collects target object data and detection equipment data and records them separately, collects switching target object data, comprehensively checks the switching target object data and the target object data to confirm the switching target difference, comprehensively checks the switching target difference and the detection equipment data to determine whether to adjust the detection equipment parameters, determines the adjustment direction by analyzing the switching target difference and the detection equipment data, reduces the workload, and after adjusting the detection equipment parameters, divides the switching target object into regions, determines the high-dimensional spectral information of each region, collects target region data, determines the detection error of each region of the switching target object, and corrects and adjusts the detection equipment according to the detection error, thereby reducing the detection error caused when switching the detection target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a schematic diagram of a high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology of the present invention.

[0027] Figure 2 This is a flow chart of the high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology of the present invention. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0029] The present invention collects target object data and detection equipment data and records them separately, collects switching target object data, comprehensively checks the switching target object data and the target object data to confirm the switching target difference, comprehensively checks the switching target difference and the detection equipment data to determine whether to adjust the detection equipment parameters, and after adjusting the detection equipment parameters, divides the switching target object into regions, determines the high-dimensional spectral information of each region, collects target region data, determines the detection error of each region of the switching target object, and corrects and adjusts the detection equipment according to the detection error, thereby reducing the detection error caused when switching the detection target object.

[0030] Examples, such as Figure 1 and Figure 2 As shown, the high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology includes: a data acquisition module, an adjustment confirmation module, a region division module and an error correction module, and the signals between the modules are connected.

[0031] The functions of each module are as follows:

[0032] The data acquisition module is used to collect and record target object data and detection equipment data, classify the target object according to the target object data and determine whether to enter the active correction mechanism, and transmit the judgment result and adjustment signal to the error correction module. When the judgment result is not to enter the correction mechanism or after receiving the correction completion signal from the error correction module, the switching target object data is collected and transmitted to the adjustment confirmation module.

[0033] The adjustment confirmation module is used to receive target object data, switching target object data and detection equipment data, confirm the switching target difference based on the target object data and the switching target object data, build an elastic net regression model based on the switching target difference and the detection equipment data, and perform an initial evaluation of the switching target detection accuracy through the elastic net regression model, and send the evaluation result and adjustment signal to the area division module or the error correction module according to the evaluation result type.

[0034] The region division module is used to receive the evaluation results transmitted by the adjustment confirmation module and divide the switching target object into regions, detect the high-dimensional spectral information of each region of the switching target object, collect the target region data, and use the absolute difference threshold method and the logistic regression method to determine the detection error of the switching target corresponding region based on the target region data and the high-dimensional spectral information detection results of the target object corresponding region, and transmit the detection error result and the adjustment signal to the error correction module.

[0035] The error correction module is used to receive adjustment signals from different modules and make corresponding adjustments. When receiving adjustment signals from the data acquisition module, it enters the active correction mechanism; when receiving adjustment signals from the adjustment confirmation module, it adjusts the detection equipment parameters; when receiving adjustment signals from the area division module, it performs secondary corrections on the detection equipment parameters.

[0036] It should be noted that the target object is the detected object for analysis, and the switching target object is the target object to be detected after the detection target object is switched. The active correction mechanism is aimed at the target object that has the ability to affect the detection results. The active correction mechanism improves the detection accuracy of the target object and provides a data basis for the subsequent analysis of the switching target difference. The elastic net regression model is a machine learning algorithm with multiple functions such as predictive analysis, feature selection and high-dimensional data modeling. It can be used to evaluate the detection accuracy of the switching target. The convolutional neural network is a special deep learning neural network that is widely used in the field of image and video analysis. By setting structural layers with different functions, it can extract low-level to high-level features in the image layer by layer, and finally complete tasks such as classification and detection. It can be used to process environmental data and target area data and detect the degree of impact.

[0037] The target object data includes the surface smoothness and light guide data of the target object, the switching target data includes the surface smoothness and light guide data of the switching target object, and the detection device data is the sensor sensitivity. The light guide data is whether the target object itself has luminous characteristics and its light intensity. Laser interferometers can be used to detect the light guide data of target objects and switching target objects.

