A production management system for sights

Abstract

The invention discloses a sight production management system, which relates to the technical field of machine vision and comprises an acquisition module, a processing module, a quality assessment module and a cloud platform module; the acquisition module is used for acquiring a video in the sight and transmitting the video to the processing module; the processing module is used for generating an image set according to a time series of frame divisions of the video, performing preprocessing, adaptive segmentation and feature extraction on the image set, and calculating a loss value, and transmitting the loss value to the quality assessment module; the quality assessment module is used for calculating an assessment value, and transmitting the assessment value to the cloud platform module; the invention saves the storage burden and power consumption of the timer by controlling the staggered operation of the timer; establishes an assessment model, evaluates the comprehensive quality of the sight, and outputs the assessment score to the cloud platform, displays the assessment score through the cloud platform, intuitively sees the quality of the sight, and reduces the workload of human eye recognition.

Description

Technical Field

[0001] The invention relates to the technical field of machine vision, and in particular to a sighting scope production management system. Background Art

[0002] In recent years, with the continuous development of science and technology, machine vision has been widely used. Machine vision is based on optical systems, industrial digital cameras and image processing technology to simulate human vision and thinking. Machine vision has the advantages of non-touch, fast speed, high precision and anti-interference. Its application scope covers industry, agriculture, medicine, military, aerospace, meteorology, astronomy, public security, transportation, security, scientific research and other industries. Improving the automation level of weapons and equipment information acquisition capabilities is the key to improving the intelligence and automation level of equipment.

[0003] At present, a Chinese invention patent with application number CN 117146647 A discloses a method and system for quick adjustment calibration of an optical sight. The method comprises the following steps: receiving a determination signal from a user, transmitting a distance measurement signal from a laser rangefinder, receiving a feedback signal, performing distance grading according to basic information of the optical sight, establishing a distance grade set, generating a predetermined distance based on the feedback signal, and determining a matching relationship. The optical sight is adjusted to a limit point and a median point, performing image retention respectively, taking the median point as an initial point, determining an optimization direction according to the image retention result, and performing adaptive optimization adjustment, including magnification optimization and focal length optimization, determining calibration parameters according to the adaptive optimization adjustment result, and performing quick adjustment calibration of the optical sight. However, the prior art does not establish a comprehensive quality evaluation system for the sight, which is not conducive to qualitatively characterizing the quality of the sight during detection, and does not correct the zero position error of the sight, which is not conducive to the accuracy of the sight. Summary of the invention

[0004] The technical problem solved by the present invention is that: in the prior art, a comprehensive quality evaluation system is not established for the sight, which is not conducive to qualitatively characterizing the quality of the sight during detection, and the zero position error of the sight is not corrected, which is not conducive to the accuracy of the sight.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a sight production management system includes a collection module, a processing module, a quality assessment module and a cloud platform module;

[0006] The acquisition module is used to acquire the video in the sight and transmit the video to the processing module;

[0007] The processing module is used to generate an image set according to a time series of frame divisions of the video, and select one of the frames in the image set as a first image, perform grayscale processing and segmentation processing on the first image to generate a two-dimensional grayscale image and a binary image, the two-dimensional grayscale image performs foreground coverage on the binary image to generate a second image, extract a feature vector of the second image, and calculate a loss value, obtain a contour grayscale curve of the two-dimensional grayscale image, perform adaptive segmentation on the remaining images in the image set according to a first threshold, extract the foreground image of the remaining images in the image set according to the contour grayscale curve, generate a third image, extract the angle feature, distance feature and shape feature of the third image, calculate the loss value of the angle feature, distance feature and shape feature, and transmit the loss value to the quality assessment module;

[0008] The quality assessment module is used to calculate the assessment value and transmit the assessment value to the cloud platform module;

[0009] The cloud platform module is used for zero error correction, display evaluation value and control of the processing module

[0010] As a preferred solution of the sight production management system of the present invention, wherein: the pre-processing unit includes a grayscale processing sub-unit and a segmentation sub-unit;

[0011] The grayscale processing subunit generates an image set by dividing the time series of video frames, selects one frame of the image set as a first image, performs grayscale processing on the first image to generate a two-dimensional grayscale image, and transmits the grayscale image to the segmentation subunit;

[0012] The segmentation subunit segments the foreground image and the background image in the two-dimensional grayscale image based on the separation threshold, defines the pixel points in the two-dimensional grayscale image that are greater than the separation threshold as target information, and records the grayscale value of the target information, extracts the pixel points in the two-dimensional grayscale image that are less than the separation threshold to generate a background image, and records the grayscale value of the background image, binarizes the foreground image and generates a binary image, and covers the binary image with the foreground image of the two-dimensional grayscale image to generate a second image;

[0013] The segmentation subunit obtains the contour grayscale values ​​of the foreground image and the background image and generates a contour grayscale curve. The segmentation subunit adaptively segments the remaining images in the image set based on the first threshold, extracts the foreground image of the remaining images in the image set based on the contour grayscale curve, generates a third image, transmits the second image to the feature extraction unit, transmits the third image to the quality assessment module, and transmits the binary image to the control unit.

