Image data acquisition quality control method and device and storage medium

Abstract

The invention relates to the field of image processing, in particular to an image data collection quality control method and device and a storage medium, and the method comprises the steps: screening and filtering collected pictures through preset detection targets including a background standard, a collection frame standard, a shooting standard, a light source standard, a scale standard and the like, and obtaining a detection result; the consistency and the accuracy of the stored pictures in all aspects are ensured, so that the acquisition quality of the picture data is improved, and bad picture data is prevented from entering the database. Meanwhile, optimization and perfection of the feature set and the sample library are ensured. And the target object in the picture is identified and classified through the feature classification model, that is, the extracted features are analyzed and modeled by using an intelligent algorithm, so that accurate identification of the target object in the picture is realized, new features and categories are screened out, and a feature set and a sample library are further enriched.

Description

Technical Field

[0001] The present application relates to the field of image processing, and in particular to a method, device and storage medium for image data acquisition quality control. Background Art

[0002] With the advancement of the Industrialization 2025 policy, the combination of industrial production with informatization and intelligence has become an important way to improve production capacity and quality. In the intelligent era, the application of artificial intelligence methods such as machine vision to reduce manual operation and intervention is an inevitable trend in the development of the industry. Especially in the field of object identification, the use of artificial intelligence to accurately identify objects is equivalent to equipping the machine with "eyes", which is a key link in realizing intelligent industrialization.

[0003] However, in the process of informationization of objects, the collection of visual sample images faces many challenges. The lack of high-quality object shooting sampling methods and detection standards has led to uneven quality of databases and sample images. Traditional collection methods mainly rely on manual work, which is easily affected by factors such as the technical level and work attitude of the collectors, resulting in low quality of sample data, most of which are dirty data and difficult to use effectively. These problems have seriously hindered the establishment and application of high-precision image recognition algorithms. Specifically, the problems existing in the existing technology in image acquisition include: the lack of unified acquisition standards, resulting in differences in background, light, shooting angle, etc. in the collected images, affecting the subsequent image recognition effect; there is no effective quality detection algorithm to screen the collected images, resulting in bad image data entering the database, reducing the accuracy of the recognition algorithm; the existing feature extraction and recognition algorithms have problems such as low recognition accuracy and weak generalization ability when dealing with complex scenes and diversified objects.

[0004] Therefore, there is an urgent need for a method that can solve the above problems to improve the quality of data collection, optimize and improve the feature set and sample library, and thus promote the development of intelligent industrialization. Summary of the invention

[0005] The present application provides a method for controlling the acquisition quality of image data. By formulating a standard image acquisition method, establishing an image quality detection algorithm and a recognition model, accurate recognition and classification of targets in images can be achieved.

[0006] In order to achieve the above object, the present invention adopts the following technical scheme:

[0007] In a first aspect, the present invention provides a method for controlling the acquisition quality of image data, comprising:

[0008] Collect image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to the sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value;

[0009] Establish a feature set using the features of the images in the sample library, extract the features of the images in the sample library based on standard images of various features, and calculate the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features;

[0010] Establishing a feature classification model corresponding to each feature according to each feature, and using the feature classification model to identify and classify the target object in the image;

[0011] According to the recognition results of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

[0012] In a preferred example of the present application, it can be further configured to include:

[0013] The Tenengrad gradient method is used to detect whether the clarity of the target object reaches the target value.

[0014] In a preferred example of the present application, it can be further configured to use a gray level co-occurrence matrix, a histogram of oriented gradients, a local binary pattern (LBP) or a Gaussian filter to extract texture features of the images in the sample library.

[0015] In a preferred example of the present application, it can be further configured that the matching degree between images is calculated using a similarity measurement criterion, including:

[0016] The matching degree between images is calculated using the phase correlation method, Hausdorff distance function, Euclidean distance discriminant method or normalized product correlation method.

[0017] In a preferred example of the present application, it can be further configured that the feature classification model is established based on a naive Bayes model.

[0018] In a preferred example of the present application, it can be further configured that the image is identified according to the preset detection target, and the image that meets the preset detection target is screened out, including:

[0019] Standardize the input image to make it meet the requirements of the SVM algorithm;

[0020] Use kernel functions to map the input space to a high-dimensional space;

[0021] According to the loss function, use the gradient descent algorithm to find the optimal category boundary;

[0022] The position of the category boundary is determined by the support vector, the data set is divided into different categories, and images that meet the preset detection target are screened out.