[0038] It should be noted that the hyperspectral imaging system can obtain high-dimensional spectral information of the target object from visible light to near infrared. The greater the illumination intensity of the self-luminous target object, the more likely the detector of the hyperspectral imaging system will be saturated, causing spectral distortion and affecting the accuracy of the target object analysis. Surface smoothness can also be detected using a laser interferometer. The sensor sensitivity in the equipment data is the response sensitivity of the hyperspectral imaging sensor to photons. The hyperspectral imaging sensor can capture the high-dimensional spectral characteristics of the target object. The higher the sensor sensitivity, the more accurately the high-dimensional spectral characteristics of the target object are captured. The quantum efficiency test can be used to detect the sensor sensitivity. The sensor sensitivity is determined by measuring the sensor's response to incident photons under standard spectral irradiance conditions. The quantum efficiency test is an existing method and will not be described in detail here.

[0039] When the data acquisition module classifies the target object, it is connected to a laser interferometer and uses the light guide data provided by the laser interferometer to determine whether the target object has luminous features and analyze them. When the target object has luminous features, the target object is classified as a Class I object; when the target object does not have luminous features, the target object is classified as a Class II object.

[0040] The contrast change of interference fringes measured by the laser interferometer at different temperatures can reflect the light intensity of the target object. The data acquisition module accesses the detection history, randomly selects N detection targets in the detection history, obtains the contrast corresponding to the detection targets through the laser interferometer, and calculates the average value as the contrast threshold. When the data acquisition module classifies the target object, it compares the contrast of the target object with the contrast threshold to determine whether the target object has luminous characteristics. When the contrast of the target object is lower than the contrast threshold, it is determined that the target object has luminous characteristics.

[0041] It should be noted that the working principle of the laser interferometer is to irradiate the laser onto the surface of the object to be measured to generate an interference fringe image. When the object itself is luminous, uneven light and dark areas will appear in the interference fringe image. The brightness of the interference fringes in the luminous area will be higher than that in the non-luminous area. The contrast of the interference fringes is measured, that is, the ratio of the brightness difference between the light and dark fringes to the average brightness. The luminescence of the object to be measured will increase the background light intensity, thereby reducing the contrast of the interference fringes. When the contrast of the interference fringes of the object to be measured is lower than the contrast threshold, it is judged that the object to be measured has a high-intensity luminescence characteristic.

[0042] When the target object is a type I object, the ratio of the contrast threshold to the contrast of the target object is used as an adjustment coefficient, and the data acquisition module transmits the adjustment coefficient and the adjustment signal to the error correction module.

[0043] The data acquisition module detects the surface smoothness of the target object and the switching target object at different temperatures and takes the difference to obtain the absolute value to obtain the target object change coefficient. The data acquisition module detects the sensor sensitivity at the corresponding temperature. The target object change coefficient at different temperatures is used as the object change data set; the sensor sensitivity detected at the corresponding temperature is used as the sensitivity data set. The data acquisition module transmits the object change data set and the sensitivity data set to the adjustment confirmation module.

[0044] It should be noted that the data acquisition module classifies the target objects and chooses whether to pass them to the error correction module for active correction according to the type of the target object. By screening first and then adjusting, the targeting of the detection target is improved and the operating efficiency of the overall system is improved.

[0045] The adjustment confirmation module constructs an elastic net regression model based on the received object change dataset and sensitivity dataset, and initially evaluates the accuracy of switching target detection based on the elastic net regression model. The specific steps are as follows:

[0046] Data preparation: Combine the object change dataset and the sensor sensitivity dataset into the input matrix X, mark the target as y, and divide the input matrix into a training set and a test set;

[0047] Model construction: Construct the objective function of the elastic net regression model:

[0048] ,in, Output the minimum value of the objective function. is the i-th test set matrix, is the i-th training set matrix, represents the inner product of the i-th sample training set matrix and the weight vector, Represents the difference between the true value and the predicted value, that is, the difference between the test set matrix and the training set matrix. is the sum of squares of all sample prediction errors, and the denominator It is only used to normalize the sum of squared prediction errors to make it an average value; in addition, is the L1 regularization term to prevent the model from overfitting. is the regularization coefficient, L1 regularization coefficient, the value range is [0,1], used to control the weight of L1 regularization term , is the L2 regularization term, which is used to control the weight of the L2 regularization term .

