Intelligent perception optimal assessment sample set construction method guided by image features

Abstract

The invention discloses an intelligent perception optimal assessment sample set construction method guided by image features, and belongs to the technical field of sample construction, and the method comprises the steps: employing a basic sample set to train an algorithm model, obtaining the influence degree of each image feature on the performance of the algorithm model, and screening out key image features; calculating an index value range of each key image feature and dividing an index interval; a plurality of samples are randomly selected from each index interval to form an interval training set for algorithm model training, the influence degree of each index interval on the performance of the algorithm model is obtained, and important intervals are screened out; and based on the key image features and the corresponding important intervals, an optimal assessment sample set is screened out from the basic sample set, so that the problem that the assessment sample set required for verification and assessment of the intelligent sensing algorithm cannot be effectively constructed by the existing method is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of sample construction, and in particular to a method for constructing an optimal assessment sample set of intelligent perception guided by image features. Background Art

[0002] At present, the method for constructing the assessment sample set required for intelligent perception algorithm verification, evaluation and optimization design usually adopts the "human in the loop" approach. First, the scene conditions that the assessment set should have are summarized and listed according to the application requirements, such as target size, natural conditions, environment and other information. Then, based on expert experience, large-scale assessment samples that meet the scene condition requirements are selected from the existing data set. Finally, based on prior knowledge, the selected sample set is manually analyzed to supplement the missing samples to form an assessment sample set. This assessment set construction method is a qualitative construction method based on expert knowledge, which is heavily dependent on the accumulation of prior knowledge. There are problems such as low efficiency in assessment sample construction and large scale of assessment data requirements. Existing methods cannot effectively construct the assessment sample set required for intelligent perception algorithm verification and evaluation. Summary of the invention

[0003] The purpose of the present invention is to provide a method for constructing an optimal assessment sample set for intelligent perception guided by image features, so as to solve the problem that existing methods cannot effectively construct the assessment sample set required for the verification and evaluation of intelligent perception algorithms.

[0004] To achieve the above object, the present invention adopts the following technical solution: On the one hand, this specification provides a method for constructing an optimal assessment sample set for intelligent perception guided by image features, comprising: Step 102, using the basic sample set to train the algorithm model, obtain the influence of each image feature on the performance of the algorithm model and screen out key image features; Step 104, calculating the index value range of each key image feature and dividing the index interval; Step 106, randomly extracting a number of samples from each index interval to form an interval training set for algorithm model training, obtaining the influence of each index interval on the performance of the algorithm model and screening out important intervals; Step 108 , based on the key image features and their corresponding important intervals, the optimal assessment sample set is selected from the basic sample set.

[0005] On the other hand, the present specification provides an image feature-guided intelligent perception optimal assessment sample set construction device, comprising: The key image feature screening module is used to train the algorithm model using the basic sample set, obtain the influence of each image feature on the performance of the algorithm model and screen out the key image features; An index interval division module is used to calculate the index value range of each key image feature and divide the index interval; The important interval screening module is used to randomly select a number of samples from each indicator interval to form an interval training set for algorithm model training, obtain the influence of each indicator interval on the performance of the algorithm model and screen out important intervals; The optimal assessment set construction module is used to select the optimal assessment sample set from the basic sample set based on key image features and their corresponding important intervals.

[0006] Based on the above technical solution, this specification can achieve the following technical effects: This method designs image feature indicators to guide the construction of assessment sample sets, analyzes the impact of image features on algorithm performance through ablation tests, finds out key image feature indicators, further finds out the important interval range of each image feature through quantitative calculation, and supplements the sample sets based on the corresponding important intervals according to the importance of key features to form the optimal assessment sample set, which effectively reduces the number of assessment sample sets and solves the problem that existing methods cannot effectively construct the assessment sample sets required for intelligent perception algorithm verification and evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 The present invention is a flowchart of a method for constructing an optimal assessment sample set for intelligent perception guided by image features in one embodiment of the present invention.

[0008] Figure 2 The present invention is a specific flow chart of a method for constructing an optimal assessment sample set of intelligent perception guided by image features in one embodiment of the present invention.

[0009] Figure 3 The flowchart of determining key image features in one embodiment of the present invention.

[0010] Figure 4 The flowchart of determining the important interval corresponding to the key image feature in one embodiment of the present invention.