[0023] In a second aspect, the present application provides a device for controlling the acquisition quality of image data, the device comprising:

[0024] A data acquisition module is used to collect image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to a sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value;

[0025] A feature extraction module, used to establish a feature set using the features of the images in the sample library, extract the features of the images in the sample library based on standard images of various features, and calculate the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features;

[0026] A classification module is used to establish a feature classification model corresponding to each feature according to the features, and use the feature classification model to identify and classify the target objects in the image; according to the recognition results of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

[0027] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the image data acquisition quality control method as described in any one of the above items are implemented.

[0028] In a fourth aspect, the present application provides a computer-readable storage medium having a program stored thereon, wherein when the program is executed by a processor, the image data acquisition quality control method as described in any one of the above items is implemented.

[0029] In a fifth aspect, the present application provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the image data acquisition quality control method as described in any one of the above items.

[0030] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:

[0031] This application screens and filters the collected images through preset detection targets, including background standards, collection frame standards, shooting standards, light source standards, and ruler standards, etc., to ensure that the stored images maintain consistency and accuracy in all aspects, thereby improving the collection quality of image data and limiting the entry of bad image data into the database. At the same time, it also ensures the optimization and improvement of the feature set and sample library. In addition, the target objects in the image are identified and classified through the feature classification model, that is, the extracted features are analyzed and modeled using intelligent algorithms, which not only realizes the accurate identification of the target objects in the image, but also screens out new features and categories, further enriching the feature set and sample library. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flowchart of a method for controlling the acquisition quality of image data provided by one embodiment of the present application.

[0033] Figure 2 A structural diagram of an image data acquisition quality control device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] In one embodiment of the present application, a method for controlling the quality of image data acquisition is provided. Figure 1 As shown, the method includes:

[0036] S100: Collect image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to a sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value;

[0037] Specifically, when detecting the background of an image, the specific method is to first detect pixels in the background area and calculate its color consistency. The calculation method is:

[0038]

[0039] Where U is the color consistency.

[0040] Set the threshold ε u , when U<ε uWhen U>ε, the background meets the requirements and can be uploaded to the database. u When the background does not meet the requirements, it cannot be uploaded to the library.

[0041] Then the color difference of the background area is calculated. Specifically, a background color plate is set, and the background color plate of the sample image is used as a template to calculate the color difference with the background plate of the sample image actually taken. The conversion from RGB to Lab is to convert the RGB color space into the Lab color space, where the Lab color space is a color model that is more uniform in human eye perception. The conversion process includes the following steps.

[0042] First, convert the RGB color from the RGB color space to the XYZ color space using the following matrix multiplication:

[0043] [X][0.412453 0.357580 0.180423][R];

[0044] [Y]=[0.212671 0.715160 0.072169]*[G];

[0045] [Z][0.019334 0.119193 0.950227][B];

[0046] This matrix multiplication converts RGB colors to the corresponding XYZ values ​​in the CIE XYZ color space and converts XYZ colors from the XYZ color space to the Lab color space. It can be calculated using the following formula:

[0047] X=X / Xn, Y=Y / Yn, Z=Z / Zn;

[0048] fx=f(X), fy=f(Y), fz=f(Z);

[0049] L=116*fy-16, a=500*(fx-fy), b=200*(fy-fz);

[0050] Among them, Xn, Yn and Zn are the standard values ​​of the reference white point, and the D65 white point is commonly used. The final Lab value is the representation of the RGB color in the Lab color space.

[0051] Next, calculate the LAB color difference. The Lab color scale is marked as follows: the L (brightness) axis represents black and white, 0 is black, and 100 is white; the A (red-green) axis has positive values ​​of red, negative values ​​of green, and 0 is neutral color; the B (yellow-blue) axis has positive values ​​of yellow, negative values ​​of blue, and 0 is neutral color. All colors can be perceived and measured by any Lab scale. These scales can also be used to represent the color difference between the standard and the test sample, and usually have Δ as an identifier. Among them, if ΔL is positive, it means that the sample is lighter than the standard sample, and if ΔL is negative, it means that the sample is darker than the standard sample. If Δa is positive, it means that the sample is redder (or less green) than the standard sample, and if it is negative, it means that the sample is greener (or less red). If Δb is positive, it means that the sample is yellower (or less blue) than the standard sample, and if it is negative, it means that the sample is bluer (or less yellow).

