Teaching quality evaluation method and system

Abstract

The invention belongs to the field of teaching quality evaluation, and particularly relates to a teaching quality evaluation method and system, and the method comprises the following steps: S1, uploading the content of teaching courseware to a database, and carrying out the keyword labeling; s2, collecting video information during teaching, and constructing a target function P; s3, optimizing the Pm, the Pe and the Pf by using an improved DBO algorithm; and S4, feeding back the optimized parameters to the deep learning algorithm, continuously adjusting the weight coefficient of each layer, continuously reducing the error between the true value and the predicted value, substituting the predicted values Pm, Pe and Pf output by the deep learning algorithm into the step S2, and calculating to obtain a score P according to a target function. The method has the advantages that an improved DBO algorithm is utilized, and a random walk strategy and a crisscross strategy are added in a rolling behavior for disturbance; and a direction following strategy is added in the stealing process, so that the optimizing precision is improved, and the performance of the method in the aspects of global exploration and optimizing precision is improved.

Description

Technical Field

[0001] The present application belongs to the field of teaching quality evaluation, and specifically relates to a teaching quality evaluation method and system. Background Art

[0002] At present, the most commonly used evaluation methods for the teaching quality of individual teachers in schools are still lectures and discussions. The lecture listening methods here are divided into attending classes in person and attending classes using remote video methods instead of entering the classroom. However, lectures and discussions collect more subjective information and less objective information. With the development of information technology, the evaluation of teaching quality based on the education information platform has also made certain progress. For example, the teaching system CN107784883A and the teaching quality evaluation method based on the teaching system have solved the problem in the prior art that the teacher does not know that many students have already mastered the content of the explanation during the interaction between the teacher and the students, and also solved the problem in the prior art that the evaluation results of the teacher's teaching quality evaluation method are greatly deviated from the actual situation. However, it is difficult to fully reflect the teaching quality only from the interaction between teachers and students. Summary of the invention

[0003] Based on the above problems, this application not only enriches the evaluation indicators, but also uses the improved DBO algorithm to add a random walk strategy to the rolling behavior for perturbation; adds a direction following strategy to the stealing to improve the optimization accuracy, thereby improving its performance in global exploration and optimization accuracy. Its technical solution is:

[0004] A teaching quality evaluation method comprises the following steps:

[0005] S1. Upload the courseware content to the database and annotate it with keywords;

[0006] S2. Collect video information during lectures and construct the objective function P;

[0007] P = aF(P m )+kF(P e )+cF(P f );

[0008] Where a, k and c are weight coefficients;

[0009] P m Teaching matching degree: the teaching content is compared with the uploaded teaching courseware content to obtain the teaching matching degree; F(P m ) is P m The standardized function of

[0010] P e For learning concentration: whether the student faces the teacher or the teaching medium; F(P e ) is P eThe standardized function of

[0011] P f Teacher-student interaction activity: the ratio of the number of teacher-student interactions to the number of questions asked; F(P f ) is P f The standardized function of

[0012] S3. Using the improved DBO algorithm to m , P e , P f Conduct optimization search;

[0013] S4. Feed the optimized parameters back to the deep learning algorithm so that it can continuously adjust the weight coefficient of each layer, continuously reduce the error between the true value and the predicted value, and output the predicted value P of the deep learning algorithm. m , P e , P f is substituted into step S2, and the score P is calculated according to the objective function.

[0014] Preferably, the value rules for a, k and c are as follows:

[0015] If P m ≥0.8, then a=0.7; if 0.8 <P m ≤0.5, then a=0.2, if P m <0.5, then a=0.1;

[0016] If the student faces toward the teacher or the teaching medium, it is considered that the class concentration is high, k = 0.7; if the student faces away from the teacher or the teaching medium, k = 0.3; if the student sleeps at the desk, k = 0;

[0017] If the number of teacher-student interactions is greater than the number of questions asked, then c = 0.7; if the number of teacher-student interactions is equal to the number of questions asked, then c = 0.2; if the number of teacher-student interactions is less than the number of questions asked, then c = 0.1.

[0018] Preferably, in step S3, the improved dung beetle algorithm identification step is:

[0019] S31: Population initialization: initial population size, sampling time, number of iterations, parameter range; S32: Calculate the fitness function P (P m , P e , P f )value;

[0020] S33: Record and save the best results to determine the population range for the next generation;

[0021] S34: Determine the iteration condition: determine whether the number of iterations or the identification accuracy has been reached. If the termination condition is met, output the optimal parameter Pm , P e , P f , end; otherwise, continue to the next step;

[0022] S35: Update the next generation through the process of rolling, dancing, foraging, stealing and reproduction. Add random walk strategy and cross-border strategy to the rolling behavior for perturbation; add direction following strategy to stealing to improve the optimization accuracy.

