Intelligent picture data cleaning and dividing method and device and medium

Abstract

The invention discloses an intelligent picture data cleaning and dividing method and device and a medium, and the method comprises the steps: detecting the feature points of each picture one by one through a deep learning model, reading the number of the feature points of each picture, and removing the pictures of which the number of the feature points is smaller than a preset value; carrying out data set division on the pictures left after elimination, and carrying out feature point matching on the remaining pictures one by one by using another deep learning model; and splitting the two pictures of which the matching degree is greater than a threshold value into a training data set and an evaluation data set according to a matching result. According to the method, the fuzzy pictures in the picture data can be effectively eliminated through deep learning, and the pictures with high similarity can be divided into the training data set and the evaluation data set, so that the neural network model can learn richer features, the picture data processing efficiency is high, the utilization rate of the picture data is high, and the waste of the picture data is reduced.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and specifically to a method, device and medium for cleaning and dividing embodied intelligent image data based on deep learning and feature matching. Background Art

[0002] In the deep learning of artificial intelligence models, a large amount of data is needed for training and model testing. Therefore, a large amount of data needs to be collected, cleaned and classified. In order to further expand the amount of data, data enhancement processing such as scaling, rotation, and splicing is also required. In the data processing process, operations such as data cleaning and classification require the participation of humans with processing experience. In the early stage of work, data cleaning and division often rely on manual experience and judgment. It is necessary to manually check the pictures in the data set one by one, exclude blurred and repeated data, and finally divide the data into training data sets and evaluation data sets according to a fixed ratio. In the preprocessing stage of converting video data to RGB pictures, pictures are often extracted and saved in the form of fixed frames at intervals.

[0003] As the demand for artificial intelligence models increases, the demand for data also increases dramatically, and the method of manual data processing can no longer meet the current data demand. In addition, manual data processing often leads to errors due to factors such as fatigue; and dividing the training data set and the evaluation data set according to a fixed ratio often leads to the problem of imbalance between the training data set and the evaluation data set. When collecting pictures in video data by fixed frames, it is easy to collect fuzzy data. Therefore, it is necessary to improve the existing data processing method to improve the efficiency and accuracy of data processing. Summary of the invention

[0004] In view of the above problems, the present application provides an embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching, which is used to solve the technical problems that the above data processing relies on manual labor and has low efficiency and accuracy.

[0005] To achieve the above objectives, the present application provides an embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching, comprising the following steps:

[0006] Detect the feature points of each image one by one through a deep learning model, read the number of feature points of each image, and remove the images whose number of feature points is less than a preset value;

[0007] The remaining pictures after the elimination are divided into data sets, and feature points of the remaining pictures are matched one by one using another deep learning model; and the two pictures with matching degrees greater than a threshold are split into a training data set and an evaluation data set according to the matching results.

[0008] Furthermore, the feature points of the pictures are obtained by performing feature detection on each of the pictures using a superpoint model;

[0009] The using another deep learning model to match the feature points of the remaining pictures one by one includes:

[0010] Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model;

[0011] Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor;

[0012] The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

[0013] Furthermore, when the superpoint model is used to perform feature detection on each of the images, it includes:

[0014] Raise the feature point score threshold of the output of the superpoint model to above 0.7;

[0015] A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

[0016] Furthermore, it also includes:

[0017] The pictures whose feature points are greater than or equal to the preset value are saved in a new data set, and the feature points and descriptors of each picture are saved in a yml file.

[0018] In order to solve the above technical problems, this application also provides another technical solution:

[0019] An embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching, comprising:

[0020] A fuzzy image removal module is used to detect the feature points of each image one by one through a deep learning model, read the number of feature points of each image, and remove the images whose number of feature points is less than a preset value;

[0021] The image division module is used to divide the remaining images after the elimination into data sets, and use another deep learning model to match the feature points of the remaining images one by one; and split the two images with a matching degree greater than a threshold into a training data set and an evaluation data set according to the matching results.

[0022] Furthermore, the feature points of the pictures are obtained by performing feature detection on each of the pictures using a superpoint model;

[0023] The using another deep learning model to match the feature points of the remaining pictures one by one includes:

[0024] Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model;

[0025] Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor;

[0026] The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

[0027] Furthermore, when the superpoint model is used to perform feature detection on each of the images, it includes:

[0028] Raise the feature point score threshold of the output of the superpoint model to above 0.7;

[0029] A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

[0030] Furthermore, the blurred image removal module is also used to save the images whose feature points are greater than or equal to the preset value into a new data set, and save the feature points and descriptors of each image in a yml file.

