A method and device for evaluating image classification network performance
By evaluating the similarity and interleaving ratio of the image classification network, the problem of low classification efficiency caused by lack of performance evaluation in the prior art is solved, and intuitive and efficient performance evaluation and performance qualification judgment are achieved.
Patent Information
- Application Number
- CN202110712226.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-06-25
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2041-06-25
AI Technical Summary
The prior art lacks effective methods to evaluate the performance of image classification networks, resulting in inefficient classification and the need for frequent retraining of the network.
By obtaining the similarity between the actual image data of the image to be predicted and the standard image data of various categories in the image classification network, counting the topK similarities, generating a similarity interval, and calculating the interchange ratio to determine whether the classification performance of the network is qualified.
This method intuitively and efficiently evaluates the performance of the image classification network through the evaluation of similarity and interleaving ratio, improves the classification efficiency and ensures that the classification performance of the network is qualified.
Smart Images

Figure CN113344109B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of image processing, and in particular relates to a method and device for evaluating the performance of an image classification network. Background Art
[0002] Image classification, at its core, is the task of assigning a label to an image from a given set of categories. In practice, this means analyzing an input image and returning a label that classifies the image, that is, by inputting the image into a neural network, which then outputs the category of the object.
[0003] In order to improve the accuracy of image classification, it is necessary to continuously add pictures that the network recognizes incorrectly or classifies poorly to the training set of the neural network. Usually, the selection and addition of such pictures are done manually and then judged manually, which leads to low judgment efficiency. In addition, there is no effective method to evaluate the performance of image classification networks on the market. The inability to evaluate the performance of image classification networks leads to poor classification results after they are put into use, and the image classification network needs to be retrained, which in turn leads to low classification efficiency. Summary of the invention
[0004] The purpose of the present invention is to provide a method and device for evaluating the performance of an image classification network, aiming to solve the technical problem in the prior art that there is no effective method for evaluating the performance of an image classification network, resulting in low subsequent classification efficiency.
[0005] To achieve the above object, an embodiment of the present invention provides an image classification network performance evaluation method, the method comprising:
[0006] Obtain the actual image data of the picture to be predicted;
[0007] Obtaining actual similarities between the actual image data and standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity;
[0008] Count the topK actual similarities among the actual similarities corresponding to each category in the image classification network; each category has a different topK actual similarities;
[0009] Generate similarity intervals for each category based on the topK actual similarities of each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals of other categories outside the classification interval are the comparison intervals;
[0010] Calculating the intersection-and-union ratios of the classification interval and each of the comparison intervals respectively;
[0011] Determine whether each of the IoU values is less than a preset IoU threshold; if so, determine that the classification performance of the image classification network is qualified.
[0012] Optionally, the method further comprises:
[0013] Get the largest actual similarity among the topK actual similarities corresponding to each category and record it as the classification similarity;
[0014] The category corresponding to the classification similarity is defined as the actual category of the image to be predicted;
[0015] The classification interval is generated according to the topK actual similarities corresponding to the actual classification.
[0016] Optionally, generating similarity intervals for each category according to the topK actual similarities for each category includes:
[0017] Extract the minimum actual similarity among the topK actual similarities of each category;
[0018] Extract the maximum actual similarity among the topK actual similarities of each category;
[0019] The minimum actual similarity and the maximum actual similarity of each category are used as interval endpoints of the similarity interval to generate the similarity interval.
[0020] Optionally, the intersection-and-union ratio of the classification interval and each of the comparison intervals is calculated by the following formula:
[0021]
[0022] Among them, Iou is the intersection-and-union ratio, A is the classification interval, and B is the comparison interval.
[0023] Optionally, the actual image data is a feature vector; the actual image data of the image to be predicted is obtained as follows:
[0024] Train the image classification network and generate a deep learning network model;
[0025] Get the actual feature vector of the image to be predicted after the deep learning network model is operated.
[0026] Optionally, obtaining the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated includes:
[0027] Use the deep learning network model to extract features from the images in the training set and obtain standard feature vectors;
[0028] Establishing a standard vector retrieval library according to the standard feature vector;
[0029] The similarity between the actual feature vector and each standard feature vector in the standard vector retrieval library is obtained and recorded as the actual similarity.
[0030] Optionally, the method further comprises:
[0031] A classification interval distribution map is generated according to each of the similarity intervals; the classification interval distribution map is used to indicate the interval distribution of each of the similarity intervals.
