Human-oriented fine-grained visual target recognition adaptive training scheme generation, difficulty adjustment, closed-loop feedback training method and system

Abstract

The embodiment of the present invention discloses a method and system for generating a human-oriented fine-grained visual target recognition adaptive training program, adjusting difficulty, and closed-loop feedback training. The method performs an indefinite selection recognition test and a personal recognition trait test on a target in a fine-grained visual target, generates a training program based on the personal recognition trait test scores of all trainees, the personal recognition trait test score of the current trainee, and the standard fine-grained visual target indefinite selection recognition test score; performs observation learning training and feedback learning training in sequence; performs a standard fine-grained visual target indefinite selection recognition test; determines whether the trainee's standard fine-grained visual target indefinite selection recognition test score is greater than 0.95, and if so, terminates the training; otherwise, adjusts the training difficulty for the next training program until the trainee's standard fine-grained visual target indefinite selection recognition test score is greater than 0.95 and then terminates the training, thereby effectively improving the training efficiency.

Description

Technical Field

[0001] The present invention belongs to the technical field of human performance enhancement, and relates to a method and system for generating a fine-grained visual target recognition adaptive training program for humans, adjusting difficulty, and conducting closed-loop feedback training. Background Art

[0002] Humans are good at identifying and distinguishing different types of targets in complex visual environments (such as identifying tables and chairs in a classroom). This basic object recognition ability is the basis for human survival and social interaction, but it is not innate, but acquired through acquired learning. Fine-grained target recognition refers to finding targets from similar and homogeneous visual stimuli through fine visual discrimination (such as finding a certain model of vehicle among many vehicles). This skill is also not innate, but needs to be acquired through long-term passive exposure to specific targets in natural scenes (such as recognizing faces in daily life), or through a shorter period of deliberate practice under constraints (such as radiologists identifying lesions in medical images).

[0003] With the development of artificial intelligence technology, especially computer vision technology, machine-assisted visual target recognition technology has made great progress and has been promoted and applied in the fields of medicine, public safety and national defense, such as computer-assisted medical image diagnosis and artificial intelligence-guided security inspection of dangerous goods. However, due to constraints such as the size of modeling samples, scene complexity, and task openness, artificial intelligence methods will not be the main solution for fine-grained object recognition now and for a long time in the future. People are still the core and final link in the decision-making chain. With the exponential growth of information flow, the demand for personnel with fine-grained target recognition capabilities in different positions has doubled.

[0004] In the face of the demand for personnel training in the fields of medicine, public security, and national defense, existing solutions for the training and training of people's fine-grained target recognition capabilities use two modes: "human-human interaction" mode and "human-computer interaction" mode: "human-human interaction" training adopts sample teaching and typical case analysis methods; "human-computer interaction" training mode conducts homogeneous and open-loop training for the trainees, resulting in low training efficiency, long training time, and large individual differences in training effects. In short, the existing training methods and system training efficiency are insufficient, resulting in a serious disconnect between the supply of qualified personnel and the exponentially growing demand for personnel. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method and system for generating adaptive training programs, adjusting difficulty, and closed-loop feedback for human-oriented fine-grained visual target recognition, so as to solve the problems existing in the current training and training process of fine-grained visual target recognition personnel, such as large individual differences in training effects, single training methods, and low training efficiency.

[0006] The first technical solution adopted by the embodiment of the present invention is: a method for generating a human-oriented fine-grained visual target recognition adaptive training program, which is performed according to the following steps:

[0007] Step S1, generating a comprehensive training index Z based on the individual recognition trait test score and the current standard fine-grained visual target indefinite item selection recognition test score;

[0008] Step S2: Obtain all training data of all trainees and generate the i-th training parameter X based on the multivariate linear regression model. i The weight β i ;

[0009] Step S3: Based on the generated comprehensive training index Z and the estimated weight β i , combined with the training scheme generation algorithm expressed in the following formula, the training parameters X that determine the training scheme are generated i , and according to the training parameters X i Set up training programs;

[0010]

[0011] Among them, β min is the minimum value of the training weight in this training scheme, β max is the maximum value of the training weight in this training scheme; the training parameter X i There are 6 items in total, namely i=1, 2, 3, 4, 5, 6, among which X1 is the picture stimulation time of the trainees in the observational learning training, X2 is the time when the trainees can end the picture stimulation early in the observational learning training, X3 is the total number of fine-grained visual target sub-images for the trainees to receive picture stimulation in the observational learning training, X4 is the total number of fine-grained visual target sub-images for the picture stimulation in the feedback learning training, X5 is the time it takes for the trainees to make choices according to the prompts after seeing the picture stimulation in the feedback learning training, and X6 is the image difficulty of a single fine-grained visual target sub-image used in the feedback learning training.

[0012] Furthermore, the step S1 generates a comprehensive training index Z according to the following formula:

[0013]

[0014] Among them, exp is the exponential function, The results of the previous standard fine-grained visual object indeterminate item selection recognition test, A s The expected next standard fine-grained visual target indefinite item selection recognition test result can be manually input by the trainee, K m is the personal identification coefficient of the current trainee obtained based on the personal identification trait test scores;

[0015] The personal identification coefficient K of the current trainee m Obtained by the following method:

[0016] Obtain the personal identification trait test scores of all trainees and the personal identification trait test score of the current trainee, use the minimum and maximum values ​​of the personal identification trait test scores of all trainees to form a personal identification trait test score interval, then normalize the personal identification trait test score interval to the interval [0,1], and then normalize the personal identification trait test score of the current trainee to the interval [0,1] to obtain the personal identification coefficient K m ;

[0017] The image difficulty of the fine-grained visual target subgraph is obtained by normalizing the accurate recognition time spent by a fine-grained visual target recognition expert on accurate recognition test of each fine-grained visual target subgraph, mapping the accurate recognition time of all fine-grained visual target subgraphs to a difficulty coefficient in the interval of (0,1), and the difficulty coefficient obtained by mapping each fine-grained visual target subgraph is the image difficulty of the fine-grained visual target subgraph.