[0011] Figure 5 It is a structural schematic diagram of a device for constructing an intelligent perception algorithm verification and evaluation index system in one embodiment of the present invention.

[0012] Figure 6 The figure is a schematic diagram of an electronic device according to the present invention. DETAILED DESCRIPTION

[0013] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become more apparent from the following description and claims. It should be noted that the drawings are all in very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.

[0014] It should be noted that, in order to clearly explain the content of the present invention, the present invention specifically cites multiple embodiments to further illustrate different implementations of the present invention, wherein the multiple embodiments are enumerated rather than exhaustive. In addition, for the sake of brevity of explanation, the contents mentioned in the previous embodiments are often omitted in the subsequent embodiments. Therefore, the contents not mentioned in the subsequent embodiments can refer to the previous embodiments accordingly. Example 1

[0015] Please refer to Figure 1 , Figure 1 The figure shows a method for constructing an optimal assessment sample set of intelligent perception guided by image features provided by this embodiment. In this embodiment, the method includes: Step 102, using the basic sample set to train the algorithm model, obtain the influence of each image feature on the performance of the algorithm model and screen out key image features; In this embodiment, one implementation of step 102 is: Step 202, dividing the basic sample set into a training set and a test set according to a preset ratio; Step 204, using the training set to train the algorithm model, and calculating the index value range of each image feature; Step 206, constructing an image feature training set for each image feature based on the index value range and the equal principle of each image feature; In this embodiment, one implementation of step 206 is: Step 2061, dividing the index value range of each image feature into three intervals by equal amount; Step 2062, dividing the samples in the training set into corresponding intervals based on the index value of the image feature of each sample in the training set to obtain interval subsets; Step 2063, randomly select one-third of the training samples from each interval subset to form a corresponding image feature training set.

[0016] Step 208, training the algorithm model using the image feature training set to obtain a trained first algorithm model; Step 210, using the test set to test the trained first algorithm model to obtain the degree of influence of the image feature on the performance of the algorithm model; Step 212, sorting the influence of all image features on the performance of the algorithm model, and screening out key image features with a greater influence on the performance of the algorithm model.

[0017] Step 104, calculating the index value range of each key image feature and dividing the index interval; In this embodiment, one implementation of step 104 is: Step 302, based on the index value range of each image feature, obtain the index value range of the key image feature; Step 304: divide the index value range of the key image feature into a number of index intervals of equal value.

[0018] Step 106, randomly extracting a number of samples from each index interval to form an interval training set for algorithm model training, obtaining the influence of each index interval on the performance of the algorithm model and screening out important intervals; In this embodiment, one implementation of step 106 is: Step 402, dividing the samples in the training set into corresponding index intervals based on the index value of the image feature of each sample in the training set, and obtaining an index interval subset; Step 404, randomly selecting a number of samples from each indicator interval subset to form a corresponding interval training set; Step 406, training the algorithm model using the interval training set to obtain a trained second algorithm model; Step 408, using the test set to test the second algorithm model, to obtain the degree of influence of the interval training set on the performance of the algorithm model; Step 410, sorting the influence of the interval training set on the algorithm model performance, and screening out important intervals with greater influence on the algorithm model performance.

[0019] Step 108 , based on the key image features and their corresponding important intervals, the optimal assessment sample set is selected from the basic sample set.

[0020] In this embodiment, one implementation of step 108 is: Step 502, based on the ranking of the key image features, obtaining the order of importance of the key image features; Step 504, based on the importance order of the key image features and the important interval corresponding to each key image feature, the assessment sample set of each key image feature is screened out from the basic sample set; In this embodiment, one implementation of step 504 is: Based on the order of importance of key image features and the index values ​​of image features of samples in the basic sample set, samples in important intervals corresponding to key image features are selected from the basic sample set in turn to form an assessment sample set for each key image feature.

[0021] In this embodiment, if the index values ​​of the image features of the screened samples cannot completely cover the important intervals, a sample generation method is used to fill in the samples corresponding to the missing index values ​​to obtain an assessment sample set that completely covers the important intervals.

[0022] Step 506, obtaining an optimal assessment sample set based on the assessment sample sets of all key image features.