[0032] The present invention also provides an image classification network performance evaluation device, the device comprising:
[0033] An actual image data acquisition module, used to acquire actual image data of the image to be predicted;
[0034] An actual similarity acquisition module is used to acquire the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity;
[0035] The statistical module is used to count the topK actual similarities among the actual similarities corresponding to each category in the image classification network; each category has different topK actual similarities;
[0036] A similarity interval generation module is used to generate similarity intervals for each category according to the topK actual similarities for each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals for other categories outside the classification interval are the comparison intervals;
[0037] An intersection-and-union ratio calculation module, used to respectively calculate the intersection-and-union ratios of the classification interval and each of the comparison intervals;
[0038] The performance judgment module is used to judge whether each of the intersection-and-union ratio values is less than a preset intersection-and-union ratio threshold; if the judgment is yes, the classification performance of the image classification network is determined to be qualified.
[0039] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned image classification network performance evaluation method when executing the computer program.
[0040] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps in the above-mentioned image classification network performance evaluation method are implemented.
[0041] The above one or more technical solutions in the image classification network performance evaluation method and device provided by the embodiment of the present invention have at least one of the following technical effects:
[0042] The present invention counts the topK actual similarities among the actual similarities corresponding to each category in the image classification network; and generates similarity intervals for each category according to the topK actual similarities for each category; then respectively calculates the intersection-and-union ratios of the classification intervals and each of the comparison intervals; finally, determines whether each of the intersection-and-union ratios is less than a preset intersection-and-union ratio threshold; if the determination is yes, the classification performance of the image classification network is determined to be qualified, so that the classification performance of the image classification network to be evaluated is determined by using the similarity and the intersection-and-union ratio as a reference and the overlap of the intersection-and-union ratio as a measurement standard, and the evaluation is intuitive and efficient, thereby improving the subsequent classification efficiency, and has extremely high practicality. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0044] Figure 1 A flowchart of steps S100-S600 of the image classification network performance evaluation method provided by an embodiment of the present invention;
[0045] Figure 2 A flowchart of steps S450-S470 of the image classification network performance evaluation method provided by an embodiment of the present invention;
[0046] Figure 3 A flowchart of steps S410-S430 of the image classification network performance evaluation method provided by an embodiment of the present invention;
[0047] Figure 4 A flowchart of steps S110-S120 of the image classification network performance evaluation method provided by an embodiment of the present invention;
[0048] Figure 5 A flowchart of steps S210-S230 of the image classification network performance evaluation method provided by an embodiment of the present invention;
[0049] Figure 6 A schematic diagram of a classification interval distribution diagram provided by an embodiment of the present invention;
[0050] Figure 7 A structural block diagram of an image classification network performance evaluation device provided by an embodiment of the present invention;
[0051] Figure 8 A structural block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0052] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.
[0053] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.
[0054] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.
[0055] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.
[0056] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.
[0057] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.
[0058] In one embodiment of the present invention, Figure 1 As shown, a method for evaluating the performance of an image classification network is provided, the method comprising:
[0059] Step S100: Acquire actual image data of the picture to be predicted;
[0060] Step S200: obtaining the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity;
[0061] Specifically, there are different measurement standards for actual similarity, and different similarity measurement standards correspond to different similarity values.
[0062] The similarity metric includes Euclidean distance and spectral angle.
[0063] The similarity metric standard is selected by those skilled in the art according to actual conditions to achieve fast and efficient comparison of similarities.
[0064] Furthermore, in this embodiment, the image classification network includes different categories, and each category corresponds to different standard image data.
[0065] In this step, the actual image data is compared with the standard image data corresponding to each category in the image classification network, and an actual similarity is obtained respectively.
[0066] For example, when there are four categories in the image classification network, namely A, B, C and D, A, B, C and D categories respectively correspond to 100 standard image data, and the actual image data are compared with the 100 standard image data corresponding to A, B, C and D categories respectively. When the actual image data is compared with the 100 standard image data of category A, 100 actual similarities are obtained. Similarly, 100 actual similarities are obtained by comparing with categories B, C and D respectively.
[0067] Step S300: Counting the topK actual similarities among the actual similarities corresponding to each category in the image classification network; wherein each category corresponds to a different topK actual similarities;
[0068] In this step, the topK actual similarities are the K similarities that reflect the most similarity between the actual image data and the standard image data corresponding to each category in the image classification network.