[0018] Furthermore, the specific implementation process of step S2 is as follows:

[0019] First, the multiple linear regression equation is established according to the following formula:

[0020]

[0021] In the above formula, Y is the predicted score of the trainee's next standard fine-grained visual object indeterminate item selection recognition test, β0 is a constant term, and e is the residual; β i is the training weight, is the i-th training parameter X in the training scheme i The weight of

[0022] Then, using the established multiple linear regression equation, the residual sum of squares Q is minimized by the least squares method according to the following formula:

[0023]

[0024] Among them, Y k is the standard fine-grained visual object indeterminate item choice recognition test score after the k-th training of the current trainee, Y k The estimated value of X ik is the i-th training parameter in the k-th training of the current trainee, β ik is the weight of the i-th training parameter in the k-th training of the current trainee, and n is the total number of training data obtained.

[0025] The second technical solution adopted by the embodiment of the present invention is: a method for adjusting the difficulty of adaptive training of fine-grained visual target recognition for humans, which is performed according to the following steps:

[0026] Step 1: construct an ordered logistic regression model based on the generated fine-grained visual object recognition training plan for the trainee;

[0027] Step 2: According to the ordered logistic regression model, the probability P of the trainee reaching level g in the standard fine-grained visual target indefinite choice recognition test after this training is calculated according to the following formula: g :

[0028]

[0029] Among them, P g-1 is the probability that the trainee's standard fine-grained visual target indefinite item selection recognition test score reaches level g-1 after this training, exp is an exponential function, β 0g-1 is the constant term of the g-1th regression; X i To determine the training parameters of the training scheme, β i is the training weight;

[0030] Step 3: Determine the probability P that the trainee's standard fine-grained visual target indefinite item choice recognition test score reaches level g after this training g Does it meet the requirements? If so, directly output the training parameters X i And generate a training plan for training, otherwise adjust the training difficulty;

[0031] Step 4: Adjust the training difficulty.

[0032] Furthermore, the specific implementation process of step 1 is as follows:

[0033] Obtain all training data from the training database, put all training data into a decision tree for classification, and perform interval classification on the trainee's standard fine-grained visual object indeterminate choice recognition test score Y before each training. The interval step is 0.1, and all training data are divided into g different data sets, i.e. g levels;

[0034] Construct an ordered logistic regression model as follows:

[0035]

[0036] Where P(Y≤j) is the probability that the trainee's standard fine-grained visual target indeterminate item choice recognition test score reaches level Y, j = 1, 2, 3...g-1, ln is an exponential function;

[0037] The specific implementation process of step 4 for adjusting the training difficulty is as follows:

[0038] The advance and retreat method in the precise one-dimensional search in gradient descent is used to re-evaluate the Z value, and the new Z value obtained by re-evaluation is used to regenerate the training scheme, that is, to regenerate the training parameters X. i , until the probability P of the trainee reaching level g in the standard fine-grained visual target indefinite item selection recognition test after this training g Meet the requirements and output the corresponding training plan, i.e. training parameters X i .

[0039] The third technical solution adopted by the embodiment of the present invention is: a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method, which is performed according to the following steps:

[0040] S1: Conduct a standard fine-grained visual target indefinite item selection recognition test and a personal identification trait test on the trainee, and obtain the trainee's personal identification trait test score and the current standard fine-grained visual target indefinite item selection recognition test score;

[0041] S2: according to the personal identification trait test scores of all trainees, the personal identification trait test scores of the trainees in S1, and the current standard fine-grained visual target indefinite choice recognition test scores, a training plan is generated by using the method for generating a training plan for human-oriented fine-grained visual target recognition as described in any one of claims 1 to 3;

[0042] S3: Based on the generated training plan, the trainees are trained in observation learning and feedback learning in sequence;

[0043] S4: The trainees were given a standard fine-grained visual target indeterminate item choice recognition test;

[0044] S5: Determine whether the trainee's standard fine-grained visual target indeterminate item choice recognition test score in S4 is greater than 0.95. If so, end the training; otherwise, enter S6;

[0045] S6: Using the method for adjusting the training difficulty of human-oriented fine-grained visual target recognition as described in any one of claims 4 to 5, adjust the training difficulty of the trainee's next training plan, regenerate the training plan, and repeat S3 to S5 until the trainee's standard fine-grained visual target indefinite choice recognition test score is greater than 0.95, then end the training.

[0046] Furthermore, the process of the personal identification trait test is as follows: the trainee memorizes a group of 6 objects under fine-grained visual targets within a specified time. Each object will be presented in the form of 3 fine-grained visual targets from different perspectives, and each fine-grained visual target is presented for 3 seconds; then, the trainee selects the 6 objects in memory from the objects under another group of fine-grained visual targets. After the selection, no feedback is provided, and the trainee's actual selection is automatically compared with the correct answer, and the accuracy rate is output as the personal identification trait test score;

[0047] The process of the standard fine-grained visual target indefinite selection recognition test is as follows: in each standard fine-grained visual target test image, the trainee selects the target object in turn. The selection time for each target object is limited to 10 seconds. After each target object is selected, the trainee will choose "continue to select the next target object" or "I have selected all the targets". When the trainee chooses "continue to select the next target object", the trainee repeats the selection process until all the targets in the nine-square grid image are selected and "I have selected all the targets" is selected. No feedback is provided after the selection.

[0048] In the single observation learning training, the trainees will first see a picture name label that lasts for 1 second, which is the recognition result of the fine-grained visual target by the fine-grained target recognition expert. Then they will see a plus sign that lasts for 500ms. After the plus sign that lasts for 500ms ends, the trainees will see a picture stimulus that lasts for 6-18 seconds, which is determined by the training plan of the training session. The trainees can choose to end it early within a certain time range.

[0049] In the single-feedback learning training, after the 500ms plus sign, the trainees will see a fine-grained visual target image, i.e., a picture stimulus, which is determined by the current training plan and lasts for 5-15s and is composed of multiple fine-grained visual target sub-pictures. During this period, the trainees need to select a fine-grained visual target sub-picture that meets the requirements from the fine-grained visual target composed of multiple fine-grained visual target sub-pictures according to text prompts. During the picture stimulation, the trainees will be prompted with text about the remaining selection time, and a plus sign with a duration of 500ms will be displayed after the trainees have made their choice or after the picture stimulation time has ended. After the plus sign with a duration of 500ms is displayed, result feedback will be given based on the trainees' choice, and the trainees' choice of the fine-grained visual target sub-picture made according to the text prompts will be fed back as correct or incorrect.