[0052] The color difference between L, a, and b can also be expressed by a separate color difference symbol ΔE. ΔE is defined as the total color difference of the sample, but it cannot indicate the offset direction of the color difference of the sample. The larger the ΔE value, the greater the color difference. It is calculated by the following formula: ΔE = [(ΔL*)2 + (Δa*)2 + (Δb*)2]1 / 2. Generally, ΔE can be distinguished visually when it is 1.5. Color difference C = ΔE.

[0053] Finally, set the threshold ε c , when C<ε c When C>ε, the background meets the requirements and can be uploaded to the library. c When the background does not meet the requirements, it cannot be uploaded to the library.

[0054] The specific process of object detection for lighting and shadows includes:

[0055] First: Grayscale conversion - the collected color image refers to the image in the RGB color space, and the three channels of such an image are R (red) G (green) B (blue). Grayscale images refer to single-channel images. Common methods for converting three-channel images to single-channel images include YUV method and average method. According to the sensitivity of the human eye to red, green and blue, the following methods can be used for conversion:

[0056] G=R*0.299+G*0.587+B*0.114;

[0057] Second: Histogram Generation and Analysis

[0058] The basic method of generating an image histogram is to generate it by counting the distribution of each pixel value in the image. A histogram is a statistical table of the distribution of image pixels, where the horizontal axis represents the pixel value and the vertical axis represents the frequency or number of occurrences of the pixel value. The process of generating a histogram mainly includes the following steps:

[0059] 1. Initialize the histogram vector: Create a vector of length 256 (for 8-bit grayscale images) and initialize all values ​​to 0. This vector is used to record the number of times each gray level appears.

[0060] 2. Traverse the image pixels: Traverse each pixel in the image one by one, and increase the count of the corresponding gray level in the histogram vector according to the pixel value.

[0061] Third: Normalized histogram: Divide each value in the histogram vector by the total number of pixels in the image to obtain the normalized histogram, that is, the probability of each gray level appearing.

[0062] The basic principle of calculating the spot area by image histogram is to determine the spot area by analyzing the brightness distribution of the image. The histogram reflects the number of pixels at each brightness level in the image. By analyzing the histogram, areas with higher brightness can be identified. These areas usually correspond to the spot. The specific steps are as follows:

[0063] 1. Calculate the histogram of the image: count the number of pixels at each brightness level in the image;

[0064] 2. Analyze the histogram: Use the histogram to identify areas with higher brightness, which usually correspond to light spots. You can distinguish between light spots and shadow areas (for example, set the grayscale value to 230) by setting a threshold (for example, setting the grayscale value to 30).

[0065] 3. Calculate the spot area: Based on the results of the histogram analysis, use image processing technology (such as threshold segmentation, connected region analysis, etc.) to determine the specific position and area A of the spot and the specific position and area S of the shadow.

[0066] 4. Spot area setting threshold: ε g , when Ag<ε g If it meets the requirements, it can be uploaded to the warehouse. Otherwise, it cannot be uploaded to the warehouse if it does not meet the requirements.

[0067] 5. Shadow area threshold setting: When As<ε s If it meets the requirements, it can be uploaded to the warehouse. Otherwise, it cannot be uploaded to the warehouse if it does not meet the requirements.

[0068] The specific process of locating the target object includes:

[0069] 1. Detection area cropping

[0070] The shooting area is divided into upper and lower areas, left and right areas according to the rectangular frame. It is cut into top and bottom edge detection areas, left and right edge detection areas, and the rectangular frame is cut into four parts according to the position coordinates: top, bottom, left, and right.

[0071] 2. Edge detection-Hough line detection

[0072] First, use Hough to detect the straight lines on each side of the target. The Hough transform uses polar coordinates to represent straight lines, and each point in the image represents a straight line passing through the point. A straight line can be represented by two parameters: r represents the distance from the origin, and θ represents the angle with the x-axis. For each point, all possible straight line parameters are traversed and accumulated in the Hough space. For each point, the value of the corresponding position in the accumulator is added by 1 for all the straight lines passing through the point. This accumulation process can be performed in a two-dimensional array in the Hough space. When a value in the accumulator exceeds a preset threshold, the straight line is considered to exist. This threshold can be set according to application requirements. According to the peak value in the accumulator, the corresponding straight line parameters (r and θ) are inversely calculated, and the detected straight line is drawn in the image. In each partial area, find the top straight line L1 (top detection area), the bottom straight line L2 (bottom detection area), the leftmost straight line L3 (left detection area), and the rightmost straight line L4 (right detection area), that is, L1 = min(y), y is the vertical coordinate, min is the minimum value, L2 = max(y), y is the vertical coordinate, and max is the maximum value. Find the equation of the line from the two endpoints of L2. The coordinates of the two points are: (x1, y1) (x2, y2). The equation of the line is (x-x1) / (x2-x1) = (y-y1) / (y2-y1). If L4 is parallel to the bottom straight line, the equation is y = m, and the distance between the bottom straight lines D = |L2-l2| is calculated. Set the threshold D<ε d , it meets the requirements and can be uploaded to the database. Otherwise, it does not meet the requirements and cannot be uploaded to the database. If the deviation angle is calculated in parallel, the angle between the bottom edge detection line and the bottom edge detection line can be calculated according to the equation to obtain the inclination angle θ. Set the threshold D<ε d , if it meets the requirements, it can be uploaded to the database. Otherwise, if it does not meet the requirements, it cannot be uploaded to the database. Set the threshold θ<ε θ If it meets the requirements, it can be uploaded to the warehouse. Otherwise, it cannot be uploaded to the warehouse if it does not meet the requirements.