[0069] Specifically, when the actual image data is compared with 100 standard image data of category A respectively to obtain 100 actual similarities, the actual similarities corresponding to the standard image data most similar to the actual image data are ranked first, and the remaining actual similarities are ranked in descending order. When K is 4, the actual similarities ranked in the top 4 are counted.
[0070] Step S400: Generate similarity intervals for each category according to the topK actual similarities of each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals of other categories outside the classification interval are the comparison intervals;
[0071] Specifically, the standard image data corresponding to the largest actual similarity among the topK actual similarities is the image data most similar to the image to be predicted, and the category described by the standard image data is the classification category of the image to be predicted.
[0072] For example, after comparing the predicted image with the standard image data in category A, the topK actual similarities are 0.8, 0.9 and 0.92 respectively. Therefore, the similarity interval is [0.8, 0.92].
[0073] After comparing with the standard image data in category B, the topK actual similarities are 0.1, 0.11 and 0.13 respectively. Therefore, the similarity interval is [0.1, 0.13].
[0074] After comparing with the standard image data in category C, the topK actual similarities are 0.21, 0.22 and 0.23 respectively. Therefore, the similarity interval is [0.21, 0.23].
[0075] After comparing with the standard image data in category D, the topK actual similarities are 0.26, 0.28 and 0.29 respectively. Therefore, the similarity interval is [0.26, 0.29].
[0076] In this embodiment, the larger the actual similarity value is, the more similar it is.
[0077] Therefore, the actual maximum similarity is 0.92, and the corresponding classification of the image to be predicted is class A. The corresponding classification interval is [0.8, 0.92], and the rest are comparison intervals: [0.1, 0.13], [0.21, 0.23] and [0.26, 0.29].
[0078] Step S500: respectively calculating the intersection-and-union ratios of the classification interval and each of the comparison intervals;
[0079] Step S600: Determine whether each of the IoU values is less than a preset IoU threshold; if so, determine that the classification performance of the image classification network is qualified.
[0080] In this step, the intersection-and-union ratio threshold is preset. When it is judged as yes, it means that the intersection-and-union ratio is less than the preset intersection-and-union ratio threshold, which means that the classification effect is better at this time, and the classification performance of the image classification network is determined to be qualified.
[0081] The present invention counts the topK actual similarities among the actual similarities corresponding to each category in the image classification network; and generates similarity intervals for each category according to the topK actual similarities for each category; then respectively calculates the intersection-and-union ratios of the classification intervals and each of the comparison intervals; finally, determines whether each of the intersection-and-union ratios is less than a preset intersection-and-union ratio threshold; if the determination is yes, the classification performance of the image classification network is determined to be qualified, so that the classification performance of the image classification network to be evaluated is determined by using the similarity and the intersection-and-union ratio as a reference and the overlap of the intersection-and-union ratio as a measurement standard, and the evaluation is intuitive and efficient, thereby improving the subsequent classification efficiency, and has extremely high practicality.
[0082] In another embodiment of the present invention, Figure 2 As shown, the method also includes:
[0083] Step S450: obtaining the largest actual similarity among the topK actual similarities corresponding to each category and recording it as the classification similarity;
[0084] Step S460: defining the category corresponding to the classification similarity as the actual category of the image to be predicted;
[0085] Step S470: generating the classification interval according to the topK actual similarities corresponding to the actual classification.
[0086] In another embodiment of the present invention, Figure 3 As shown, generating similarity intervals for each category according to the topK actual similarities for each category includes:
[0087] Step S410: extracting the minimum actual similarity among the topK actual similarities of each category;
[0088] Step S420: extracting the maximum actual similarity among the topK actual similarities of each category;
[0089] Step S430: taking the minimum actual similarity and the maximum actual similarity of each category as the interval endpoints of the similarity interval to generate the similarity interval.
[0090] In this embodiment, the larger the actual similarity value is, the more similar it is. If the topK actual similarities are 0.8, 0.9 and 0.92 respectively, the minimum actual similarity is 0.8 and the maximum actual similarity is 0.92. Therefore, the similarity interval is [0.8, 0.92].