[0050] The fourth technical solution adopted by the embodiment of the present invention is: a human-oriented fine-grained visual target recognition adaptive training program generation system, comprising:

[0051] a memory for storing instructions executable by a processor;

[0052] A processor is used to execute the instructions to implement the method for generating a human-oriented fine-grained visual target recognition adaptive training solution as described above.

[0053] The fifth technical solution adopted by the embodiment of the present invention is: a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system, comprising:

[0054] a memory for storing instructions executable by a processor;

[0055] A processor is used to execute the instructions to implement the human-oriented fine-grained visual target recognition adaptive training difficulty adjustment method as described above.

[0056] The sixth technical solution adopted by the embodiment of the present invention is: a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system, comprising:

[0057] a memory for storing instructions executable by a processor;

[0058] A processor is used to execute the instructions to implement the human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method as described above.

[0059] The beneficial effects of the embodiments of the present invention are: a closed-loop feedback training system based on individual recognition level, a training program formulated according to the personal recognition characteristics of the trainee, and a closed-loop feedback training method combining observational learning and feedback learning are used to preliminarily improve the training efficiency of the trainee; a data-driven closed-loop feedback training difficulty adjustment algorithm is designed, including the generation of individual-based training programs and the feasibility judgment of training programs through an ordered logistic regression model, taking full account of individual recognition levels, dynamically feeding back test results after each training, adjusting the training program and difficulty, selecting the optimal training program, and further effectively improving the training efficiency. It can solve the problems of large individual differences in training effects, single training methods, and low training efficiency in the current training and training process of fine-grained visual target recognition personnel.

[0060] Synthetic Aperture Radar (SAR) is an active earth observation system deployed on long-range high-altitude detection platforms such as aircraft, space satellites, and various types of drones. By utilizing its synthetic aperture principle, high-resolution and wide-bandwidth microwave imaging can be achieved. These advantages make SAR highly valuable in the fields of earth mapping, military reconnaissance, and geological exploration. Fine-grained visual targets are determined by their own imaging characteristics, which are visually quite different from our common optical images. The information covered by the image is also different from that of the optical image, and ordinary people without experience cannot visually identify the target. The human-oriented fine-grained visual target recognition adaptive training scheme generation, difficulty adjustment, and closed-loop feedback training method and system of the embodiment of the present invention can realize the training and training of SAR image recognition personnel, and can be expanded to the training and training of personnel in positions such as radiologists and traffic station security inspectors who need fine-grained visual target recognition skills. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, 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 work.

[0062] Figure 1 It is a flowchart of an adaptive closed-loop feedback training method for fine-grained visual target recognition for humans according to an embodiment of the present invention.

[0063] Figure 2 It is a flow chart of an adaptive training difficulty adjustment algorithm for fine-grained visual target recognition for humans driven by large sample data according to an embodiment of the present invention.

[0064] Figure 3 It is a schematic diagram of the training process of observation learning and feedback learning in the training module of an embodiment of the present invention.

[0065] Figure 4 It is a schematic diagram of the system interface of feedback learning in the training module of an embodiment of the present invention.

[0066] Figure 5 It is a comparison chart of the actual measured training results of the adaptive closed-loop feedback training method for fine-grained visual target recognition for humans according to an embodiment of the present invention and the traditional open-loop training method. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Example 1

[0069] The embodiment of the present invention provides a method for generating a fine-grained visual target recognition adaptive training program for humans, such as Figures 1-2 As shown, follow these steps:

[0070] Step S1: Based on the personal identification trait test score and the current standard fine-grained visual target indefinite item selection recognition test score, generate a comprehensive training index Z according to the following formula:

[0071]

[0072] Among them, exp is the exponential function, The results of the previous standard fine-grained visual object indeterminate item selection recognition test, A s The expected next standard fine-grained visual target indefinite item selection recognition test result can be manually input by the trainee, K m is the personal identification coefficient of the current trainee obtained based on the personal identification trait test scores.

[0073] The comprehensive training index Z combines the trainee's visual recognition ability of objective objects and the recognition ability of standard fine-grained visual targets in the current training stage. The higher the value, the more effort the trainee needs to make.

[0074] The personal identification coefficient K of the current trainee m Obtained by the following method:

[0075] Obtain the personal identification trait test scores of all trainees and the personal identification trait test score of the current trainee, use the minimum and maximum values ​​of the personal identification trait test scores of all trainees to form a personal identification trait test score interval, then normalize the personal identification trait test score interval to the interval [0,1], and then normalize the personal identification trait test score of the current trainee to the interval [0,1] to obtain the personal identification coefficient K mIf the minimum and maximum values ​​of the personal identification trait test scores of all trainees are 10 and 30, and the personal identification trait test score of the current trainee is 20, when the interval [10,30] consisting of the minimum and maximum values ​​of the personal identification trait test scores of all trainees is normalized to the interval [0,1], the personal identification trait test score 20 of the current trainee is mapped to the interval [0,1], and the obtained personal identification coefficient K is m is 0.5.