[0073] The specific process of deformation detection includes:

[0074] The deformation of the left and right sides is mainly caused by the shooting angle. The vertical line is deformed into a diagonal line in the image. It is necessary to detect the angle between the left line and L3, and the right line and L4, and set the threshold. That is, the angle between L3 and the left detection line is α<ε α , the angle between L4 and the left detection line is β<ε βα .

[0075] S200: Establishing a feature set using the features of the images in the sample library, extracting features of the images in the sample library based on standard images of various features, and calculating the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features;

[0076] Specifically, a feature set Ti = {t1, t2, t3 ... ti} is constructed using various features of the sample, for example: t1 is color, t2 is contour, and t3 is texture.

[0077] According to the standard collection method, the standard template library is obtained through standard detection (one standard template image is set for each category in the existing categories), and the initialization features are obtained by extracting standard image features, such as color, contour, texture, blob, sift, hash, etc. Among them, the standard detection is to collect standard collection templates (after obtaining a certain amount of standard images for each category), establish a picture template library, and detect standard values ​​from the templates.

[0078] The extraction of various features includes color feature value extraction, contour extraction, texture extraction, Blob feature and Sift feature extraction.

[0079] The steps of color feature value extraction include:

[0080] 1. Use the image flood filling segmentation algorithm to remove the background and obtain the target object.

[0081] First: Select the seed point: Select a starting point in the image as the seed point. That is, randomly generate several seed points in the background area.

[0082] Then: Determine the neighboring pixels: Take the seed point as the center and determine the difference between the pixel values ​​of its 4-neighborhood or 8-neighborhood and the pixel value of the seed point. If the difference is less than a certain threshold, the pixel point is added to the region.

[0083] Then expand the area: use the newly added pixel as a new seed point and repeat the above steps until no new pixel is added to the area.

[0084] The obtained area is the background area. The background area is masked and the remaining part is the target area.

[0085] 2. Cluster the unmasked area and calculate the number of color types contained in the target object. The core idea of ​​the K-means algorithm is to assign data points to K clusters through an iterative process so that the sum of the distances between each data point and the center point (mean) of its cluster is minimized. The steps of the algorithm are as follows: randomly select K data points as the initial cluster centers; for each data point, calculate its distance to each cluster center and assign it to the cluster represented by the nearest cluster center; recalculate the average value of the data points in each cluster as the new cluster center; repeat the above two steps until the stopping condition is met, such as the change in the cluster center is less than a certain threshold or the preset number of iterations is reached.

[0086] The steps of extracting contour feature values ​​include:

[0087] 1. Convert to grayscale

[0088] 2. Apply threshold or canny edge detection to extract contours, for example: canny edge detection

[0089] 3. Find contours

[0090] Finding contours is an important task in image processing and is often used to extract object boundaries in an image. Contours are detected from binary images and the returned contours can be used for further analysis or visualization.

[0091] The steps for texture extraction include:

[0092] In the field of image processing, texture refers to a visual feature of visible details and structures in an image. Texture feature extraction is one of the important tasks in image analysis, and it can provide effective information about the local area of ​​the image. In this embodiment, four commonly used texture feature extraction methods are compared: gray level co-occurrence matrix (GLCM), histogram of oriented gradients (HOG), local binary pattern (LBP) and Gaussian filter.

[0093] The process of Blob feature extraction includes:

[0094] Blob feature extraction is an image processing technique used to identify and extract regions with specific shapes, sizes, and textures from images. Blob feature extraction is achieved by analyzing connected domains in images, which are composed of pixels with similar properties. Blob analysis can separate target objects in an image and calculate their characteristics such as position, shape, size, and orientation.