[0014] As a preferred solution of the sight production management system of the present invention, the calculation expression of the adaptive segmentation is:

[0015]

[0016] in, is an image in the image set to be segmented, For the third image, is a first threshold value set in the processing module, is the gradient operation of the images in the image set to be segmented.

[0017] As a preferred solution of the sight production management system of the present invention, wherein: the feature extraction unit integrates the attention mechanism into a convolutional neural network, and the convolutional neural network includes an input layer, an output layer, a pooling layer, a convolution layer, and a fully connected layer;

[0018] The pooling layer decomposes the two-dimensional pooling of each pixel in the second image into one-dimensional pooling, performs feature encoding on the one-dimensional pooling, and generates a feature map, the feature map includes the direction information and position information of each pixel in the second image, and inputs the feature map into the fully connected layer, the fully connected layer extracts and fuses the feature information in the feature map based on the classifier, and generates a fused feature vector, the calculation expression of which is:

[0019]

[0020] in, is the fusion feature vector, is the feature map, are learnable weights, is the learning rate of the convolutional neural network. The convolutional neural network uses the fused feature vector as a data set, sets the number of training times and performs comparisons, constructs a loss function, calculates the loss value, and transmits the loss value to the quality assessment module. The fused feature vector includes angle, distance, and shape. The calculation expression of the loss function is:

[0021]

[0022]

[0023] in, is the value of the loss function, is the intersection-and-union ratio, is the dynamic weight, is the distance loss, is the shape loss, is a constant.

[0024] As a preferred solution of the sight production management system of the present invention, wherein: the detection unit includes a first timer and a second timer, the control unit is electrically connected to the cloud platform module, and the control unit is used to control the staggered operation of the first timer and the second timer;

[0025] When the control unit receives a detection instruction from the cloud platform module, the control unit sends a start signal to the first timer, the first timer performs pixel scanning on the pixels of the binary image, obtains the coordinates of the central pixel of the binary image, calculates the zero position deviation m of the pixels of the binary image based on the position of the a priori calibration point pixel, transmits the zero position deviation m to the control unit, and when the zero position deviation m exceeds the zero position threshold, the first timer outputs a zero position deviation excessive signal to the cloud platform module;

[0026] The detection unit generates a zero position threshold based on the position of the a priori calibration point pixel, the first timer is electrically connected to the control unit, and the second timer is electrically connected to the cloud platform module and the control unit.

[0027] As a preferred solution of the sight production management system described in the present invention, wherein: the cloud platform module includes a display screen and a control button, the display screen is used to display an excessive deviation signal and an evaluation value, the control button is used to manually reduce the zero deviation m, and send the reduced zero deviation value to the control unit, the control unit sends a start signal to the second timer when the zero deviation is less than the zero threshold, the second timer performs pixel scanning on the third image, obtains the central pixel coordinates of the third image, obtains the neighborhood image of the central pixel coordinates, calculates the zero deviation n of the neighborhood image of the central pixel coordinates, and transmits the zero deviation n to the control unit;

[0028] The control unit calculates the difference between the zero deviation m and the zero deviation n, and starts the first timer based on a threshold of the difference between the zero deviation m and the zero deviation n. When the difference between the zero deviation m and the zero deviation n is greater than the threshold of the difference between the zero deviation m and the zero deviation n, the control unit starts the first timer, and the first timer performs a weighted update on the zero deviation m and sets the weighted updated zero deviation m as the first timer. Transmitting to the cloud platform module;

[0029] The calculation expression of the weighted update is:

[0030]

[0031]

[0032] in, is the weighted updated value of the zero-position deviation m, is the value of the zero position deviation m, is the value of the zero deviation n, It is the difference between the zero deviation m and the zero deviation n.

[0033] As a preferred solution of the sight production management system of the present invention, wherein: the quality assessment module includes a calculation unit and an assessment model;

[0034] The calculation unit extracts key quality parameters of the third image, calculates a turning threshold of the key quality parameters of the third image by using a neural network, establishes a clouding interval based on the turning threshold, and transmits the clouding interval to the evaluation model, wherein the key quality parameters of the third image include an eye relief, an exit pupil diameter, a magnification, and a parallax angle;

[0035] The evaluation model uses fuzzy hierarchical analysis method to calculate the weight vector of the key quality parameters of the third image, establishes the quality level of the key quality parameters of the third image based on the weight vector and the cloudification interval, and calculates the evaluation value based on the quality level, and the quality level includes excellent, qualified and unqualified.