[0023] S36: Update the number of iterations τ=τ+1, and return to Step 32.

[0024] Preferably, in step S35, a random walk strategy is used to disturb the dung beetle's ball rolling behavior, and the formula of the random walk strategy is as follows:

[0025] X(m)={0,c ussum [2r(m1)-1],c ussum [2r(m2)-1],…,c ussum [2r(m n )-1]

[0026] Among them, X(m) is the number of steps of random walk; C ussum is the cumulative number of walking steps; m is the current number of walking steps; m n is the number of steps for the nth walk, r(m) is a random function with a value of [0,1];

[0027] Normalize it as follows:

[0028]

[0029] Where: is the parameter P of the i-th dung beetle in the τ-th iteration m or P e or P f Particle position; α i and b i are the minimum and maximum values ​​of the random walk variable of the i-th dimension respectively; and are the minimum and maximum values ​​of the random walk variable of the i-th dimension at the τ-th iteration, respectively.

[0030] Preferably, in step S35, the horizontal cross hypothesis SM i1 and SM i2 is the parent individual after pairing, and its offspring is and Obtained by the following formula:

[0031]

[0032] Among them SMi1j and SM i2j Respectively represent SM i1 and SM i2 The jth dimension of It's SM i1j and SM i2j The horizontal crossover in the jth dimension generates the parameter P m , P e , P f In the j-th dimension of the offspring, r1 and r2 are random numbers uniformly distributed in the range (0,1), and λ1 and λ2 are random numbers uniformly distributed in the range (-1,1); the generated parameter P m , P e , P f The offspring are compared with their parents respectively, and the individuals with smaller objective function values ​​are retained;

[0033] The vertical cross is as follows:

[0034]

[0035] Among them, SM ij1 and SM ij2 Respectively represent SM j1 and SM j2 The i-th dimension of It's SM j1 and SM j2 , in the i-th dimension of the vertical crossover to generate offspring, r is a random number uniformly distributed in the range of (0,1), and the generated parameter P m , P e , P f The offspring are compared with their parents, and the individuals with smaller target values ​​are kept;

[0036] Set the threshold function pv:

[0037] pv=2*(1-(τ / T) 1.5 ) / 3;

[0038] τ is the current iteration number, T is the maximum iteration number, when r≥pv, the vertical and horizontal cross strategy is adopted; when r<pv, the vertical and horizontal cross strategy is skipped in this iteration.

[0039] Preferably, in step S35, the direction following strategy steps are as follows:

[0040] S35-1. After all thief dung beetles have updated their positions, they adopt a direction-following strategy to update their positions again;

[0041] S35-2. Before stealing, the thief dung beetle needs to observe the ball-pushing dung beetle, randomly select a ball-pushing dung beetle based on their respective observations, and follow it to facilitate the next step of stealing;

[0042] S35-3. Assuming that the global optimal position is the most suitable position for observation, the position update formula using the direction following strategy is as follows:

[0043]

[0044] δ=pv*(||X b +Ψ(U max -U lim )||)

[0045] In the formula, X b is the global optimal position, R τ The thief dung beetle randomly selects the location of the target pusher dung beetle; D (R τ -X b ) represents the unit vector of the straight line direction from the optimal position to the randomly selected position of the ball-pushing dung beetle; Ψ is a random number with a value of [0,1], δ is the moving step length, U max , U lim They are the upper and lower limits of the optimized position respectively.

[0046] Preferably, in step S5, the loss function of the deep learning algorithm uses mean square error MSE, and the error between the true value and the predicted value is calculated as follows:

[0047]

[0048]

[0049] Where W1-W4 is the weight of each layer, b1-b4 is the bias of each layer; N is the number of samples, is the predicted value (about P m , P e , P f The predicted value of i is the true value, The feature vector value for each layer input.

[0050] A teaching quality evaluation system; comprising a data acquisition module, a data processing module, a data calculation module and an output module;

[0051] The data acquisition module is used to obtain video information during the lecture through devices such as cameras and microphones;

[0052] The data processing module uses a deep learning method to process video information to obtain teaching matching, learning concentration, and the number of teacher-student interactions;

[0053] The data calculation module: constructs the objective function, and uses the improved DBO algorithm to search for the best solution, and feeds the search result back to the deep learning algorithm so that it continuously adjusts the weight coefficient of each layer to obtain the final score P;

[0054] The output module: outputs the score P visually.