[0095] The process of Sift feature extraction includes:

[0096] Extraction steps - scale space extreme value detection: By Gaussian blurring images of different scales, the local minimum and maximum values ​​in the image are detected to form a scale space.

[0097] Key point positioning: In the scale space, the exact position of the key point is found by fitting the local area of ​​the extreme point.

[0098] Keypoint direction assignment: Assign a main direction to each keypoint so that the descriptor is rotation invariant.

[0099] Descriptor generation: Calculate local gradients in the area around the key points to generate feature descriptors that are scale and rotation invariant.

[0100] S300: establishing a feature classification model corresponding to each feature according to the features, and using the feature classification model to identify and classify the target object in the image;

[0101] S400: According to the recognition result of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

[0102] In this embodiment, the collected images are screened and filtered through preset detection targets, including background standards, collection frame standards, shooting standards, light source standards, and ruler standards, ensuring that the stored images maintain consistency and accuracy in all aspects, thereby improving the collection quality of image data and limiting the entry of bad image data into the database. At the same time, it also ensures the optimization and improvement of the feature set and sample library. In addition, the target objects in the image are identified and classified through the feature classification model, that is, the extracted features are analyzed and modeled using intelligent algorithms, which not only realizes the accurate identification of the target objects in the image, but also screens out new features and categories, further enriching the feature set and sample library.

[0103] In some embodiments, including:

[0104] The Tenengrad gradient method is used to detect whether the clarity of the target object reaches the target value.

[0105] In specific implementation, the Tenengrad gradient method is an algorithm based on image edge information. It evaluates the clarity of an image by calculating the gradient change of the edge in the image. In specific implementation, the Sobel operator is first used to detect the edge of the image, and then the gradient amplitude of the detected edge is calculated, and these amplitudes are statistically analyzed to obtain a measure representing the clarity of the image. Usually, the sum or average of the edge gradient amplitudes can be used as a clarity evaluation index. The larger the value, the clearer the image. The definition of calculating image clarity is as follows:

[0106] D(f)=∑ y ∑ x |G(x,y)| (G(x,y)>T);

[0107] The form of G(x, y) is as follows:

[0108]

[0109] Where T is the given edge detection threshold, Gx and Gy are the convolution of the Sobel horizontal and vertical edge detection operators at the pixel point (x, y), respectively, and the Sobel operator template is used to detect the edge and calculate Q = D(f).

[0110] Set the threshold Q < ε q If it meets the requirements, it can be uploaded to the warehouse. Otherwise, it cannot be uploaded to the warehouse if it does not meet the requirements.

[0111] In this embodiment, the accuracy of clarity recognition is improved, thereby improving the accuracy of image quality screening.

[0112] In some embodiments, the similarity metric is used to calculate the matching degree between images, including:

[0113] The matching degree between images is calculated using the phase correlation method, Hausdorff distance function, Euclidean distance discriminant method or normalized product correlation method.

[0114] In some embodiments, the feature classification model is established based on a Naive Bayes model.

[0115] In specific implementation, the Naive Bayes model is a classification method based on the Bayesian theorem and the conditional independence assumption of features. It calculates the conditional probability of the feature under each category and selects the category with the highest probability as the classification result.

[0116] The basic principle of the naive Bayes model is based on Bayes' theorem and the assumption of conditional independence of features. Bayes' theorem describes the conditional probability relationship between two events, and the naive Bayes model further assumes that each feature is independent of each other. This allows the model to greatly simplify the calculation process when calculating the joint probability of features. Specifically, for a given sample feature X = (X1, X2...Xn) and category y, the naive Bayes model selects the category with the maximum probability as the classification result by calculating P(y)∏i=1nP(xi|y). The construction steps include:

[0117] 1. Divide the dataset into training set and test set

[0118] 2. Calculate the probability, calculate the prior probability of each category and calculate the conditional probability of each feature in each category, that is, p(ti)

[0119] 3. Training model, according to the calculated probability, train the naive Bayes classifier

[0120] In this embodiment, the core idea of ​​the Naive Bayes model lies in its concise assumptions and efficient calculation method, which enables it to perform well in many practical applications, thereby improving the efficiency and accuracy of model classification.