[0049] Model evaluation: By setting multiple groups of different ratios to divide the training set and test set, set the weight vector and the weight of the regularization term respectively. The weight vector can be set with reference to the data size of the object change data set and the sensor sensitivity data set. For example, the weight vector is set to the normalized result of the data in the object change data set and the sensor sensitivity data set at different temperatures. The data can be divided by the maximum value in the corresponding data set to obtain the normalized result of the corresponding data. The larger the data in the object change data set or the sensor sensitivity data set, the larger the corresponding weight vector.

[0050] Evaluate detection accuracy: Select the minimum value of each group of calculated objective function data results and compare it with the preset evaluation threshold. If the selected minimum value exceeds the evaluation threshold, it is judged that the switching target detection accuracy is low, and the evaluation result and adjustment signal are sent to the error correction module; otherwise, it is judged that the switching target detection accuracy is high, and the evaluation result is sent to the area division module.

[0051] The adjustment confirmation module analyzes the detection accuracy of the switching target through the elastic net regression model, and performs error correction on the switching target with low detection accuracy, thereby reducing the uncertainty impact caused by other factors in the subsequent error analysis.

[0052] It should be noted that the regularization item weights, regularization coefficients and L1 regularization coefficients involved in the above steps can be set according to actual conditions. For example, and are set to 0.4 and 0.6 respectively, and the regularization coefficient Set to 0.8, the L1 regularization coefficient It is set to 0.5, etc., and there is no restriction here. The evaluation threshold can be calculated by randomly setting multiple groups of different regularization item weights to calculate the output results of the objective function, and the average value is selected as the evaluation threshold in the obtained output results of the objective function.

[0053] After receiving the evaluation results, the region division module divides the image to be tested of the switching target object into regions and sets a sliding window of appropriate size to ensure that the sliding window contains all image features of the switching target objects in the image. For example, the image is cropped so that the image only contains the image features of the switching target object, and then the sliding window is set to 1 / 4 of the cropped image, so that the image features of the switching target object can be divided into 4 regions.

[0054] Detect the spectral features of each area separately, that is, obtain the spectral curve graph of each area, use the absolute difference threshold method to set the difference threshold, the difference threshold can be set to the peak position difference N and the reflection intensity difference M, for example, set the peak position difference to 10nm, and the reflection intensity difference to 5%, calculate the average value of the peak position and the average value of the reflection intensity in the spectral curve graph of all areas, and subtract the peak position and the reflection intensity in the spectral curve graph of each area from the peak position average value and the reflection intensity average value to obtain the peak difference coefficient and the reflection difference coefficient, compare the peak difference coefficient and the reflection difference coefficient with the peak position difference and the reflection intensity difference in the difference threshold, set the regional measurement coefficient, and for areas where the peak difference coefficient exceeds the peak position difference and the reflection difference coefficient exceeds the reflection intensity difference, set the regional measurement coefficient to 1; otherwise, set the regional measurement coefficient to 0.

[0055] It should be noted that for the switching target object, due to the same material, the spectral curves obtained from each part should be highly similar. The difference threshold is set to detect each part separately. The detection error can be judged by whether the detection difference of each part reaches the preset difference threshold. The greater the difference between the parts, the greater the detection error. The detection error is digitized by dividing the area and comparing the spectral curves to provide data support for analyzing the detection error and facilitate subsequent adjustment of the detection setting parameters. The difference threshold can be set according to the expected effect. The difference threshold in this example is only used as an example.

[0056] After setting the regional coefficient, the regional division module collects the target area data for in-depth analysis. The target area data includes regional heat and regional area. When a substance is heated, it will change the molecular bonding state, thereby affecting the absorption or reflection spectrum of the substance; the larger the regional area, the larger the area to be detected, and the detection error increases due to the cumulative effect. Therefore, the greater the regional heat or the larger the regional area, the greater the high-dimensional spectrum detection error.

[0057] It should be noted that it is recommended to use an infrared thermal imager to detect and determine the regional heat. An infrared thermal imager is a non-contact temperature measurement device that can be used to measure the temperature distribution and heat information on the surface of an object. Since physical cutting cannot be performed when switching the target object to detect heat, it is difficult to use traditional thermal sensors to measure regional heat. The infrared thermal imager is used to detect the overall temperature distribution of the switching target object, and then the area is divided. After the temperature distribution of each area is obtained, the average temperature of each area is calculated as the regional temperature of the corresponding area; the regional area can also be detected using a laser interferometer. Since temperature detection and area detection are existing technical means, they are not analyzed here.