[0091] In another embodiment of the present invention, the intersection-and-union ratio of the classification interval and each of the comparison intervals is calculated by the following formula:
[0092]
[0093] Among them, Iou is the intersection-and-union ratio, A is the classification interval, and B is the comparison interval.
[0094] In another embodiment of the present invention, Figure 4 As shown, the actual image data is a feature vector; the actual image data of the image to be predicted is obtained as follows:
[0095] Step S110: training the image classification network and generating a deep learning network model;
[0096] Step S120: Obtain the actual feature vector of the image to be predicted after the deep learning network model is operated.
[0097] Specifically, compared with the prior art, the acquisition of feature vectors facilitates subsequent similarity comparison and improves efficiency.
[0098] In another embodiment of the present invention, Figure 5 As shown, obtaining the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated includes:
[0099] Step S210: extracting features from the images in the training set through a deep learning network model and obtaining a standard feature vector;
[0100] Specifically, in this step, the feature extraction method is set according to the actual situation, and different network models are different, usually the last fully connected layer or convolutional layer of the neural network.
[0101] Step S220: establishing a standard vector retrieval library according to the standard feature vector;
[0102] Step S230: obtaining the similarity between the actual feature vector and each standard feature vector in the standard vector search library, and recording it as the actual similarity.
[0103] In another embodiment of the present invention, Figure 6 As shown, the method also includes:
[0104] A classification interval distribution map is generated according to each of the similarity intervals; the classification interval distribution map is used to indicate the interval distribution of each of the similarity intervals.
[0105] Specifically, the classification interval distribution diagram is as follows Figure 6 As shown in FIG. 1 , each vertical bar in the figure is a classification interval distribution diagram. Through the classification interval distribution diagram, the interval distribution of each similarity interval can be intuitively seen.
[0106] In another embodiment of the present invention, Figure 7 As shown, the device comprises:
[0107] An actual image data acquisition module, used to acquire actual image data of the image to be predicted;
[0108] An actual similarity acquisition module is used to acquire the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity;
[0109] The statistical module is used to count the topK actual similarities among the actual similarities corresponding to each category in the image classification network; each category has different topK actual similarities;
[0110] A similarity interval generation module is used to generate similarity intervals for each category according to the topK actual similarities for each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals for other categories outside the classification interval are the comparison intervals;
[0111] An intersection-and-union ratio calculation module, used to respectively calculate the intersection-and-union ratios of the classification interval and each of the comparison intervals;
[0112] The performance judgment module is used to judge whether each of the intersection-and-union ratio values is less than a preset intersection-and-union ratio threshold; if the judgment is yes, the classification performance of the image classification network is determined to be qualified.
[0113] In another embodiment of the present invention, the similarity interval generation module is also used to: extract the minimum actual similarity among the topK actual similarities of each category; extract the maximum actual similarity among the topK actual similarities of each category; use the minimum actual similarity and the maximum actual similarity of each category as the interval endpoints of the similarity interval to generate the similarity interval; obtain the maximum actual similarity among the topK actual similarities corresponding to each category and record it as the classification similarity; define the category corresponding to the classification similarity as the actual category of the picture to be predicted; and generate the classification interval according to the topK actual similarities corresponding to the actual category.
[0114] In another embodiment of the present invention, the actual image data acquisition module is also used to train the image classification network and generate a deep learning network model; and obtain the actual feature vector obtained after the image to be predicted is operated by the deep learning network model.
[0115] In another embodiment of the present invention, the actual similarity acquisition module is also used to extract features of images in the training set through a deep learning network model and obtain standard feature vectors; establish a standard vector retrieval library based on the standard feature vectors; obtain the similarity between the actual feature vector and each standard feature vector in the standard vector retrieval library, and record it as the actual similarity.
[0116] In another embodiment of the present invention, the device further comprises a distribution map generating module, which is used to generate a classification interval distribution map according to each of the similarity intervals; the classification interval distribution map is used to indicate the interval distribution of each of the similarity intervals.
[0117] In another embodiment of the present invention, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above-mentioned image classification network performance evaluation method when executing the computer program.
[0118] In another embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned image classification network performance evaluation method are implemented.
[0119] 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 this 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).