[0121] In some embodiments, identifying images according to a preset detection target and screening out images that meet the preset detection target includes:

[0122] Standardize the input image to make it meet the requirements of the SVM algorithm;

[0123] Use kernel functions to map the input space to a high-dimensional space;

[0124] According to the loss function, use the gradient descent algorithm to find the optimal category boundary;

[0125] The position of the category boundary is determined by the support vector, the data set is divided into different categories, and images that meet the preset detection target are screened out.

[0126] This application also provides a device for controlling the quality of image data acquisition. Figure 2 As shown, the device comprises:

[0127] The data acquisition module 100 is used to acquire image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to the sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value;

[0128] The feature extraction module 200 is used to establish a feature set using the features of the images in the sample library, extract the features of the images in the sample library based on standard images of various features, and calculate the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features;

[0129] The classification module 300 is used to establish a feature classification model corresponding to each feature according to the features, and use the feature classification model to identify and classify the target object in the image; according to the recognition result of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

[0130] The functional implementation of each module in the above-mentioned image data-based acquisition quality control device corresponds to the steps in the above-mentioned image data acquisition quality control method embodiment, and its functions and implementation processes are no longer repeated here.

[0131] The present application also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the steps of the image data acquisition quality control method as described in any of the above embodiments are implemented.

[0132] The present application also provides a computer-readable storage medium, on which a program is stored, wherein the computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive, and / or a memory stick, etc., and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the above embodiment of a method for controlling the acquisition quality of image data, and will not be repeated here.

[0133] The application also provides a computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the image data acquisition quality control method as described in any of the above embodiments.

[0134] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in the present application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0135] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, all possible combinations of the technical features in the above embodiments are not described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification. The above-mentioned embodiments only express several implementation methods of the present application, and their descriptions are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, without departing from the concept of the present application, several deformations and improvements can be made, which all belong to the protection scope of the present application. Therefore, the protection scope of the patent of this application shall be based on the attached claims.

Claims

1. A method for controlling the acquisition quality of image data, characterized in that: include: Collect image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to the sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value; Establish a feature set using the features of the images in the sample library, extract the features of the images in the sample library based on standard images of various features, and calculate the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features; Establishing a feature classification model corresponding to each feature according to each feature, and using the feature classification model to identify and classify the target object in the image; According to the recognition results of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

2. The image data acquisition quality control method according to claim 1, characterized in that: include: The Tenengrad gradient method is used to detect whether the clarity of the target object reaches the target value.

3. The image data acquisition quality control method according to claim 2, characterized in that: The texture features of the images in the sample library are extracted using a gray level co-occurrence matrix, a histogram of oriented gradients, a local binary pattern (LBP) or a Gaussian filter.

4. The image data acquisition quality control method according to claim 3, characterized in that: The matching degree between images is calculated using similarity measurement criteria, including: The matching degree between images is calculated using the phase correlation method, Hausdorff distance function, Euclidean distance discriminant method or normalized product correlation method.

5. The image data acquisition quality control method according to claim 4, characterized in that: The feature classification model is established based on the Naive Bayes model.

6. The image data acquisition quality control method according to claim 1, characterized in that: Identify images according to preset detection targets and select images that meet the preset detection targets, including: Standardize the input image to make it meet the requirements of the SVM algorithm; Use kernel functions to map the input space to a high-dimensional space; According to the loss function, use the gradient descent algorithm to find the optimal category boundary; The position of the category boundary is determined by the support vector, the data set is divided into different categories, and images that meet the preset detection target are screened out.

7. A device for controlling the acquisition quality of image data, characterized in that: include: A data acquisition module is used to collect image data, identify images according to preset detection targets, screen out images that meet the preset detection targets, and add the images that meet the preset detection targets to a sample library. The preset detection targets include background, lighting, shadow, the position of the target object, and whether the clarity of the target object reaches the target value; A feature extraction module, used to establish a feature set using the features of the images in the sample library, extract the features of the images in the sample library based on standard images of various features, and calculate the matching degree between the images using a similarity measurement criterion, wherein the features include color, contour, texture, Blob or Sift features; A classification module is used to establish a feature classification model corresponding to each feature according to the features, and use the feature classification model to identify and classify the target objects in the image; according to the recognition results of the feature classification model, new features and categories are screened out and added to the feature set and the sample library.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for controlling the acquisition quality of image data according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, wherein when the program is executed by a processor, the image data acquisition quality control method according to any one of claims 1 to 6 is implemented.

10. A computer program product comprising computer instructions, characterized in that When executed by a processor, the computer instructions implement the steps of the image data acquisition quality control method described in claims 1 to 6.

Images