[0076] Step S2: Obtain all training data of all trainees and generate the i-th training parameter X based on the multivariate linear regression model. i The weight β i The specific implementation process is as follows:

[0077] First, the multiple linear regression equation is established according to the following formula:

[0078]

[0079] Among them, Y is the predicted score of the trainee's next standard fine-grained visual object indeterminate item choice recognition test, β0 is a constant term, and e is the residual; β i is the training weight, is the i-th training parameter X in the training scheme X value i The weights of , i = 1, 2, 3, 4, 5, 6;

[0080] The training scheme X consists of training parameters X i Composition, training parameters X i There are 6 items in total. These 6 training parameters jointly determine the training plan of this training. Among them, X1 is the picture stimulation time of the trainees in observational learning training, X2 is the time when the trainees can end the picture stimulation in advance in observational learning training, X3 is the total number of fine-grained visual target sub-images that the trainees receive picture stimulation in observational learning training. When the picture is stimulated, generally an image name label is displayed, and a fine-grained visual target sub-image is observed. For high-order and difficult images, multiple image name labels are displayed, and multiple fine-grained visual target sub-images are observed. X4 is the total number of fine-grained visual target sub-images stimulated by the picture in feedback learning training, X5 is the time it takes for the trainees to make a choice according to the prompt after seeing the picture stimulation in feedback learning training, and X6 is the image difficulty of a single fine-grained visual target sub-image used in feedback learning training. When the feedback learning training uses a fine-grained visual target image composed of multiple fine-grained visual target sub-images, the image difficulty of all fine-grained visual target sub-images in the fine-grained visual target image is consistent. The fine-grained visual target sub-images and their image difficulty are obtained from the training image library.

[0081] Then, using the established multiple linear regression equation, the residual sum of squares Q is minimized by the least squares method according to the following formula:

[0082]

[0083] Among them, Y k is the standard fine-grained visual object indeterminate item choice recognition test score after the k-th training of the current trainee, Y k The estimated value of i The average value of Y when taking the value, X ik is the i-th training parameter in the k-th training of the current trainee, β ik is the weight of the i-th training parameter in the k-th training of the current trainee, and n is the total number of training data obtained.

[0084] Step S3: Based on the generated comprehensive training index Z and the estimated training weight β i , combined with the training scheme generation algorithm expressed in the following formula, the training parameters X that determine the training scheme are generated i , and according to the training parameters X i Set up a training plan:

[0085]

[0086] Among them, β min is the minimum value of the training weight in this training scheme, β max is the maximum value of the training weight in this training scheme.

[0087] The training scheme consists of the training parameters X i Determine, and train the parameters X i The comprehensive training index Z and the training weight β i The comprehensive training index Z is determined by the large sample personal identification characteristics in the training database, namely the personal identification coefficient K m Determined by the training parameter X i The weight β i Determined by the accumulated training data in the training database, the training plan is generated by integrating historical training data and personal identification characteristics, making the resulting training plan more objective and scientific.

[0088] Example 2

[0089] The embodiment of the present invention also provides a method for adjusting the difficulty of adaptive training of fine-grained visual target recognition for humans. Figures 1-2 As shown, follow these steps:

[0090] Step 1: According to the training parameters X i The trainees composed this training program and constructed an ordered logistic regression model:

[0091] All training data are obtained from the training database, and all training data are put into the decision tree for classification. The decision tree performs interval classification on the trainee's standard fine-grained visual target indefinite choice recognition test score Y before each training, with an interval step size of 0.1. All training data are divided into g different data sets, i.e. g levels, as the training set of the ordered logistic regression model, serving the g levels of the ordered logistic regression model. The ordered logistic regression model is based on the cumulative probability to build a regression model. The independent variable Y is a rank variable. The trainee's standard fine-grained visual target indefinite choice recognition test score Y after a single training is divided into g levels (Y takes values ​​of 1, 2, 3...g). The ordered logistic regression model is constructed according to the following formula:

[0092]

[0093] Wherein, j=1,2,3…g-1, ln is an exponential function, and P(Y≤j) is the probability that the trainee's standard fine-grained visual target indeterminate item choice recognition test score reaches level Y.

[0094] Step 2: According to the ordered logistic regression model, the probability P that the trainee's standard fine-grained visual target indefinite choice recognition test score reaches level g after this training can be calculated according to the following formula: g :

[0095]

[0096] Among them, P g-1 is the probability that the trainee's standard fine-grained visual target indefinite item selection recognition test score reaches level g-1 after this training, exp is an exponential function, β 0g-1 is the constant term of the g-1th regression.

[0097] Step 3: Determine the probability P that the trainee's standard fine-grained visual target indefinite item choice recognition test score reaches level g after this training g Whether it meets the requirements (such as whether it is greater than 80%), if so, directly output the training parameter X i The training module will generate a training plan for training, otherwise the training difficulty will be adjusted;

[0098] Step 4: Adjust the training difficulty:

[0099] The advance and retreat method in the precise one-dimensional search in gradient descent is used to re-evaluate the Z value, and the new Z value obtained by re-evaluation is used to regenerate the training scheme, that is, to regenerate the training parameters X. i , until the probability P of the trainee reaching level g in the standard fine-grained visual target indefinite item selection recognition test after this training gMeet the requirements and output the corresponding training plan, i.e. training parameters X i .

[0100] The training difficulty adjustment method of this embodiment provides a feasible training plan tailored to the individual and having mathematical and statistical significance according to the test results and personal characteristics of the trainee.

[0101] The specific steps of exact one-dimensional search in gradient descent are as follows:

[0102] Step 41: Select the initial point as the standard fine-grained visual target indefinite item selection recognition test score A after the training in step S1 s , search step h = 0.05, precision ε = 0.025, the adjustment function selects the Z value generating function in step S1, and calculates

[0103] Step 42: Calculate

[0104] Step 43: and As new Z values, regenerate the corresponding training parameters X i , and calculated by The ordered logistic regression model used as the Z value determines the probability that the standard fine-grained visual target indefinite choice recognition test score reaches level g after this training. and by The ordered logistic regression model used as the Z value determines the probability that the standard fine-grained visual target indefinite choice recognition test score reaches level g after this training. like Less than and If less than 80%, the search is successful and the process goes to step 44; Less than The search fails, and the process goes to step 45; at any time, When the requirement is met (greater than 80%), the corresponding training scheme, i.e., training parameter X, is output. i ;

[0105] Step 44: Set A s :=A s -h, h: = 2h, that is, first update the standard fine-grained visual target indefinite item selection recognition test score to be equal to the previous standard fine-grained visual target indefinite item selection recognition test score A s Subtract the search step size h, and then update A s Bring in renew Finally, the search step size is doubled and the search is continued to step 42. renew

[0106] Step 45, determine whether h satisfies |h|≤ε; if |h|≤ε, stop iteration, otherwise update h=-1 / 4h and go to step 42.