[0031] In order to solve the above technical problems, this application also provides another technical solution:

[0032] A computer-readable storage medium stores a computer program, which, when executed by a processor, executes the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching described in any of the above technical solutions.

[0033] Different from the existing technology, the above technical solution detects the feature points of the picture through a deep learning model, and can remove blurry pictures with feature points lower than the preset value, retaining clear high-quality pictures; and can match the detected feature points through another deep learning model to obtain the matching degree of any two pictures, and divide the two pictures with matching degree (i.e. similarity) greater than the threshold into training data sets and evaluation data sets, thereby avoiding data waste and maintaining a relative balance between training data sets and evaluation data sets. In addition, the above technical solution can be fully automated without the need for manual experience, greatly improving the efficiency and accuracy of image processing.

[0034] The above-mentioned records related to the invention content are only an overview of the technical solution of the present application. In order to enable ordinary technicians in the field to more clearly understand the technical solution of the present application, and then implement it according to the text of the specification and the contents recorded in the drawings, and to make the above-mentioned purpose and other purposes, features and advantages of the present application easier to understand, the following is an explanation in combination with the specific implementation mode and drawings of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings are only used to illustrate the principles, implementation methods, applications, characteristics and effects of the specific embodiments of the present invention and other related contents, and shall not be considered as limiting the present application.

[0036] In the drawings of the specification:

[0037] Figure 1 It is a flowchart of the method for cleaning and dividing embodied intelligent image data based on deep learning and feature matching described in the specific implementation method;

[0038] Figure 2 A specific flow chart of the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching described in the specific implementation method;

[0039] Figure 3 It is a module block diagram of the embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching described in the specific implementation method;

[0040] Figure 4 A schematic diagram of a computer-readable storage medium according to a specific embodiment;

[0041] The reference numerals in the above drawings are described as follows:

[0042] 300. Embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching;

[0043] 301. Blurred image removal module; 302. Image segmentation module;

[0044] 400. Computer-readable storage medium. DETAILED DESCRIPTION

[0045] In order to explain in detail the possible application scenarios, technical principles, specific schemes that can be implemented, and the purposes and effects that can be achieved, the following is a detailed description of the specific embodiments listed in conjunction with the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, and are therefore only used as examples, and cannot be used to limit the scope of protection of the present application.

[0046] Reference to "embodiment" herein means that the specific features, structures or characteristics described in conjunction with the embodiment may be included in at least one embodiment of the present application. The term "embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment, nor does it particularly limit its independence or association with other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, the various technical features mentioned in the embodiments can be combined in any way to form a corresponding implementable technical solution.

[0047] Unless otherwise defined, the technical terms used in this document have the same meanings as those generally understood by those skilled in the art to which this application belongs; the use of relevant terms in this document is only for describing specific embodiments and is not intended to limit this application.

[0048] In the description of this application, the term "and / or" is an expression used to describe the logical relationship between objects, indicating that three relationships may exist, for example, A and / or B, which means: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this article generally indicates that the objects before and after are in an "or" logical relationship.

[0049] In the present application, terms such as “first” and “second” are merely used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship of quantity, priority or sequence between these entities or operations.

[0050] Without further limitations, in this application, the words "include", "comprises", "has" or other similar open-ended expressions used in sentences are intended to cover non-exclusive inclusion. These expressions do not exclude the presence of additional elements in the process, method or product including the elements, so that the process, method or product including a series of elements may include not only those limited elements, but also other elements not explicitly listed, or also include elements inherent to such process, method or product.

[0051] Similar to the understanding in the Examination Guidelines, in this application, expressions such as "greater than", "less than", "exceed" and the like are understood to exclude the number itself; expressions such as "above", "below", "within" and the like are understood to include the number itself. In addition, in the description of the embodiments of this application, "multiple" means more than two (including two), and similar expressions related to "multiple" are also understood in this way, such as "multiple groups", "multiple times", etc., unless otherwise clearly and specifically limited.