[0036] As a preferred solution of the sight production management system of the present invention, wherein: the evaluation model constructs a judgment matrix based on the fuzzy analytic hierarchy process, calculates a consistency matrix based on the judgment matrix, and calculates a weight vector of the key quality parameters of the third image based on the consistency matrix;

[0037] The calculation expression of the judgment matrix is:

[0038]

[0039] The calculation expression of the consistency matrix is:

[0040]

[0041] The calculation expression of the weight vector of the key quality parameter of the third image is:

[0042]

[0043] in, is the judgment matrix, is the consistency matrix, is the key quality parameter in the third image Key quality parameters in the third image The comparison value of is the key quality parameter in the third image Key quality parameters in the third image The reciprocal of the comparison value, is the weight vector, is the new matrix obtained after n iterations of the key quality parameters in the third image, is the key quality parameter in the third image go through Iterates and integrates the key quality parameters in the third image go through The new matrix obtained by iteration.

[0044] As a preferred solution of the sight production management system of the present invention, the neural network respectively determines the key quality parameters of the third image. and Normalization is performed to make the third image quality key parameter and Distributed between 0 and 1, set the initial weight vector of competition , topological neighborhood radius and the initial learning rate , based on the weight vector and the key quality parameter of the third image and Determine the winning neuron, obtain the adjacent interval image centered on the winning neuron, and update the adjacent interval image topology and learning rate centered on the winning neuron, based on the quality key parameters of the third image and The convergence of the number of iterations is calculated, the eye distance turning threshold a, the exit pupil diameter turning threshold b, the magnification turning threshold c and the parallax angle turning threshold d are calculated, clouding intervals a~b, b~c and c~d are generated, and the clouding intervals are transmitted to the evaluation model.

[0045] As a preferred solution of the sight production management system of the present invention, the evaluation model is based on the key quality parameters of the clouded interval and the third image. and The weight vector establishes the excellent level p, qualified level q and unqualified level s, and the key quality parameters of the third image and The weight vector is weighted and integrated to calculate the key quality parameters of the third image and The evaluation center of gravity of the weight vector is calculated based on the evaluation center of gravity and the loss value, and the evaluation value is transmitted to the cloud platform module;

[0046] The calculation expression of the evaluation value is:

[0047]

[0048]

[0049] in, is the evaluation value, To evaluate the center of gravity, To estimate the expected parameters of the model, To evaluate the entropy of the model, is the loss function.

[0050] The beneficial effects of the present invention are as follows: the present invention divides the video in the camera into frames, obtains a static image as the first image, performs grayscale processing and segmentation processing on the first image, generates a two-dimensional grayscale image and a binary image, uses the two-dimensional grayscale image to cover the foreground of the binary image, generates a second image, and extracts the feature vector of the second image to calculate the loss value; adopts the zero-position deviation of the binary image calculation system to simplify the calculation and reduce the difficulty of practical application; adopts a control unit and combines the human-computer interaction of the cloud platform to control the staggered operation of the first timer and the second timer, saves the storage burden and power loss of the first timer and the second timer, and is beneficial to energy conservation and environmental protection; establishes an evaluation model, evaluates the comprehensive quality of the sight by comprehensively evaluating the loss value, key quality parameters and the evaluation center of gravity of the weight vector, and quantifies it into an evaluation score. The larger the evaluation score, the better the quality of the sight. The evaluation score is displayed on the cloud platform, and the quality of the sight can be intuitively seen, which is easy to distinguish and reduces the workload of human eye recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the basic flow of a sight production management system provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments. Example 1

[0053] Reference Figure 1 , as an embodiment of the present invention, provides a sight production management system, including a collection module, a processing module, a quality assessment module and a cloud platform module;

[0054] The acquisition module is used to acquire the video in the sight and transmit the video to the processing module;

[0055] The processing module is used to generate an image set according to a time series of frame divisions of the video, and select one of the frames in the image set as a first image, perform grayscale processing and segmentation processing on the first image to generate a two-dimensional grayscale image and a binary image, the two-dimensional grayscale image performs foreground coverage on the binary image to generate a second image, extract a feature vector of the second image, and calculate a loss value, obtain a contour grayscale curve of the two-dimensional grayscale image, perform adaptive segmentation on the remaining images in the image set according to a first threshold, extract the foreground image of the remaining images in the image set according to the contour grayscale curve, generate a third image, extract the angle feature, distance feature and shape feature of the third image, calculate the loss value of the angle feature, distance feature and shape feature, and transmit the loss value to the quality assessment module;

[0056] The quality assessment module is used to calculate the assessment value and transmit the assessment value to the cloud platform module;

[0057] The cloud platform module is used for zero-position error correction, displaying evaluation values ​​and controlling the processing module.