[0023] In this embodiment, the image features include brightness, contrast, information entropy, standard deviation and gradient; the brightness is used to characterize the brightness level of the image; the contrast is used to characterize the difference between different brightness levels of the image; the information entropy is used to characterize the amount of information in the image; the standard deviation is used to characterize the brightness change of the image; the gradient is used to characterize the rate of change of pixels on the X-axis and Y-axis.

[0024] Specifically, refer to Figure 2 , a method for constructing an optimal assessment sample set of intelligent perception guided by image features, comprising the following steps: The first step is to design and construct image feature evaluation indicators An evaluation index consisting of image features such as brightness, contrast, information entropy, standard deviation and gradient was designed and constructed to analyze the correlation between samples and the performance of intelligent perception algorithms.

[0025] (1) Brightness is usually related to the image grayscale value, which represents the brightness level or intensity of the image pixels.

[0026] (2) Contrast refers to the degree of difference between different brightness levels in an image. The contrast depends on the distribution of pixel values.

[0027] (3) Information entropy is an indicator used to measure the amount of image information, indicating the uncertainty or randomness of pixel values. Its calculation formula is as follows:

[0028] in, Represents an image, p(i) represents the pixel value The higher the information entropy value, the more information the image has.

[0029] (4) Standard deviation is a statistic used to measure the brightness change of an image, reflecting the discrete degree of the image pixel value. Its calculation formula is as follows:

[0030] in, represents the standard deviation, represents the pixel value, represents the average value of pixel values, Indicates the total number of pixels.

[0031] (5) Gradient refers to the rate of change of a pixel in an image in both the X-axis and Y-axis directions. The image gradient is a two-dimensional vector composed of the change components of the X-axis and Y-axis. Its calculation formula is as follows:

[0032] in, and Respectively represent the images in The first-order partial derivatives along the X-axis and Y-axis.

[0033] The second step is to quantitatively analyze the impact of image features on algorithm performance and determine the key image features Collect and obtain the basic sample set covering the image features listed in the first step, and calculate the value of each image feature index. For each image feature index, use the "equal amount principle" to construct the corresponding training sample set by random sampling, train the algorithm model separately, and obtain different algorithm performances on the same test set. By sorting the algorithm performance, find out the key image features that have a significant impact on the intelligent perception algorithm.

[0034] The specific process is as follows Figure 3 As shown: (1) For the basic data set, divide it into training set and test set according to a fixed ratio (such as 4:1); (2) Calculate the brightness, contrast, information entropy, standard deviation, and gradient indicators of the training set; (3) For each image feature, the training set is divided into three interval subsets according to the calculation results of the index value range, and a random sample is selected from each interval subset. training samples, forming a training set of m samples representing the corresponding image features; For example, for the contrast index, the calculation result of the contrast index value range of the training set is [0, 1], then it is divided into three equal value intervals, namely [0, 0.33], [0.33, 0.66], and [0.66, 1].

[0035] (4) Use the training set of each indicator to train the algorithm model and test it on the same test set; (5) Sort the performance of each indicator obtained and select the sample quality evaluation indicators that have a greater impact on the performance of the intelligent perception algorithm.

[0036] The third step is to quantitatively analyze the impact of samples in different intervals on algorithm performance and find out the important intervals For each key image feature found in the second step, the image feature value is calculated and further divided into multiple intervals, and the samples are divided into corresponding intervals to form an interval training set. The algorithm model is trained using the interval training set, and the algorithm performance is measured on the same test set. The important interval range is determined based on the interval performance.

[0037] The specific process is as follows Figure 4 As shown: (1) Count the calculation results of the value range of each key image feature, and divide the training data set into multiple index intervals. For example, if the value range of the key image feature index is [a1, a2], divide the index value range into N equal intervals, and divide each sample into the corresponding interval according to its image feature value. The value of N can be appropriately adjusted according to the number of samples to ensure that a sufficient number of samples for model training can be randomly extracted from each divided interval; (2) For each key image feature, n samples are randomly selected from each interval to form the interval training set; (3) For each key image feature, use the training sets corresponding to different intervals to train the algorithm model, and test it on the same test set to obtain the algorithm performance of multiple intervals of the key image feature; (4) Sort the algorithm performance by interval and determine the important intervals of each key image feature that have a significant impact on the algorithm performance.