[0052] In the description of the embodiments of the present application, space-related expressions used, such as "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", etc., indicate the orientation or position relationship based on the orientation or position relationship shown in the specific embodiments or drawings, and are only for the convenience of describing the specific embodiments of the present application or facilitating the reader's understanding, and do not indicate or imply that the referred device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the present application.

[0053] Unless otherwise expressly specified or limited, in the description of the embodiments of the present application, the terms such as "install", "connect", "connect", "fix", "set", etc. used should be understood in a broad sense. For example, the "connection" can be a fixed connection, a detachable connection, or an integrated setting; it can be a mechanical connection, an electrical connection, or a communication connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium; it can be the internal connection of two elements or the interaction relationship between two elements. For technicians in the technical field to which the present application belongs, the specific meanings of the above terms in the embodiments of the present application can be understood according to the specific circumstances.

[0054] In the related art, data such as pictures used in artificial intelligence model training require manual participation in data processing such as data cleaning and classification. Therefore, there are technical problems such as low processing efficiency and easy errors, which cannot meet the current demand for data volume in the development of artificial intelligence. Based on this technical problem, the present embodiment provides an embodied intelligent picture data cleaning and division device based on deep learning and feature matching. The embodied intelligent picture data cleaning and division device based on deep learning and feature matching in this embodiment can perform feature point detection and feature point matching on pictures based on a deep learning model, so it can further eliminate blurred pictures based on the number of feature points, and identify the similarity of any two pictures based on feature point matching, and divide pictures with similarity greater than a threshold into a training data set and an evaluation data set, wherein the training data set is used to train the artificial intelligence model, and the evaluation data set is used to test the function of the trained artificial intelligence model.

[0055] See also Figures 1 to 4 This embodiment provides a method for cleaning and dividing embodied intelligent image data based on deep learning and feature matching. The method for cleaning and dividing embodied intelligent image data based on deep learning and feature matching includes the following steps:

[0056] S101, detecting feature points of each image one by one through a deep learning model, reading the number of feature points of each image, and eliminating images whose number of feature points is less than a preset value;

[0057] S102, dividing the remaining pictures after the elimination into data sets, and using another deep learning model to match feature points of the remaining pictures one by one; and splitting the two pictures with matching degrees greater than a threshold into a training data set and an evaluation data set according to the matching results.

[0058] In step S101, the feature points of the pictures may be obtained by performing feature detection on each of the pictures using a superpoint model.

[0059] The using another deep learning model to match the feature points of the remaining pictures one by one includes:

[0060] Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model;

[0061] Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor;

[0062] The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

[0063] Among them, the superpoint model is a CNN-based network model for image key point and descriptor extraction. Its main features include: Encoder-Decoder structure: The model adopts the Encoder-Decoder structure, the feature point detector and the descriptor network share a forward encoder, and then learn different network parameters according to different tasks.

[0064] The Superglue model is a model structure based on deep learning and is widely used in natural language processing tasks.

[0065] The MLP model, or Multilayer Perceptron, is a feedforward artificial neural network model that maps multiple input data sets to a single output data set.

[0066] A descriptor is a vector that encodes and represents image feature points or regions.

[0067] In other embodiments, similar models may also be used to replace the above-mentioned superglue model and Superpoint model.

[0068] In one embodiment, the use of the superpoint model to perform feature detection on each of the images includes:

[0069] Raise the feature point score threshold of the output of the superpoint model to above 0.7;

[0070] A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

[0071] In order to ensure that the number of feature points of the blurred image is as small as possible, the feature point score threshold of the superpoint model output is increased to above 0.7, where 0.7 is the value after normalization. In this embodiment, the feature value greater than 0 is classified as 1, and the feature value less than 0 is classified as -1, so as to optimize the instability of the loss function of binarization after clustering that occurs during the network training process.

[0072] In one embodiment, the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching further includes:

[0073] The pictures whose feature points are greater than or equal to the preset value are saved in a new data set, and the feature points and descriptors of each picture are saved in a yml file.

[0074] like Figure 2 As shown, it is a specific flow chart of the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching; the specific steps of the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching include:

[0075] 1. Blurry image removal ( Figure 2 The left part of the flow chart):

[0076] 1.1. Read an image from the image dataset;

[0077] 1.2. Use the superpoint model to detect feature points of the image. In order to ensure that the number of feature points of the blurred image is as small as possible, the feature point score threshold of the model output is increased to 0.7 (after normalization); and a corresponding binary layer is added to the output part of the feature vector, and the eigenvalues ​​greater than 0 are classified as 1, and those less than 0 are classified as -1, so as to optimize the instability of the binarization loss function after clustering that occurs during the network training process.