[0058] The camera is an industrial camera, and the target surface model of the industrial camera is , target surface diagonal 4mm, eyepiece lens optical magnification 0.5X, display size 14 inches, display magnification 44.45X, magnification 133.35X.

[0059] The acquisition module divides the frames through the video in the camera, the processing module obtains the static image as the first image, performs grayscale processing and segmentation processing on the first image, generates a two-dimensional grayscale image and a binary image, uses the two-dimensional grayscale image to cover the foreground of the binary image, generates a second image, and extracts the feature vector of the second image to calculate the loss value; the zero-position deviation of the binary image calculation system is used to simplify the calculation and reduce the difficulty of practical application; the control unit is used in combination with the human-computer interaction of the cloud platform to control the staggered operation of the first timer and the second timer, saving the storage burden and power loss of the first timer and the second timer, which is beneficial to energy conservation and environmental protection; the quality assessment module establishes an assessment model, and the comprehensive loss value, key quality parameters and weight vector assessment center of gravity are used to evaluate the comprehensive quality of the sight, and quantified into an assessment score. The larger the assessment score, the better the quality of the sight. The cloud platform displays the assessment score, which can intuitively see the quality of the sight, is easy to distinguish, and reduces the workload of human eye recognition.

[0060] The preprocessing unit includes a grayscale processing subunit and a segmentation subunit;

[0061] The grayscale processing subunit generates an image set by dividing the time series of video frames, selects one frame of the image set as a first image, performs grayscale processing on the first image to generate a two-dimensional grayscale image, and transmits the grayscale image to the segmentation subunit. The two-dimensional grayscale image occupies less memory and has a faster calculation speed, while visually increasing contrast and highlighting the target area;

[0062] The segmentation subunit segments the foreground image and the background image in the two-dimensional grayscale image based on the separation threshold, defines the pixel points in the two-dimensional grayscale image that are greater than the separation threshold as target information, and records the grayscale value of the target information, extracts the pixel points in the two-dimensional grayscale image that are less than the separation threshold to generate a background image, and records the grayscale value of the background image, binarizes the foreground image and generates a binary image, and covers the binary image with the foreground image of the two-dimensional grayscale image to generate a second image;

[0063] The segmentation subunit obtains the contour grayscale values ​​of the foreground image and the background image, and generates a contour grayscale curve. The segmentation subunit adaptively segments the remaining images in the image set based on the first threshold, extracts the foreground image of the remaining images based on the contour grayscale curve, and generates a third image. The segmentation subunit automatically extracts the foreground image of the remaining images based on the contour line, thereby realizing automatic separation of the foreground image and the background image, eliminating repeated work steps, saving calculation time, transmitting the second image to the feature extraction unit, transmitting the third image to the quality assessment module, and transmitting the binary image to the control unit.

[0064] The grayscale processing subunit converts the first image into a two-dimensional grayscale image based on the RGB component weighting method, and the calculation expression of the RGB component weighting method is:

[0065]

[0066] The expression of the binary image is:

[0067]

[0068] in, is the red channel, For the green channel, is the blue channel, , and are the grayscale conversion parameters, is a binary image, is the two-dimensional grayscale image to be tested, The first threshold to set for segmenting subunits.

[0069] The calculation expression of the adaptive segmentation is:

[0070]

[0071] in, is an image in the image set to be segmented, For the third image, is a first threshold value set in the processing module, is the gradient operation of the images in the image set to be segmented.

[0072] The feature extraction unit integrates the attention mechanism into the convolutional neural network, which includes an input layer, an output layer, a pooling layer, a convolution layer and a fully connected layer. By introducing the attention mechanism, the neural network can automatically learn and selectively focus on important information in the input, thereby improving the performance and generalization ability of the model.