[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for evaluating the performance of an image classification network, characterized in that: The method comprises: Obtain the actual image data of the picture to be predicted; Obtaining actual similarities between the actual image data and standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity; Count the topK actual similarities among the actual similarities corresponding to each category in the image classification network; each category has a different topK actual similarities; Generate similarity intervals for each category based on the topK actual similarities of each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals of other categories outside the classification interval are the comparison intervals; Calculating the intersection-and-union ratios of the classification interval and each of the comparison intervals respectively; Determine whether each of the intersection-over-union ratio values is less than a preset intersection-over-union ratio threshold; if the determination is yes, determine that the classification performance of the image classification network is qualified; The actual image data is a feature vector; the actual image data of the image to be predicted is obtained as follows: Train the image classification network and generate a deep learning network model; Get the actual feature vector of the image to be predicted after the deep learning network model is operated; The method further comprises: Get the largest actual similarity among the topK actual similarities corresponding to each category and record it as the classification similarity; The category corresponding to the classification similarity is defined as the actual category of the image to be predicted; Generate the classification interval according to the topK actual similarities corresponding to the actual classification; The similarity intervals for each category are generated based on the topK actual similarities of each category: Extract the minimum actual similarity among the topK actual similarities of each category; Extract the maximum actual similarity among the topK actual similarities of each category; Taking the minimum actual similarity and the maximum actual similarity of each category as the interval endpoints of the similarity interval to generate the similarity interval; Obtaining the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated includes: Use the deep learning network model to extract features from the images in the training set and obtain standard feature vectors; Establishing a standard vector retrieval library according to the standard feature vector; The similarity between the actual feature vector and each standard feature vector in the standard vector retrieval library is obtained and recorded as the actual similarity.
2. The image classification network performance evaluation method according to claim 1, characterized in that: The intersection-and-union ratio of the classification interval and each comparison interval is calculated by the following formula: ; Among them, Iou is the intersection-union ratio, A is the classification interval, and B is the comparison interval.
3. The image classification network performance evaluation method according to claim 1, characterized in that: The method further comprises: A classification interval distribution map is generated according to each of the similarity intervals; the classification interval distribution map is used to indicate the interval distribution of each of the similarity intervals.
4. An image classification network performance evaluation device, characterized in that: The device comprises: An actual image data acquisition module, used to acquire actual image data of the image to be predicted; An actual similarity acquisition module is used to acquire the actual similarity between the actual image data and the standard image data corresponding to each category in the image classification network to be evaluated; wherein each category corresponds to a different actual similarity; The statistical module is used to count the topK actual similarities among the actual similarities corresponding to each category in the image classification network; each category has different topK actual similarities; A similarity interval generation module is used to generate similarity intervals for each category according to the topK actual similarities for each category; wherein the similarity interval corresponding to the largest actual similarity among the topK actual similarities is the classification interval, and the similarity intervals for other categories outside the classification interval are the comparison intervals; An intersection-and-union ratio calculation module, used to respectively calculate the intersection-and-union ratios of the classification interval and each of the comparison intervals; A performance judgment module, used to judge whether each of the intersection-and-union ratio values is less than a preset intersection-and-union ratio threshold; if the judgment is yes, the classification performance of the image classification network is determined to be qualified; The actual image data is a feature vector; the actual image data of the image to be predicted is obtained as follows: Train the image classification network and generate a deep learning network model; Get the actual feature vector of the image to be predicted after the deep learning network model is operated; The similarity interval generation module is also used to: obtain the largest actual similarity among the topK actual similarities corresponding to each category and record it as the classification similarity; define the category corresponding to the classification similarity as the actual category of the image to be predicted; and generate the classification interval according to the topK actual similarities corresponding to the actual category; The similarity interval generation module is further used to: extract the minimum actual similarity among the topK actual similarities of each category; extract the maximum actual similarity among the topK actual similarities of each category; use the minimum actual similarity and the maximum actual similarity of each category as the interval endpoints of the similarity interval to generate the similarity interval; The actual similarity acquisition module is also used to: extract features from images in the training set through a deep learning network model and obtain standard feature vectors; establish a standard vector retrieval library based on the standard feature vectors; obtain the similarity between the actual feature vector and each standard feature vector in the standard vector retrieval library, and record it as the actual similarity.
5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 3 are implemented.
6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
Citation Information
Patent Citations
Damaged corn particle detection and classification method based on deep learning
CN111310756A
Image processing method and device, and equipment
CN111986217A
Automated evaluation of neural networks using trained classifier
US20190122096A1
Interactive visualization evaluation for classification models
US20200104753A1