[0107] Example 3

[0108] The embodiment of the present invention also provides a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method, such as Figures 1-2 As shown, follow these steps:

[0109] S1: Conducting a fine-grained visual target recognition test and a personal recognition trait test on the trainee. Specifically, the fine-grained visual target recognition test is to conduct a standard fine-grained visual target indefinite choice recognition test on the trainee, and obtain the trainee's personal recognition trait test score and the current fine-grained visual target recognition test score (current standard fine-grained visual target indefinite choice recognition test score);

[0110] The personal identification trait test tests the trainee's ability to identify objects in general fields (which can be a personal talent). The personal identification trait test is only conducted when the trainee uses it for the first time. The personal identification trait test has a total of 54 multiple-choice questions, and the test results are output as the correct rate;

[0111] The standard fine-grained visual target indefinite selection recognition test tests the trainee's fine-grained visual target recognition ability. The standard fine-grained visual target test image is composed of nine fine-grained visual target sub-images spliced ​​in a nine-square grid manner. The score of the standard fine-grained visual target indefinite selection recognition test can be given by the score calculation method in the invention patent named "A method, system and computer-readable medium for quantifying human-oriented SAR image target visual recognition ability."

[0112] The process of the personal identification trait test is as follows: the trainee memorizes a group of 6 objects under fine-grained visual targets within a specified time. Each object will be presented in the form of 3 fine-grained visual target images from different perspectives, and each fine-grained visual target image is presented for 3 seconds. Then, the trainee selects the 6 objects in memory from the objects under another group of fine-grained visual target images and selects them through the keyboard (pressing the number keys) or the mouse (clicking to select when the display screen is a touch screen). No feedback is provided after the selection, and the trainee's actual selection is automatically compared with the correct answer, and the accuracy rate is output as the personal identification trait test score.

[0113] The test process of fine-grained visual target indefinite selection recognition is as follows: in each standard fine-grained visual target test image, the trainee selects the target objects in turn. The selection time limit for each target object is 10s. After selecting each target object, the trainee will choose "continue to select the next target object" or "I have selected all targets". When the trainee chooses "continue to select the next target object", the trainee repeats the selection process until all targets in the nine-square grid image are selected (the number of targets is 1-9) and selects "I have selected all targets". The standard fine-grained visual target indefinite selection recognition test has a total of 20 questions. The entire test takes about 8-10 minutes. No feedback is provided after the selection.

[0114] S2: generating a training plan according to the method for generating a human-oriented fine-grained visual object recognition adaptive training plan described in Example 1, based on the personal recognition trait test scores of all trainees, the personal recognition trait test scores of the trainees in S1, and the current standard fine-grained visual object indefinite choice recognition test scores;

[0115] S3: Based on the generated training plan, the trainees are trained in observation learning and feedback learning in sequence;

[0116] like Figure 3 As shown, in the single observation learning training, the trainees will first see a picture name label lasting 1 second, which is the recognition result of the fine-grained visual target recognition expert on the fine-grained visual target image, and then see a plus sign lasting 500ms. After the plus sign lasting 500ms is displayed, the trainees will see a picture stimulus that lasts 6-18s and is determined by the training plan of the current training. The trainees can choose to end it early within a certain time range. In the single feedback learning training, after the plus sign of 500ms is displayed, the trainees will see a fine-grained visual target image composed of multiple fine-grained visual target sub-images, which is determined by the training plan of the current training and lasts for 5-15s. During this period, the trainees need to select the fine-grained visual target sub-image that meets the requirements from the fine-grained visual target image composed of multiple fine-grained visual target sub-images according to the text prompts, as shown in FIG. Figure 4 As shown, during the picture stimulation period, the trainees are prompted with text about the remaining selection time, and after the trainees have made their choice or the picture stimulation time has ended, a plus sign lasting 500ms is displayed. After the plus sign lasting 500ms is displayed, result feedback is given based on the trainees' choice, and the trainees' choice of the fine-grained visual target sub-image made according to the text prompt is fed back as correct or incorrect.

[0117] S4: The trainees were given a standard fine-grained visual target indeterminate item choice recognition test;

[0118] S5: Determine whether the trainee's standard fine-grained visual target indeterminate item choice recognition test score in S4 is greater than 0.95. If so, end the training; otherwise, enter S6;

[0119] S6: Adjust the training difficulty of the trainee's next training plan, regenerate the training plan, repeat S3 to S5, until the trainee's standard fine-grained visual target indefinite choice recognition test score is greater than 0.95, then end the training; the training difficulty of the trainee's next training plan can be adjusted by using the human-oriented fine-grained visual target recognition adaptive training difficulty adjustment method described in Example 2.

[0120] For the training of a certain type of tank recognition in SAR images of 10 samples, the measured training results of the adaptive closed-loop feedback training method for fine-grained visual target recognition for humans according to the embodiment of the present invention and the existing open-loop training method without feedback are as follows: Figure 5 As shown by Figure 5 It can be seen that the human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method ( Figure 5 The training effect of the embodiment method in the embodiment method is that the trainee's judgment accuracy of the SAR image increases with the increase of training days, and the training effect is good.

[0121] Example 4

[0122] An embodiment of the present invention also provides a system for generating an adaptive training program for fine-grained visual target recognition for humans, comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a method for generating an adaptive training program for fine-grained visual target recognition for humans as described in Example 1 above.

[0123] A system for generating a fine-grained visual target recognition adaptive training program for people may include an internal communication bus, a processor, a read-only memory (ROM), a random access memory (RAM), a communication port, and a hard disk. The internal communication bus can realize data communication between components of a system for generating a fine-grained visual target recognition adaptive training program for people. The processor can make judgments and issue prompts. In some embodiments, the processor can be composed of one or more processors. The communication port can realize data communication outside a system for generating a fine-grained visual target recognition adaptive training program for people. In some embodiments, a system for generating a fine-grained visual target recognition adaptive training program for people can also send and receive information and data from a network through a communication port. A system for generating a fine-grained visual target recognition adaptive training program for people may also include different forms of program storage units and data storage units, such as a hard disk, a read-only memory (ROM) and a random access memory (RAM), which can store various data files used for computer processing and / or communication, and possible program instructions executed by the processor. The processor executes these instructions to implement the main part of the method. The result of the processor processing is transmitted to the subject device through the communication port and displayed on the subject interface.