[0073] The convolutional neural network uses a 3*3 convolution kernel. After the convolution layer, the dimension of the feature map output by the convolution layer is reduced by pooling, which effectively reduces the network parameters and prevents overfitting. The pooling layer decomposes the two-dimensional pooling of each pixel in the second image into one-dimensional pooling, and performs feature encoding on the one-dimensional pooling respectively, and generates a feature map. The feature map includes the direction information and position information of each pixel in the second image. The feature map is input into the fully connected layer. The fully connected layer extracts and fuses the feature information in the feature map based on the classifier, and generates a fused feature vector, and its calculation expression is:

[0074]

[0075] in, is the fusion feature vector, is the feature map, are learnable weights, is the learning rate of the convolutional neural network. The convolutional neural network uses the fused feature vector as a data set, sets the number of training times and performs comparisons, constructs a loss function, calculates the loss value, and transmits the loss value to the quality assessment module. The fused feature vector includes angle, distance, and shape. The calculation expression of the loss function is:

[0076]

[0077]

[0078] in, is the value of the loss function, is the intersection-and-union ratio, is the dynamic weight, is the distance loss, is the shape loss, is a constant, and the feature vectors of different dimensions are fused to increase the accuracy of the loss value. When the real box and the predicted box of the convolutional neural network are closer, the influence of the distance loss is smaller, and the influence of the shape loss is greater, forming a dynamic process. The dynamic weight is used to adjust the change of the distance loss to improve the stability of the distance loss.

[0079] The detection unit includes a first timer and a second timer, and the control unit is electrically connected to the cloud platform module, and the control unit is used to control the staggered operation of the first timer and the second timer;

[0080] When the control unit receives a detection instruction from the cloud platform module, the control unit sends a start signal to the first timer, the first timer performs pixel scanning on the pixels of the binary image, obtains the coordinates of the central pixel of the binary image, calculates the zero position deviation m of the pixels of the binary image based on the position of the a priori calibration point pixel, transmits the zero position deviation m to the control unit, and when the zero position deviation m exceeds the zero position threshold, the first timer outputs a zero position deviation excessive signal to the cloud platform module;

[0081] The detection unit generates a zero position threshold based on the position of the a priori calibration point pixel, the first timer is electrically connected to the control unit, and the second timer is electrically connected to the cloud platform module and the control unit.

[0082] A control unit is used in combination with human-computer interaction of a cloud platform to control the staggered operation of the first timer and the second timer, thereby saving the storage burden and power consumption of the first timer and the second timer, and being beneficial to energy conservation and environmental protection.

[0083] The cloud platform module includes a display screen and a control button, wherein the display screen is used to display an excessive deviation signal and an evaluation value, and the control button is used to manually reduce the zero deviation m, and send the reduced zero deviation threshold to the control unit, and the control unit sends a start signal to the second timer when the zero deviation is less than the zero threshold, and the second timer performs pixel scanning on the third image, obtains the central pixel coordinates of the third image, obtains the neighborhood image of the central pixel coordinates, calculates the zero deviation n of the neighborhood image of the central pixel coordinates, and transmits the zero deviation n to the control unit;

[0084] The control unit calculates the difference between the zero deviation m and the zero deviation n, and starts the first timer based on a threshold of the difference between the zero deviation m and the zero deviation n. When the difference between the zero deviation m and the zero deviation n is greater than the threshold of the difference between the zero deviation m and the zero deviation n, the control unit starts the first timer, and the first timer performs a weighted update on the zero deviation m and sets the weighted updated zero deviation m as the first timer. Transmitting to the cloud platform module;

[0085] The calculation expression of the weighted update is:

[0086]

[0087]

[0088] in, is the weighted updated value of the zero-position deviation m, is the value of the zero position deviation m, is the value of the zero deviation n, The difference between zero deviation m and zero deviation n

[0089] The control unit and the cloud platform module are interconnected, and the cloud platform receives the deviation signal of the control unit. The staff makes a rough adjustment to the zero-position deviation based on the deviation signal, thereby saving the calculation steps of the computer and speeding up the detection work efficiency.

[0090] The cloud platform includes control buttons for manual coarse adjustment of zero position deviation, which reduces the system's cyclic calculation volume, saves system storage space, and increases the system's service life.

[0091] The quality assessment module includes a calculation unit and an assessment model;

[0092] The calculation unit extracts key quality parameters of the third image, calculates a turning threshold of the key quality parameters of the third image by using a neural network, establishes a clouding interval based on the turning threshold, and transmits the clouding interval to the evaluation model, wherein the key quality parameters of the third image include an eye relief, an exit pupil diameter, a magnification, and a parallax angle;

[0093] The evaluation model uses fuzzy hierarchical analysis method to calculate the weight vector of the key quality parameters of the third image, establishes the quality level of the key quality parameters of the third image based on the weight vector and the cloudification interval, and calculates the evaluation value based on the quality level, and the quality level includes excellent, qualified and unqualified.