[0078] 1.3. Read the number of feature points unm of the image through the feature point output head of the superpoint model.

[0079] 1.31 Determine whether the number of feature points is greater than a preset value (for example, the preset value is 50). If the number is lower than the set feature point number threshold, remove them.

[0080] 1.32 If it is higher than the preset value, save the image image to a new dataset, and save the feature points and descriptors of each image in the yml file.

[0081] 1.4. Repeat steps 1.1-1.3 until all the data sets are completed. All blurry images can be removed through the above steps 1.1-1.4.

[0082] 2. Removal of high similarity images and data division ( Figure 2 The process on the right side):

[0083] 2.1. Read an image img1 in the dataset;

[0084] 2.2. Read another picture img2 in the dataset;

[0085] 2.3. Use the superglue model to match the features of img1 and img2 to ensure the matching accuracy:

[0086] A. The input features and descriptor parts of the superglue model directly use the data saved by the superpoint model.

[0087] B. Use the MLP network to increase the dimension of the descriptor and normalize the descriptor after the increase in dimension.

[0088] 2.4. The similarity between images is calculated by calculating the ratio of the matched feature point pairs in the matching results to all the feature points in the image. Images with a similarity higher than a threshold (e.g. 0.5) are considered to be images with repeated features and are moved to the evaluation dataset. The remaining images are used as the training dataset.

[0089] 2.5. Repeat the above steps 2.2-2.4 until image img1 matches all images.

[0090] 2.6. Repeat steps 2.1-2.5 above until there are no unread images in the data set. At this time, the data set contains images with different features that are conducive to training, while the evaluation set contains images with relatively repeated features. Some images in the training set or the evaluation set can be used as tests.

[0091] In the above embodiment, the feature points of the image are extracted through the deep learning model, and the number of feature points is used to judge the clarity of the image, so that the blurred images in the point image can be effectively filtered out and the quality of the image data can be improved. In this embodiment, the extracted feature points are matched through the deep learning model, and the similarity between different images is obtained according to the matching results, and the images with high similarity are accurately divided into the training data set and the evaluation data set, so that the image data can be fully utilized, and the image features in the obtained training data set are more obvious, which is convenient for learning more abundant features when training the neural network model.

[0092] like Figure 3 As shown, in another embodiment, an embodied intelligent image data cleaning and segmentation device 300 based on deep learning and feature matching is provided. The embodied intelligent image data cleaning and segmentation device 300 based on deep learning and feature matching includes: a fuzzy image removal module 301 and an image segmentation module 302.

[0093] The blurred image removal module 301 is used to detect the feature points of each image one by one through a deep learning model, read the number of feature points of each image, and remove the images whose number of feature points is less than a preset value.

[0094] The image division module 302 is used to divide the remaining images after the elimination into data sets, and use another deep learning model to match the feature points of the remaining images one by one; and split the two images with a matching degree greater than a threshold into a training data set and an evaluation data set according to the matching results.

[0095] The feature points of the pictures are obtained by performing feature detection on each of the pictures using a superpoint model; and the feature points of the remaining pictures are matched one by one using another deep learning model, including:

[0096] Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model;

[0097] Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor;

[0098] The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

[0099] When the superpoint model is used to perform feature detection on each of the images, it includes:

[0100] Raise the feature point score threshold of the output of the superpoint model to above 0.7;

[0101] A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

[0102] The blurred image removal module 301 is also used to save the images whose feature points are greater than or equal to the preset value into a new data set, and save the feature points and descriptors of each image in a yml file.

[0103] The description of each technical feature and the description of the technical effects in this embodiment are all based on the above-mentioned embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching, and will not be repeated here.

[0104] like Figure 4 As shown, in another embodiment, a computer-readable storage medium 400 is provided, in which a computer program is stored. When the computer program is executed by a processor, the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching described in any one of the above embodiments is executed.

[0105] The computer-readable storage medium may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a ferromagnetic random access memory (FRAM), a flash memory, a magnetic surface memory, an optical disk, or a compact disc read-only memory (CD-ROM); the magnetic surface memory may be a disk memory or a tape memory.