[0058] The regional heat and regional area of ​​each region are used to calculate the regional influence coefficient through the logistic regression formula, and the calculation formula is: ,in is the regional influence coefficient, To be the influencing factor, the regional heat and regional area of ​​the corresponding area can be normalized and summed up to obtain, It is a natural base. The regional heat normalization of the corresponding area can be achieved by summing up the regional heat of all areas, and taking the ratio of the regional heat of the corresponding area to the summed result as the result of the regional heat normalization of the corresponding area. The regional area can be normalized in the same way.

[0059] The area division module comprehensively considers the area measurement coefficient and the area influence coefficient to determine the detection error of the switching target corresponding area. The sum of the area measurement coefficient and the area influence coefficient is used as the area coefficient. The area coefficient is compared with the preset detection error threshold to analyze the detection error size. When the area coefficient exceeds the detection error threshold, the detection error is judged to be large; otherwise, the detection error is judged to be small. The detection error result, the area coefficient and the adjustment signal are transmitted to the error correction module.

[0060] The area division module analyzes the detection error of the region by calculating the measurement error and influence error of each region separately, reducing the risk of misjudgment of detection results caused by the singleness of error analysis, greatly improving the accuracy of error analysis, and providing convenience for subsequent adjustments and corrections.

[0061] It should be noted that the above-mentioned detection error threshold in this example provides a setting method, in which the detection error is set to 1. When the regional coefficient exceeds 1, the sum of the regional measurement coefficient and the regional influence coefficient is greater than 1, that is, when the regional measurement coefficient is 1, the measurement error is judged to be large even if the regional influence coefficient is extremely small; when the regional measurement coefficient is 0, the measurement error is judged to be large only when the regional influence coefficient exceeds 1. The setting and adjustment of the detection error threshold invisibly controls the weights of the measurement error and the influence error. The detection error threshold can be set and adjusted according to actual conditions, which will not be repeated here.

[0062] When the error correction module receives the adjustment signal transmitted by the data acquisition module, it performs an active correction mechanism. The active correction mechanism shortens the exposure time by adjusting the coefficient, reduces the accumulation of the target object's self-luminescence on the single-frame image, and thus avoids overexposure and distortion. In the active correction mechanism, the product of the initial exposure time and the adjustment coefficient is used as the new exposure time to complete the correction.

[0063] When the error correction module receives the adjustment signal transmitted by the adjustment confirmation module, it adjusts the parameters of the detection equipment. The error correction module changes the signal-to-noise ratio of the switching target object image by randomly adjusting the spectral resolution. When the signal-to-noise ratio of the switching target object image is within the default signal-to-noise ratio judgment range, the current detection band of the detection equipment is used to detect the switching target object.

[0064] When the error correction module receives the detection error result, regional coefficient and adjustment signal transmitted by the area division module, it allocates the image signal-to-noise ratio judgment interval to each area according to the regional coefficient. When the detection error is judged to be small, the default signal-to-noise ratio judgment interval is adopted; when the detection error is judged to be large, the product obtained by dividing the signal-to-noise ratio judgment interval by the regional coefficient is used as the corrected signal-to-noise ratio judgment interval and recorded.

[0065] After adjusting the signal-to-noise ratio judgment interval of different areas according to the degree of error reflected by the regional coefficient, images within the corresponding signal-to-noise ratio judgment interval of different areas are collected, which improves the accuracy of hyperspectral high-dimensional visual recognition. At the same time, the corrected signal-to-noise ratio judgment interval is included in the database, which is convenient for subsequent detection of similar objects.

[0066] It should be noted that the signal-to-noise ratio is used to measure the ratio of signal strength to noise strength, reflecting the problem of low signal power of the detection equipment due to external reasons. The larger the signal-to-noise ratio, the better the system performance and the more accurate the switch target object detection. When the detection equipment collects images, the signal-to-noise ratio will be reduced due to detection errors caused by non-standard detection or external reasons.

[0067] The error correction module receives the adjustment signal transmitted by the adjustment confirmation module, re-determines the spectral range of the measurement switching target object, and monitors the image signal-to-noise ratio by randomly adjusting the spectral resolution when narrowing the detection band of the detection equipment. When the image signal-to-noise ratio is within the default signal-to-noise ratio judgment interval, the corresponding detection band is selected for detection. The default signal-to-noise ratio judgment interval can be set according to the required accuracy, for example, the default signal-to-noise ratio judgment interval is set to between 30dp and 60dp.