[0124] The above-mentioned method for generating an adaptive training program for fine-grained visual target recognition for humans can be implemented as a computer program, stored in a hard disk, and recorded in a processor for execution. Therefore, an embodiment of the present invention also provides a computer-readable medium storing a computer program code, which, when executed by a processor, implements the above-mentioned method for generating an adaptive training program for fine-grained visual target recognition for humans.

[0125] When a method for generating a human-oriented fine-grained visual target recognition adaptive training program is implemented as a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described in the embodiments of the present invention can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0126] Example 5

[0127] An embodiment of the present invention also provides a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system, comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment method as described in Example 2 above.

[0128] A human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system may include an internal communication bus, a processor, a read-only memory (ROM), a random access memory (RAM), a communication port, and a hard disk. The internal communication bus can realize data communication between components of a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system. The processor can make judgments and issue prompts. In some embodiments, the processor can be composed of one or more processors. The communication port can realize data communication outside a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system. In some embodiments, a human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system can also send and receive information and data from a network through a communication port. A human-oriented fine-grained visual target recognition adaptive training difficulty adjustment system may also include different forms of program storage units and data storage units, such as a hard disk, a read-only memory (ROM) and a random access memory (RAM), which can store various data files used for computer processing and / or communication, and possible program instructions executed by the processor. The processor executes these instructions to implement the main part of the method. The result of the processor processing is transmitted to the subject device through the communication port and displayed on the subject interface.

[0129] The above-mentioned method for adjusting the difficulty of adaptive training of fine-grained visual target recognition for humans can be implemented as a computer program, stored in a hard disk, and recorded in a processor for execution. Therefore, an embodiment of the present invention also provides a computer-readable medium storing a computer program code, which, when executed by a processor, implements the above-mentioned method for adjusting the difficulty of adaptive training of fine-grained visual target recognition for humans.

[0130] When a method for generating a human-oriented fine-grained visual target recognition adaptive training program is implemented as a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described in the embodiments of the present invention can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0131] Example 6

[0132] An embodiment of the present invention also provides a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system, comprising: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method as described in Example 3 above.

[0133] A human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system may include an internal communication bus, a processor, a read-only memory (ROM), a random access memory (RAM), a communication port, and a hard disk. The internal communication bus can realize data communication between components of a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system. The processor can make judgments and issue prompts. In some embodiments, the processor can be composed of one or more processors. The communication port can realize data communication outside a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system. In some embodiments, a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system can also send and receive information and data from a network through a communication port. A human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system may also include different forms of program storage units and data storage units, such as a hard disk, a read-only memory (ROM) and a random access memory (RAM), which can store various data files used for computer processing and / or communication, and possible program instructions executed by the processor. The processor executes these instructions to implement the main part of the method. The result of the processor processing is transmitted to the subject device through the communication port and displayed on the subject interface.

[0134] The above-mentioned method for adaptive closed-loop feedback training of fine-grained visual target recognition for humans can be implemented as a computer program, stored in a hard disk, and recorded in a processor for execution. Therefore, an embodiment of the present invention also provides a computer-readable medium storing a computer program code, which, when executed by a processor, implements the above-mentioned method for adaptive closed-loop feedback training of fine-grained visual target recognition for humans.

[0135] When a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training method is implemented as a computer program, it can also be stored in a computer-readable storage medium as a product. For example, a computer-readable storage medium may include, but is not limited to, a magnetic storage device (e.g., a hard disk, a floppy disk, a magnetic strip), an optical disk (e.g., a compact disk (CD), a digital versatile disk (DVD)), a smart card, and a flash memory device (e.g., an electrically erasable programmable read-only memory (EPROM), a card, a stick, a key drive). In addition, the various storage media described in the embodiments of the present invention can represent one or more devices and / or other machine-readable media for storing information. The term "machine-readable medium" may include, but is not limited to, wireless channels and various other media (and / or storage media) that can store, contain and / or carry code and / or instructions and / or data.

[0136] It should be understood that the above-mentioned embodiments are only illustrative. The embodiments described in the present invention may be implemented in hardware, software, firmware, middleware, microcode or any combination thereof. For hardware implementation, the processing unit may be implemented in one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors and / or other electronic units designed to perform the functions described in the present invention or in combination thereof.

[0137] like Figures 1-2 As shown, a human-oriented fine-grained visual target recognition adaptive closed-loop feedback training system includes a training database, a training image library, a training module, a training difficulty adjustment module, and a testing module, wherein:

[0138] The training database is used to store all the early training data of the trainees (such as in the previous version of this system), including the scores of the standard fine-grained visual object indefinite choice recognition test after each single training and the training parameters of the training program.

[0139] The training image library is used to store fine-grained visual target images for training trainees, standard fine-grained visual target test images for testing trainees, and corresponding image difficulty and image name labels. The training image library stores multiple fine-grained visual target sub-images containing a single target object to be identified (there may be multiple objects), fine-grained visual target images composed of fine-grained visual target sub-images containing one or more targets to be identified (there may be multiple objects), and standard fine-grained visual target test images.

[0140] The image name labels refer to the recognition results of fine-grained visual object recognition experts on the fine-grained visual object sub-images used for training and testing.

[0141] Image difficulty is obtained by normalizing the accurate recognition time spent by fine-grained visual target recognition experts on each fine-grained visual target sub-image for accurate recognition testing, and mapping the accurate recognition time of all fine-grained visual target sub-images to a difficulty coefficient in the interval (0,1). The difficulty coefficient obtained by mapping each fine-grained visual target sub-image is the image difficulty of the fine-grained visual target sub-image.