[0106] The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static random access memory (SR AM), synchronous static random access memory (SSR AM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), direct memory bus random access memory (DRRAM). The computer-readable storage medium described in the embodiments of the present invention is intended to include these and any other suitable types of memory.

[0107] In some embodiments, the processor can be implemented by software, hardware, firmware or a combination thereof, and can use circuits, single or multiple application specific integrated circuits (Application Specific Integrated Circuit, ASIC), digital signal processors (Digital Signal Processor, DSP), digital signal processing devices (Digital Signal Processing Device, DSPD), programmable logic devices (Programmable Logic Device, PLD), field programmable gate arrays (Field Programmable Gate Array, FPGA), central processing units (Central Processing Unit, CPU), controllers, microcontrollers, microprocessors, at least one of which, so that the processor can execute some or all of the steps or any combination of the steps in the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching described in the various embodiments of the present application.

[0108] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of this application, this does not limit the scope of patent protection of this application. All technical solutions generated by replacing or modifying equivalent structures or equivalent processes based on the essential concept of this application using the contents recorded in the specification and drawings of this application, as well as directly or indirectly implementing the technical solutions of the above embodiments in other related technical fields, are included in the scope of patent protection of this application.

Claims

1. A method for cleaning and segmenting embodied intelligent image data based on deep learning and feature matching, characterized in that: The following steps are involved: Detect the feature points of each image one by one through a deep learning model, read the number of feature points of each image, and remove the images whose number of feature points is less than a preset value; Dividing the remaining pictures after the elimination into data sets, and using another deep learning model to match feature points of the remaining pictures one by one; And according to the matching results, the two pictures with a matching degree greater than a threshold are split into a training data set and an evaluation data set.

2. The method for cleaning and segmenting embodied intelligent image data based on deep learning and feature matching according to claim 1, characterized in that: The feature points of the images are obtained by performing feature detection on each of the images using a superpoint model; The using another deep learning model to match the feature points of the remaining pictures one by one includes: Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model; Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor; The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

3. The method for cleaning and segmenting embodied intelligent image data based on deep learning and feature matching according to claim 2, characterized in that: When the superpoint model is used to perform feature detection on each of the images, it includes: Raise the feature point score threshold of the output of the superpoint model to above 0.7; A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

4. The method for cleaning and segmenting embodied intelligent image data based on deep learning and feature matching according to claim 1, characterized in that: Also includes: The pictures whose feature points are greater than or equal to the preset value are saved in a new data set, and the feature points and descriptors of each picture are saved in a yml file.

5. An embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching, characterized in that: include: A fuzzy image removal module is used to detect the feature points of each image one by one through a deep learning model, read the number of feature points of each image, and remove the images whose number of feature points is less than a preset value; A picture division module, used for dividing the pictures remaining after the elimination into data sets, and using another deep learning model to match feature points of each of the remaining pictures one by one; And according to the matching results, the two pictures with a matching degree greater than a threshold are split into a training data set and an evaluation data set.

6. The embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching according to claim 5, characterized in that: The feature points of the images are obtained by performing feature detection on each of the images using a superpoint model; The using another deep learning model to match the feature points of the remaining pictures one by one includes: Reading feature points and descriptors of two images, wherein the feature points and the descriptors are data obtained by performing feature point detection using the superpoint model; Using the well-trained MLP model to perform dimension-upgrading and normalization processing on the descriptor; The feature points and the descriptors after dimension increase and normalization are input into the superglue model to perform feature point matching.

7. The embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching according to claim 6, characterized in that: When the superpoint model is used to perform feature detection on each of the images, it includes: Raise the feature point score threshold of the output of the superpoint model to above 0.7; A binarization corresponding layer is set at the output part of the feature vector of the superpoint model to classify the values ​​with feature values ​​greater than 0 as 1 and the values ​​with feature values ​​less than 0 as -1.

8. The embodied intelligent image data cleaning and segmentation device based on deep learning and feature matching according to claim 5, characterized in that: The blurred image removal module is also used to save the images whose feature points are greater than or equal to the preset value into a new data set, and save the feature points and descriptors of each image in a yml file.

9. A computer-readable storage medium having a computer program stored therein, characterized in that: When the computer program is executed by the processor, the embodied intelligent image data cleaning and segmentation method based on deep learning and feature matching as described in any one of claims 1 to 4 is executed.

Images