[0094] The evaluation model constructs a judgment matrix based on the fuzzy analytic hierarchy process, calculates a consistency matrix based on the judgment matrix, and calculates a weight vector of the key quality parameters of the third image based on the consistency matrix;

[0095] The calculation expression of the judgment matrix is:

[0096]

[0097] The calculation expression of the consistency matrix is:

[0098]

[0099] The calculation expression of the weight vector of the key quality parameter of the third image is:

[0100]

[0101] in, is the judgment matrix, is the consistency matrix, is the key quality parameter in the third image Key quality parameters in the third image The comparison value of is the key quality parameter in the third image Key quality parameters in the third image The reciprocal of the comparison value, is the weight vector, is the new matrix obtained after n iterations of the key quality parameters in the third image, is the key quality parameter in the third image go through Iterates and integrates the key quality parameters in the third image go through The new matrix is ​​obtained by the iteration, and the quality assessment module establishes an assessment model. The comprehensive quality of the sight is evaluated by the comprehensive loss value, key quality parameters and assessment center of gravity of the weight vector, and quantified into an assessment score. The larger the assessment score, the better the quality of the sight.

[0102] The neural network is the third key parameter of image quality and Normalization is performed to make the third image quality key parameter and Distributed between 0 and 1, set the initial weight vector of competition , topological neighborhood radius and the initial learning rate , based on the weight vector and the key quality parameter of the third image and Determine the winning neuron, obtain the adjacent interval image centered on the winning neuron, and update the adjacent interval image topology and learning rate centered on the winning neuron, based on the quality key parameters of the third image and The convergence of the number of iterations is calculated, the eye distance turning threshold a, the exit pupil diameter turning threshold b, the magnification turning threshold c and the parallax angle turning threshold d are calculated, and clouded intervals a~b, b~c and c~d are generated. The clouded intervals are transmitted to the evaluation model, and the cloud platform displays the evaluation scores, which is convenient for the staff to intuitively see the quality of the sight and easy to distinguish, reducing the workload of human eye recognition.

[0103] The video in the camera is framed to obtain a static image as the first image, the first image is gray-processed and segmented to generate a two-dimensional gray-scale image and a binary image, the two-dimensional gray-scale image is used to cover the foreground of the binary image to generate a second image, and the feature vector of the second image is extracted to calculate the loss value; the zero-position deviation of the binary image calculation system is used to simplify the calculation and reduce the difficulty of practical application; a control unit is used in combination with the human-computer interaction of the cloud platform to control the staggered operation of the first timer and the second timer, saving the storage burden and power loss of the first timer and the second timer, which is beneficial to energy conservation and environmental protection; an evaluation model is established to evaluate the comprehensive quality of the sight based on the evaluation center of gravity of the comprehensive loss value, key quality parameters and weight vector, and quantified into an evaluation score. The larger the evaluation score, the better the quality of the sight. The evaluation score is displayed on the cloud platform, so the quality of the sight can be intuitively seen, which is easy to distinguish and reduces the workload of human eye recognition. Example 2

[0104] This is another embodiment of the present invention. Different from the first embodiment, this embodiment provides a sight production management system. In order to verify and illustrate the technical effects adopted in this method, this embodiment adopts a traditional technical solution and compares it with the method of the present invention. The test results are compared by means of scientific demonstration to verify the real effect of this method.

[0105] Based on the historical production quality data of the scope, it is shown that the scope will have the following quality levels when it is produced: excellent level, qualified level and unqualified level; based on this, this paper has produced excellent level type, qualified level type and unqualified level type as data sets;

[0106] Based on neural network and fuzzy analysis, an evaluation model for the riflescope is established and the evaluation value is calculated. Tests are conducted to verify the effectiveness of this method.

[0107] The neural network is the third key parameter of image quality and Normalization is performed to make the third image quality key parameter and Distributed between 0 and 1, set the initial weight vector of competition , topological neighborhood radius and the initial learning rate , based on the weight vector and the key quality parameter of the third image and Determine the winning neuron, obtain the adjacent interval image centered on the winning neuron, and update the adjacent interval image topology and learning rate centered on the winning neuron, based on the quality key parameters of the third image and The convergence of the number of iterations is calculated, the eye distance turning threshold a, the exit pupil diameter turning threshold b, the magnification turning threshold c and the parallax angle turning threshold d are calculated, clouding intervals a~b, b~c and c~d are generated, and the clouding intervals are transmitted to the evaluation model.

[0108] The evaluation model is based on the cloudification interval and the key quality parameters of the third image and The weight vector establishes the excellent level p, qualified level q and unqualified level s, and the key quality parameters of the third image and The weight vector is weighted and integrated to calculate the key quality parameters of the third image and The evaluation center of gravity of the weight vector is calculated based on the evaluation center of gravity and the loss value, and the evaluation value is transmitted to the cloud platform module;

[0109] The calculation expression of the evaluation value is:

[0110]

[0111]

[0112] in, is the evaluation value, To evaluate the center of gravity, To estimate the expected parameters of the model, To evaluate the entropy of the model, is the loss function

[0113] In the experiment, 500 sets of data were collected for each quality level of the sight, and the training set and the test set were set at a ratio of 8:2; based on the comparison between the output results and the actual results, the data were obtained: the recognition accuracy of the excellent grade was the highest, reaching 100%; the recognition rate of the unqualified grade was 95%; the recognition rate of the qualified grade reached 96%; the average recognition accuracy of the above three quality levels of the sight was 97%, and the larger the evaluation score, the better the quality of the sight.