[0142] The test module is used to test the trainee's general field object recognition ability and fine-grained visual target recognition ability, and send the test results to the training database for storage. The test module includes a personal recognition trait test module and a standard fine-grained visual target indefinite choice recognition test module, wherein:

[0143] The personal identification trait test module is used to test the trainee's personal identification trait and test the trainee's general field object recognition ability (i.e., talent). The personal identification trait test is only conducted when the trainee uses it for the first time. The personal identification trait test module has a total of 54 multiple-choice questions, and the test results are output as the correct rate;

[0144] The standard fine-grained visual target indefinite selection recognition test module is used to conduct a standard fine-grained visual target indefinite selection recognition test on trainees using a standard fine-grained visual target test image to test the trainees' fine-grained visual target recognition ability. The standard fine-grained visual target test image is composed of nine fine-grained visual target sub-images spliced ​​in a nine-square grid manner. The results of the standard fine-grained visual target indefinite selection recognition test can be given by the score calculation method in the invention patent named "A method, system and computer-readable medium for quantifying human-oriented SAR multi-target visual recognition ability", and the test results are fed back to the training difficulty adjustment module to adjust the trainees' next training plan.

[0145] The training module includes an observation learning module and a feedback learning module, that is, it includes two training methods, observation learning and feedback learning. Before each training, the feedback result output by the training difficulty adjustment module after comprehensively considering the personal characteristics of the trainee and the training effect is obtained, that is, the training plan output by the training difficulty adjustment module is obtained. The training module adjusts the six training parameters in the observation learning module and the feedback learning module according to the training plan output by the training difficulty adjustment module, and then performs fine-grained visual target visual recognition training on the trainee based on the adjusted observation learning module and feedback learning module.

[0146] The training difficulty adjustment module receives the test results and personal characteristics of the trainee fed back by the test module, provides a feasible training plan for the individual and has mathematical and statistical significance, and sends the feasible training plan to the training module to train the trainee. The training difficulty adjustment module further includes a personal characteristic Z value generation module, a training characteristic β value generation module, a training plan X value generation module and a training plan feasibility judgment module, wherein:

[0147] The personal trait Z value generation module generates the personal trait Z value based on the personal identification trait test structure, the training trait β value generation module generates the training trait β value based on the multivariate linear regression model, the training scheme X value generation module generates the X value that determines the training scheme, and the training scheme feasibility judgment module judges the feasibility of the training scheme determined by the X value based on the ordered logistic regression model.

[0148] The problem of high training difficulty can be solved by the training difficulty adjustment module, and the problem of single training method can be solved by the closed-loop feedback training method combining observation learning and feedback learning. The two training methods adopted in the embodiment of the present invention, namely observation learning training and feedback learning training, are essentially training methods that integrate conceptual information and image perception information. This training method itself can improve the recognition training efficiency of the trainees; and the training program formulated according to the personal recognition characteristics of the trainees can also improve the training efficiency; therefore, the training efficiency is improved, and the training process required to reach the level of image recognizers is shortened.

[0149] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for generating adaptive training schemes for fine-grained visual object recognition for humans, characterized in that: Follow these steps: Step S1, generating a comprehensive training index Z based on the individual recognition trait test score and the current standard fine-grained visual target indefinite item selection recognition test score; Step S2: Obtain all training data of all trainees and generate the i-th training parameter X based on the multivariate linear regression model. i The weight β i ; Step S3: Based on the generated comprehensive training index Z and the estimated weight β i , combined with the training scheme generation algorithm expressed in the following formula, the training parameters X that determine the training scheme are generated i , and according to the training parameters X i Set up training programs; Among them, β min is the minimum value of the training weight in this training scheme, β max is the maximum value of the training weight in this training scheme; the training parameter X i There are 6 items in total, i = 1, 2, 3, 4, 5, 6, where X1 is the picture stimulation time of the trainees in observational learning training, X2 is the time when the trainees can end the picture stimulation in advance in observational learning training, X3 is the total number of fine-grained visual target sub-images that the trainees receive picture stimulation in observational learning training, X4 is the total number of fine-grained visual target sub-images of picture stimulation in feedback learning training, X5 is the time it takes for the trainees to make choices according to the prompts after seeing the picture stimulation in feedback learning training, and X6 is the image difficulty of a single fine-grained visual target sub-image used in feedback learning training; The step S1 generates a comprehensive training index Z according to the following formula: Among them, exp is the exponential function, The results of the previous standard fine-grained visual object indeterminate item selection recognition test, A s The expected next standard fine-grained visual target indeterminate item selection recognition test results are manually input by the trainee, K m is the personal identification coefficient of the current trainee obtained based on the personal identification trait test scores; The personal identification coefficient K of the current trainee m Obtained by the following method: Obtain the personal identification trait test scores of all trainees and the personal identification trait test score of the current trainee, use the minimum and maximum values ​​of the personal identification trait test scores of all trainees to form a personal identification trait test score interval, then normalize the personal identification trait test score interval to the interval [0,1], and then normalize the personal identification trait test score of the current trainee to the interval [0,1] to obtain the personal identification coefficient K m ; The image difficulty of the fine-grained visual target subgraph is obtained by normalizing the accurate recognition time spent by a fine-grained visual target recognition expert on accurate recognition test of each fine-grained visual target subgraph, mapping the accurate recognition time of all fine-grained visual target subgraphs to a difficulty coefficient in the interval of (0,1), and the difficulty coefficient obtained by mapping each fine-grained visual target subgraph is the image difficulty of the fine-grained visual target subgraph.

2. The method for generating a human-oriented fine-grained visual object recognition adaptive training program according to claim 1, characterized in that: The specific implementation process of step S2 is as follows: First, the multiple linear regression equation is established according to the following formula: In the above formula, Y is the predicted score of the trainee's next standard fine-grained visual object indeterminate item selection recognition test, β0 is a constant term, and e is the residual; β i is the training weight, is the i-th training parameter X in the training scheme i The weight of Then, using the established multiple linear regression equation, the residual sum of squares Q is minimized by the least squares method according to the following formula: Among them, Y k is the standard fine-grained visual object indeterminate item choice recognition test score after the k-th training of the current trainee, Y k The estimated value of X ik is the i-th training parameter in the k-th training of the current trainee, β ik is the weight of the i-th training parameter in the k-th training of the current trainee, and n is the total number of training data obtained.