[0068] When the error correction module receives the adjustment signal transmitted by the area division module, it corrects the signal-to-noise ratio judgment interval according to the detection error result, and divides the signal-to-noise ratio judgment interval by the area coefficient to obtain the product as the new signal-to-noise ratio judgment interval. For example, if the area coefficient is 1.2 and the default signal-to-noise ratio judgment interval is between 30dp and 60dp, the default signal-to-noise ratio judgment interval is corrected to between 36dp and 72dp.

[0069] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product.

[0070] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application of the technical solution and the invention constraints. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0071] In addition, each functional module in each embodiment of the present application may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0072] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0073] Finally: 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 spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology, characterized in that: include: Data acquisition module, adjustment confirmation module, area division module and error correction module, and signal connection between each module; The data acquisition module is used to collect target object data and detection equipment data, classify the target object according to the target object data and determine whether to enter the active correction mechanism. If the active correction mechanism is entered, the adjustment signal is transmitted to the error correction module. Otherwise, the switching target data is collected and transmitted together with the target object data and the detection equipment data to the adjustment confirmation module; The adjustment confirmation module is used to confirm the switching target difference, build an elastic net regression model based on the switching target difference and the detection equipment data, and make an initial evaluation of the switching target detection accuracy through the elastic net regression model, and send the evaluation result to the area division module or the error correction module; The region division module is used to divide the switching target object into regions, detect the high-dimensional spectral information of each region of the switching target object, collect target region data, use the absolute difference threshold method and the logistic regression method to determine the detection error of the switching target corresponding region, and transmit the detection error result and the adjustment signal to the error correction module; The error correction module is used to make corresponding adjustments to the detection equipment according to the adjustment signals transmitted from different modules; The target object data includes the surface smoothness and light guide data of the target object, the switching target data includes the surface smoothness and light guide data of the switching target object, the detection device data is the sensor sensitivity, and the light guide data is whether the object itself has a luminous feature; The adjustment confirmation module compares the surface smoothness of the target object and the switching target object to confirm the switching target difference, and constructs an elastic net regression model by recording the switching target difference and the sensor sensitivity at multiple temperatures and merging them into an object change data set and a sensitivity data set respectively; The target area data in the area division module includes regional heat and regional area. The impact error is analyzed according to the target area data, and the measurement error is analyzed according to the high-dimensional spectral information of each area. The detection error of the switching target corresponding area is determined by combining the impact error and measurement error.

2. The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology according to claim 1 is characterized in that: In the data acquisition module, the target objects are classified according to the target object data, that is, the target objects with luminous characteristics are classified as Class I objects according to the light guide data; and the target objects without luminous characteristics are classified as Class II objects.

3. The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology according to claim 2 is characterized in that: The active correction mechanism shortens the exposure time of the detection equipment, reduces the accumulation of the target object's self-luminescence on a single frame image, and avoids detection errors caused by the target object's self-luminescence.

4. The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology according to claim 1 is characterized in that: The specific steps for adjusting the confirmation module to build the elastic net regression model are as follows: Data preparation: Combine the object change dataset and the sensor sensitivity dataset into the input matrix X, mark the target as y, and set the ratio to divide the input matrix into a training set and a test set; Model construction: Construct the objective function of the elastic net regression model; Model evaluation: Use the training set and test set to calculate the objective function and set the parameter weight of the objective function according to the calculation results; Evaluate detection accuracy: Select the minimum value in the calculated objective function and compare it with the preset evaluation threshold to evaluate the detection accuracy.

5. The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology according to claim 1 is characterized in that: When the region division module divides the switching target object into regions, a sliding window is set to crop the image first. When the cropped image only contains the image features of the switching target object, N sliding windows are set to divide the image features of the switching target object into N regions.

6. The high-precision instrument and equipment detection system based on hyperspectral and high-dimensional vision technology according to claim 1 is characterized in that: The error correction module performs an active correction mechanism when receiving the adjustment signal transmitted by the data acquisition module; adjusts the detection equipment parameters when receiving the adjustment signal transmitted by the adjustment confirmation module; and detects error correction parameters according to the switching target corresponding area when receiving the adjustment signal transmitted by the area division module.

Images