[0038] Step 4: Construct the optimal assessment sample set According to the key image features and important intervals found in the second and third steps, the corresponding samples are selected from the basic sample set according to their corresponding important intervals in the order of importance of the key evaluation indicators as the test set samples. For the missing samples in the important interval, the sample generation based on image features is used to fill the missing samples to ensure that the constructed test sample set covers the important intervals of the key features.

[0039] For example, the importance of key feature indicators is contrast>information entropy>average gradient>standard deviation, and the corresponding important intervals are [0.20,0.44], [5.1,6.0], [10,29], [10,14]. First, select samples with contrast in the interval [0.20,0.44] from the basic sample set as the test set samples. If the selected samples cannot fully cover the contrast interval [0.20,0.44], the sample generation technology is used to fill in the missing samples in the contrast interval to ensure that the test set meets the contrast interval requirements of [0.20,0.44]. Then, select samples with information entropy in the interval [5.1,6.0] from the basic sample set. If the selected samples cannot fully cover the information entropy interval [5.1,6.0], the sample generation technology is also used to fill in the missing samples in the information entropy interval to ensure that the test set samples meet the information entropy interval requirements of [0.20,0.44]. By analogy, the assessment set covers important intervals of average gradient and standard deviation.

[0040] In summary, this method designs image feature indicators to guide the construction of assessment sample sets, analyzes the impact of image features on algorithm performance through ablation test method, finds out key image feature indicators, and further finds out the important interval range of each image feature through quantitative calculation. According to the importance of key features, the sample set is supplemented based on the corresponding important intervals to form the optimal assessment sample set, which effectively reduces the number of assessment sample sets. Example 2

[0041] Please refer to Figure 5 , Figure 5 The figure shows an image feature-guided intelligent perception optimal assessment sample set construction device provided in this embodiment. In this embodiment, the device includes: The key image feature screening module is used to train the algorithm model using the basic sample set, obtain the influence of each image feature on the performance of the algorithm model and screen out the key image features; An index interval division module is used to calculate the index value range of each key image feature and divide the index interval; The important interval screening module is used to randomly select a number of samples from each indicator interval to form an interval training set for algorithm model training, obtain the influence of each indicator interval on the performance of the algorithm model and screen out important intervals; The optimal assessment set construction module is used to select the optimal assessment sample set from the basic sample set based on key image features and their corresponding important intervals.

[0042] Optionally, the key image feature screening module includes: A sample division unit, used to divide the basic sample set into a training set and a test set according to a preset ratio; An indicator value range calculation unit is used to train the algorithm model using a training set and calculate the indicator value range of each image feature; An image feature training set construction unit, used to construct an image feature training set for each image feature based on an index value range and an equal amount principle of each image feature; A first algorithm model training unit, used to train the algorithm model using the image feature training set to obtain a trained first algorithm model; A first algorithm model testing unit, used to test the trained first algorithm model using a test set to obtain the degree of influence of the image feature on the performance of the algorithm model; The key image feature screening unit is used to sort the degree of influence of all image features on the performance of the algorithm model and screen out the key image features that have a greater influence on the performance of the algorithm model.

[0043] Optionally, the image feature training set construction unit includes: An interval division subunit is used to divide the index value range of each image feature into three intervals with equal quantity; An interval subset division subunit, used to divide the samples of the training set into corresponding intervals based on the index value of the image feature of each sample in the training set, so as to obtain interval subsets; The image feature training set construction subunit is used to randomly extract one-third of the training samples from each interval subset to form a corresponding image feature training set.

[0044] Optionally, the indicator interval division module includes: A key image feature index value range acquisition unit, used to obtain the index value range of the key image feature based on the index value range of each image feature; The key image feature index value range division unit is used to divide the index value range of the key image feature into a number of index intervals in equal amounts.

[0045] Optional, important interval screening modules include: An index interval subset establishing unit, used to divide the samples in the training set into corresponding index intervals based on the index value of the image feature of each sample in the training set, so as to obtain an index interval subset; An interval training set establishment unit is used to randomly select a number of samples from each indicator interval subset to form a corresponding interval training set; A second algorithm model training unit, used to train the algorithm model using the interval training set to obtain a trained second algorithm model; A second algorithm model testing unit, used to test the second algorithm model using the test set to obtain the influence of the interval training set on the performance of the algorithm model; The important interval screening unit is used to sort the influence of the interval training set on the performance of the algorithm model and screen out the important intervals with greater influence on the performance of the algorithm model.