[0114] It should be appreciated that embodiments of the present invention may be implemented or implemented by a combination of computer hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program may be implemented in assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed dedicated integrated circuit for this purpose.

[0115] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A sight production management system, characterized in that: It includes acquisition module, processing module, quality assessment module and cloud platform module; The acquisition module is used to acquire the video in the sight and transmit the video to the processing module; The processing module is used to generate an image set according to a time series of frame divisions of the video, and select one of the frames in the image set as a first image, perform grayscale processing and segmentation processing on the first image to generate a two-dimensional grayscale image and a binary image, the two-dimensional grayscale image performs foreground coverage on the binary image to generate a second image, extract a feature vector of the second image, and calculate a loss value, obtain a contour grayscale curve of the two-dimensional grayscale image, perform adaptive segmentation on the remaining images in the image set according to a first threshold, extract the foreground image of the remaining images in the image set according to the contour grayscale curve, generate a third image, extract the angle feature, distance feature and shape feature of the third image, calculate the loss value of the angle feature, distance feature and shape feature, and transmit the loss value to the quality assessment module; The quality assessment module is used to calculate the assessment value and transmit the assessment value to the cloud platform module; The cloud platform module is used for zero error correction, displaying evaluation values ​​and controlling the processing module; Evaluation model based on cloudification interval and key quality parameters of the third image and The weight vector establishes the excellent level p, qualified level q and unqualified level s, and the key quality parameters of the third image and The weight vector is weighted and integrated to calculate the key quality parameters of the third image and The evaluation center of gravity of the weight vector is calculated based on the evaluation center of gravity and the loss value, and the evaluation value is transmitted to the cloud platform module; The calculation expression of the evaluation value is: ； ； in, is the evaluation value, To evaluate the center of gravity, To estimate the expected parameters of the model, To evaluate the entropy of the model, is the loss function; The quality assessment module includes a calculation unit and an assessment model; The calculation unit extracts key quality parameters of the third image, calculates a turning threshold of the key quality parameters of the third image by using a neural network, establishes a clouding interval based on the turning threshold, and transmits the clouding interval to the evaluation model, wherein the key quality parameters of the third image include an eye relief, an exit pupil diameter, a magnification, and a parallax angle; The evaluation model uses fuzzy hierarchical analysis method to calculate the weight vector of the key quality parameters of the third image, establishes the quality level of the key quality parameters of the third image based on the weight vector and the cloudification interval, and calculates the evaluation value based on the quality level, and the quality level includes excellent, qualified and unqualified.

2. The sight production management system as claimed in claim 1, characterized in that: The preprocessing unit includes a grayscale processing subunit and a segmentation subunit; The grayscale processing subunit generates an image set by dividing the time series of video frames, selects one frame of the image set as a first image, performs grayscale processing on the first image to generate a two-dimensional grayscale image, and transmits the grayscale image to the segmentation subunit; The segmentation subunit segments the foreground image and the background image in the two-dimensional grayscale image based on the separation threshold, defines the pixel points in the two-dimensional grayscale image that are greater than the separation threshold as target information, and records the grayscale value of the target information, extracts the pixel points in the two-dimensional grayscale image that are less than the separation threshold to generate a background image, and records the grayscale value of the background image, binarizes the foreground image and generates a binary image, and covers the binary image with the foreground image of the two-dimensional grayscale image to generate a second image; The segmentation subunit obtains the contour grayscale values ​​of the foreground image and the background image and generates a contour grayscale curve. The segmentation subunit adaptively segments the remaining images in the image set based on the first threshold, extracts the foreground image of the remaining images in the image set based on the contour grayscale curve, generates a third image, transmits the second image to the feature extraction unit, transmits the third image to the quality assessment module, and transmits the binary image to the control unit.

3. The sight production management system as claimed in claim 2, characterized in that: The calculation expression of the adaptive segmentation is: ； in, is an image in the image set to be segmented, For the third image, is a first threshold value set in the processing module, is the gradient operation of the images in the image set to be segmented.