3. A method for adaptive training difficulty adjustment for fine-grained visual object recognition for humans, characterized in that: Follow these steps: Step 1: construct an ordered logistic regression model based on the generated fine-grained visual object recognition training plan for the trainee; Step 2: According to the ordered logistic regression model, the probability P of the trainee reaching level g in the standard fine-grained visual target indefinite choice recognition test after this training is calculated according to the following formula: g : Among them, P g-1 is the probability that the trainee's standard fine-grained visual target indefinite item selection recognition test score reaches level g-1 after this training, exp is an exponential function, β 0g-1 is the constant term of the g-1th regression; X i To determine the training parameters of the training scheme, β i is the training weight; Step 3: Determine the probability P that the trainee's standard fine-grained visual target indefinite item choice recognition test score reaches level g after this training g Does it meet the requirements? If so, directly output the training parameters X i And generate a training plan for training, otherwise adjust the training difficulty; Step 4: Adjust the training difficulty.

4. The method for adjusting the difficulty of adaptive training of fine-grained visual target recognition for humans according to claim 3, characterized in that: The specific implementation process of step 1 is as follows: Obtain all training data from the training database, put all training data into a decision tree for classification, and perform interval classification on the trainee's standard fine-grained visual object indeterminate choice recognition test score Y before each training. The interval step is 0.1, and all training data are divided into g different data sets, i.e. g levels; Construct an ordered logistic regression model as follows: Where P(Y≤j) is the probability that the trainee's standard fine-grained visual target indeterminate item choice recognition test score reaches level Y, j = 1, 2, 3L g-1, ln is an exponential function; The specific implementation process of step 4 for adjusting the training difficulty is as follows: The advance and retreat method in the precise one-dimensional search in gradient descent is used to re-evaluate the Z value, and the new Z value obtained by re-evaluation is used to regenerate the training scheme, that is, to regenerate the training parameters X. i , until the probability P of the trainee reaching level g in the standard fine-grained visual target indefinite item selection recognition test after this training g Meet the requirements and output the corresponding training plan, i.e. training parameters X i .

5. An adaptive closed-loop feedback training method for fine-grained visual object recognition for humans, characterized in that: Follow these steps: S1: Conduct a standard fine-grained visual target indefinite item selection recognition test and a personal identification trait test on the trainee, and obtain the trainee's personal identification trait test score and the current standard fine-grained visual target indefinite item selection recognition test score; S2: according to the personal identification trait test scores of all trainees, the personal identification trait test scores of the trainees in S1, and the current standard fine-grained visual target indefinite choice recognition test scores, a training plan is generated by using the method for generating a training plan for human-oriented fine-grained visual target recognition as described in any one of claims 1 to 2; S3: Based on the generated training plan, the trainees are trained in observation learning and feedback learning in sequence; S4: The trainees were given a standard fine-grained visual target indeterminate item choice recognition test; S5: Determine whether the trainee's standard fine-grained visual target indeterminate item choice recognition test score in S4 is greater than 0.

95. If so, end the training; otherwise, enter S6; S6: Using the human-oriented fine-grained visual target recognition training difficulty adjustment method as described in any one of claims 3 to 4, adjust the training difficulty of the trainee's next training plan, regenerate the training plan, and repeat S3 to S5 until the trainee's standard fine-grained visual target indefinite choice recognition test score is greater than 0.95, then end the training.

6. The method for adaptive closed-loop feedback training of fine-grained visual object recognition for humans according to claim 5, characterized in that: The process of the personal identification trait test is as follows: the trainees memorize a set of 6 objects under fine-grained visual targets within a specified time. Each object will be presented in the form of 3 fine-grained visual targets from different perspectives, and each fine-grained visual target will be presented for 3 seconds; Then, the trainee is asked to select six objects from the memory among the objects under another set of fine-grained visual targets. No feedback is provided after the selection. The trainee's actual selection is automatically compared with the correct answer, and the accuracy rate is output as the personal identification trait test score. The process of the standard fine-grained visual target indefinite selection recognition test is as follows: in each standard fine-grained visual target test image, the trainee selects the target object in turn. The selection time for each target object is limited to 10 seconds. After each target object is selected, the trainee will choose "continue to select the next target object" or "I have selected all targets". When the trainee chooses "continue to select the next target object", the trainee repeats the selection process until all the targets in the nine-square grid image are selected and "I have selected all targets" is selected. No feedback is provided after the selection. In the single observation learning training, the trainees will first see a picture name label that lasts for 1 second, which is the recognition result of the fine-grained visual target by the fine-grained target recognition expert. Then they will see a plus sign that lasts for 500ms. After the plus sign that lasts for 500ms ends, the trainees will see a picture stimulus that lasts for 6-18 seconds, which is determined by the training plan of the training session. The trainees can choose to end it early within a certain time range. In the single-feedback learning training, after the 500ms plus sign, the trainees will see a fine-grained visual target image, i.e., a picture stimulus, which is determined by the current training plan and lasts for 5-15s and is composed of multiple fine-grained visual target sub-pictures. During this period, the trainees need to select a fine-grained visual target sub-picture that meets the requirements from the fine-grained visual target composed of multiple fine-grained visual target sub-pictures according to text prompts. During the picture stimulation, the trainees will be prompted with text about the remaining selection time, and a plus sign with a duration of 500ms will be displayed after the trainees have made their choice or after the picture stimulation time has ended. After the plus sign with a duration of 500ms is displayed, result feedback will be given based on the trainees' choice, and the trainees' choice of the fine-grained visual target sub-picture made according to the text prompts will be fed back as correct or incorrect.

7. A human-oriented fine-grained visual object recognition adaptive training scheme generation system, characterized by: include: a memory for storing instructions executable by a processor; A processor, configured to execute the instructions to implement the method according to any one of claims 1 to 2.

8. A human-oriented fine-grained visual object recognition adaptive training difficulty adjustment system, characterized by: include: a memory for storing instructions executable by a processor; A processor, configured to execute the instructions to implement the method according to any one of claims 3 to 4.

9. An adaptive closed-loop feedback training system for fine-grained visual object recognition for humans, characterized by: include: a memory for storing instructions executable by a processor; A processor is used to execute the instructions to implement the method according to claim 5.

Images