[0046] Optionally, the optimal assessment set building blocks include: A key image feature sorting unit, used to obtain the order of importance of the key image features based on the sorting of the key image features; An assessment sample set screening unit is used to screen out an assessment sample set for each key image feature from a basic sample set based on the order of importance of the key image features and the important interval corresponding to each key image feature; The optimal assessment sample set construction unit is used to obtain the optimal assessment sample set based on the assessment sample set of all key image features.

[0047] Optionally, an assessment sample set screening unit is used to screen out samples in important intervals corresponding to key image features from the basic sample set in turn based on the order of importance of key image features and the index values ​​of the image features of the samples in the basic sample set, to form an assessment sample set for each key image feature.

[0048] Optionally, if the index values ​​of the image features of the screened samples cannot completely cover the important intervals, a sample generation method is used to fill in the samples corresponding to the missing index values ​​to obtain an assessment sample set that completely covers the important intervals.

[0049] Optionally, the image features include brightness, contrast, information entropy, standard deviation and gradient; the brightness is used to characterize the brightness level of the image; the contrast is used to characterize the difference between different brightness levels of the image; the information entropy is used to characterize the amount of information in the image; the standard deviation is used to characterize the brightness change of the image; the gradient is used to characterize the rate of change of pixels on the X-axis and Y-axis.

[0050] Based on this, this device designs image feature indicators to guide the construction of assessment sample sets, analyzes the impact of image features on algorithm performance through ablation test method, finds out key image feature indicators, and further finds out the important interval range of each image feature through quantitative calculation. According to the importance of key features, the sample set is supplemented based on the corresponding important intervals in turn to form the optimal assessment sample set, which effectively reduces the number of assessment sample sets. Example 3

[0051] Please refer to Figure 6 , this embodiment provides an electronic device, which includes a processor, an internal bus, a network interface, a memory and a non-volatile memory, and of course may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it, forming a method for constructing an optimal assessment sample set for intelligent perception guided by image features at the logical level. Of course, in addition to software implementation methods, this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0052] The network interface, processor and memory can be connected to each other through a bus system. The above bus can be divided into an address bus, a data bus, a control bus, etc.

[0053] The memory is used to store programs. Specifically, the program may include program code, and the program code includes computer operation instructions. The memory may include a read-only memory and a random access memory, and provides instructions and data to the processor.

[0054] The processor is used to execute the program stored in the above memory, and specifically perform: Step 102, using the basic sample set to train the algorithm model, obtain the influence of each image feature on the performance of the algorithm model and screen out key image features; Step 104, calculating the index value range of each key image feature and dividing the index interval; Step 106, randomly extracting a number of samples from each index interval to form an interval training set for algorithm model training, obtaining the influence of each index interval on the performance of the algorithm model and screening out important intervals; Step 108 , based on the key image features and their corresponding important intervals, the optimal assessment sample set is selected from the basic sample set.

[0055] The processor may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above method may be completed by an integrated logic circuit of the processor hardware or by instructions in the form of software.

[0056] Based on the same invention, the embodiment of this specification also provides a computer-readable storage medium, which stores one or more programs. When the one or more programs are executed by an electronic device including multiple application programs, the electronic device executes Figure 1-Figure 4 The corresponding embodiment provides a method for constructing an optimal assessment sample set of intelligent perception guided by image features.

[0057] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-readable storage media containing computer-usable program code.

[0058] In addition, for the above-mentioned specific implementation of the system, since it is basically similar to the method implementation, the description is relatively simple, and the relevant parts can be referred to the partial description of the method implementation. Moreover, it should be noted that in each module of the system of the present application, the components therein are logically divided according to the functions to be implemented, but the present application is not limited thereto, and the components can be re-divided or combined as needed.

[0059] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other. Each embodiment focuses on the differences between it and other embodiments.

[0060] The above is a description of a specific embodiment of the present specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the specific order or sequential order shown in the process depicted in the drawings is not necessarily required to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for constructing an optimal assessment sample set of intelligent perception guided by image features, characterized in that: include: The algorithm model is trained using a basic sample set to obtain the degree of influence of each image feature on the performance of the algorithm model and to screen out key image features; Calculate the index value range of each key image feature and divide the index interval; Randomly select a number of samples from each index interval to form an interval training set for algorithm model training, obtain the impact of each index interval on the performance of the algorithm model and screen out important intervals; Based on key image features and their corresponding important intervals, the optimal assessment sample set is selected from the basic sample set.