4. The sight production management system as claimed in claim 2, characterized in that: The feature extraction unit integrates the attention mechanism into a convolutional neural network, which includes an input layer, an output layer, a pooling layer, a convolution layer, and a fully connected layer; The pooling layer decomposes the two-dimensional pooling of each pixel in the second image into one-dimensional pooling, performs feature encoding on the one-dimensional pooling, and generates a feature map, the feature map includes the direction information and position information of each pixel in the second image, and inputs the feature map into the fully connected layer, the fully connected layer extracts and fuses the feature information in the feature map based on the classifier, and generates a fused feature vector, the calculation expression of which is: ； in, is the fusion feature vector, is the feature map, are learnable weights, is the learning rate of the convolutional neural network. The convolutional neural network uses the fused feature vector as a data set, sets the number of training times and performs comparisons, constructs a loss function, calculates the loss value, and transmits the loss value to the quality assessment module. The fused feature vector includes angle, distance, and shape. The calculation expression of the loss function is: ； ； in, is the value of the loss function, is the intersection-and-union ratio, is the dynamic weight, is the distance loss, is the shape loss, is a constant.

5. The sight production management system as claimed in claim 1, characterized in that: The detection unit includes a first timer and a second timer, and the control unit is electrically connected to the cloud platform module, and the control unit is used to control the staggered operation of the first timer and the second timer; When the control unit receives a detection instruction from the cloud platform module, the control unit sends a start signal to the first timer, the first timer performs pixel scanning on the pixels of the binary image, obtains the coordinates of the central pixel of the binary image, calculates the zero position deviation m of the pixels of the binary image based on the position of the a priori calibration point pixel, transmits the zero position deviation m to the control unit, and when the zero position deviation m exceeds the zero position threshold, the first timer outputs a zero position deviation excessive signal to the cloud platform module; The detection unit generates a zero position threshold based on the position of the a priori calibration point pixel, the first timer is electrically connected to the control unit, and the second timer is electrically connected to the cloud platform module and the control unit.

6. The sight production management system as claimed in claim 5, characterized in that: The cloud platform module includes a display screen and a control button, the display screen is used to display an excessive deviation signal and an evaluation value, the control button is used to manually reduce the zero deviation m, and send the reduced zero deviation value to the control unit, the control unit sends a start signal to the second timer when the zero deviation is less than the zero threshold, the second timer performs pixel scanning on the third image, obtains the central pixel coordinates of the third image, obtains the neighborhood image of the central pixel coordinates, calculates the zero deviation n of the neighborhood image of the central pixel coordinates, and transmits the zero deviation n to the control unit; The control unit calculates the difference between the zero deviation m and the zero deviation n, and starts the first timer based on a threshold of the difference between the zero deviation m and the zero deviation n. When the difference between the zero deviation m and the zero deviation n is greater than the threshold of the difference between the zero deviation m and the zero deviation n, the control unit starts the first timer, and the first timer performs a weighted update on the zero deviation m and sets the weighted updated zero deviation m as the first timer. Transmitting to the cloud platform module; The calculation expression of the weighted update is: ； ； in, is the weighted updated value of the zero-position deviation m, is the value of the zero position deviation m, is the value of the zero deviation n, is the difference between the zero position deviation m and the zero position deviation n.

7. The sight production management system as claimed in claim 1, characterized in that: The evaluation model constructs a judgment matrix based on the fuzzy analytic hierarchy process, calculates a consistency matrix based on the judgment matrix, and calculates a weight vector of the key quality parameters of the third image based on the consistency matrix; The calculation expression of the judgment matrix is: ； Where i, j = 1, 2, ..., n; The calculation expression of the consistency matrix is: ； The calculation expression of the weight vector of the key quality parameter of the third image is: ； in, is the judgment matrix, is the consistency matrix, is the key quality parameter in the third image Key quality parameters in the third image The comparison value of is the key quality parameter in the third image Key quality parameters in the third image The reciprocal of the comparison value, is the weight vector, is the new matrix obtained after n iterations of the key quality parameters in the third image, is the key quality parameter in the third image go through Iterations and integration of key quality parameters in the third image go through The new matrix obtained by iteration.

8. The sight production management system as claimed in claim 1, characterized in that: The neural network is the third key parameter of image quality and Normalization is performed to make the third image quality key parameter and Distributed between 0 and 1, set the initial weight vector of competition , topological neighborhood radius and the initial learning rate , based on the weight vector and the key quality parameter of the third image and Determine the winning neuron, obtain the adjacent interval image centered on the winning neuron, and update the adjacent interval image topology and learning rate centered on the winning neuron, based on the quality key parameters of the third image and The convergence of the number of iterations is calculated, the eye distance turning threshold a, the exit pupil diameter turning threshold b, the magnification turning threshold c and the parallax angle turning threshold d are calculated, clouding intervals a~b, b~c and c~d are generated, and the clouding intervals are transmitted to the evaluation model.

Images