2. The method according to claim 1, characterized in that The algorithm model is trained using the basic sample set to obtain the influence of each image feature on the performance of the algorithm model and screen out key image features, including: Divide the basic sample set into a training set and a test set according to a preset ratio; The algorithm model is trained using the training set to calculate the value range of the index for each image feature; Based on the index value range and equal principle of each image feature, an image feature training set for each image feature is constructed; The algorithm model is trained using the image feature training set to obtain a first algorithm model after training; Using the test set to test the trained first algorithm model, to obtain the degree of influence of the image feature on the performance of the algorithm model; The influence of all image features on the performance of the algorithm model is ranked, and the key image features with the greatest influence on the performance of the algorithm model are screened out.

3. The method according to claim 2, characterized in that The image feature training set for each image feature is constructed based on the index value range and the equal principle of each image feature, including: Divide the index value range of each image feature into three intervals equally; Divide the samples of the training set into corresponding intervals based on the index value of the image feature of each sample in the training set to obtain interval subsets; One third of the training samples are randomly selected from each interval subset to form the corresponding image feature training set.

4. The method according to claim 3, characterized in that The step of calculating the index value range of each key image feature and dividing the index intervals includes: Based on the index value range of each image feature, obtain the index value range of the key image feature; The index value range of the key image features is equally divided into several index intervals.

5. The method according to claim 4, characterized in that The method of randomly extracting a number of samples from each index interval to form an interval training set for algorithm model training, obtaining the influence of each index interval on the performance of the algorithm model and screening out important intervals includes: Based on the index value of the image feature of each sample in the training set, the samples of the training set are divided into corresponding index intervals to obtain the index interval subset; Randomly select several samples from each indicator interval subset to form the corresponding interval training set; The algorithm model is trained using the interval training set to obtain a trained second algorithm model; The second algorithm model is tested using the test set to obtain the degree of influence of the interval training set on the performance of the algorithm model; The influence of the interval training set on the performance of the algorithm model is sorted, and the important intervals with greater influence on the performance of the algorithm model are screened out.

6. The method according to claim 1, characterized in that The method of selecting the optimal assessment sample set from the basic sample set based on the key image features and their corresponding important intervals includes: Based on the sorting of key image features, the order of importance of the key image features is obtained; Based on the order of importance of key image features and the important interval corresponding to each key image feature, the assessment sample set of each key image feature is screened out from the basic sample set; Based on the assessment sample set of all key image features, the optimal assessment sample set is obtained.

7. The method according to claim 6, characterized in that The method of selecting an assessment sample set for each key image feature from a basic sample set based on the order of importance of key image features and the important interval corresponding to each key image feature includes: selecting samples in the important interval corresponding to the key image feature from the basic sample set in sequence based on the order of importance of key image features and the index values ​​of the image features of the samples in the basic sample set to form an assessment sample set for each key image feature.

8. The method according to claim 7, characterized in that If the index values ​​of the image features of the screened samples cannot completely cover the important intervals, the sample generation method is used to fill in the samples corresponding to the missing index values ​​to obtain an assessment sample set that completely covers the important intervals.

9. The method according to claim 1, characterized in that: The image features include brightness, contrast, information entropy, standard deviation and gradient; the brightness is used to characterize the brightness level of the image; the contrast is used to characterize the difference between different brightness levels of the image; the information entropy is used to characterize the amount of information in the image; The standard deviation is used to characterize the brightness change of the image; the gradient is used to characterize the change rate of the pixel on the X-axis and the Y-axis.

10. An image feature-guided intelligent perception optimal assessment sample set construction device, characterized in that: include: The key image feature screening module is used to train the algorithm model using the basic sample set, obtain the influence of each image feature on the performance of the algorithm model and screen out the key image features; An index interval division module is used to calculate the index value range of each key image feature and divide the index interval; The important interval screening module is used to randomly select a number of samples from each indicator interval to form an interval training set for algorithm model training, obtain the influence of each indicator interval on the performance of the algorithm model and screen out important intervals; The optimal assessment set construction module is used to select the optimal assessment sample set from the basic sample set based on key image features and their corresponding important intervals.

Images