[0055] Preferably, the recognition of learning concentration adopts convolutional neural network technology based on deep learning, and the head posture data is obtained by using dimensionality reduction and adaptive gradient methods; head posture refers to the azimuth of a person's head in three-dimensional space, and the rotation angles of the head around the X, Y, and Z axes are recorded as pitch, yaw, and roll, respectively; when the student's face is facing the teacher or the teaching carrier, the classroom concentration is high, and the influence of the student's position in the classroom should also be considered, and the yaw angle is 0°~20° at this time; when the yaw angle exceeds 60°, it is considered that the student is looking around and the classroom concentration is average; if the student sleeps at the desk, the pitch angle will exceed 60°, and the classroom concentration is seriously insufficient.

[0056] Compared with the prior art, the present invention has the following beneficial effects:

[0057] 1. Evaluation indicators are set from two dimensions: teachers and students. Deep learning is used for feature extraction, and the data obtained is relatively objective.

[0058] 2. In the early stage of iteration, the thief dung beetle tends to quickly approach the optimal individual, causing the algorithm to fall into the local optimum. Therefore, this application adds the random walk strategy and the vertical and horizontal cross strategy to the rolling ball behavior, thereby improving the global exploration ability of the algorithm; adding a threshold function to the vertical and horizontal cross strategy to reduce the search cost and improve the search efficiency; adding a direction following strategy to the stealing to improve the optimization accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 This is the flow chart of this application.

[0060] Figure 2 This is the schematic diagram of the direction following strategy.

[0061] Figure 3 This is a simulation test chart of the mean square error of deep learning. DETAILED DESCRIPTION

[0062] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0063] A teaching quality evaluation method comprises the following steps:

[0064] S1. Upload the courseware content to the database and annotate it with keywords;

[0065] S2. Collect video information during lectures and construct the objective function P;

[0066] P = aF(P m )+kF(P e )+cF(P f );

[0067] P m Teaching matching degree: the teaching content is compared with the uploaded teaching courseware content to obtain the teaching matching degree; F(P m ) is P m The standardized function of

[0068] P e For learning concentration: whether the student faces the teacher or the teaching medium; F(P e ) is P e The standardized function of

[0069] P f Teacher-student interaction activity: the ratio of the number of teacher-student interactions to the number of questions asked; F(P f ) is P f The standardized function of

[0070] a, k and c are weight coefficients, and the value rules for a, k and c are as follows:

[0071] If P m ≥0.8, then a=0.7; if 0.8 <P m ≤0.5, then a=0.2, if P m <0.5, then a=0.1;

[0072] If the student faces toward the teacher or the teaching medium, it is considered that the class concentration is high, k = 0.7; if the student faces away from the teacher or the teaching medium, k = 0.3; if the student sleeps at the desk, k = 0;

[0073] If the number of teacher-student interactions P f is greater than the number of questions asked, then c = 0.7. If the number of teacher-student interactions P f is equal to the number of questions asked, then c = 0.2. If the number of teacher-student interactions P f If the number of questions is smaller than N, then c = 0.1. S3. Use the improved DBO algorithm to m , P e , P f Conduct optimization search;

[0074] In step S3, the improved dung beetle algorithm identification steps are:

[0075] S31: Population initialization: initial population size, sampling time, number of iterations, parameter range; S32: Calculate the fitness function P (P m , P e , P f )value;

[0076] S33: Record and save the best results to determine the population range for the next generation;

[0077] S34: Determine the iteration condition: determine whether the number of iterations or the identification accuracy has been reached. If the termination condition is met, output the optimal parameter P m , P e , P f , end; otherwise, continue to the next step;

[0078] S35: Update the next generation through the process of rolling, dancing, foraging, stealing and reproduction. Add random walk strategy and cross-border strategy to the rolling behavior for perturbation; add direction following strategy to stealing to improve the optimization accuracy.

[0079] In step S35, a random walk strategy is used to disturb the dung beetle's ball rolling behavior. The formula of the random walk strategy is as follows:

[0080] X(m)={0,c ussum [2r(m1)-1],c ussum [2r(m2)-1],…,c ussum [2r(m n )-1]

[0081] Among them, X(m) is the number of steps of random walk; C ussum is the cumulative number of walking steps; m is the current number of walking steps; m n is the number of steps for the nth walk, r(m) is a random function with a value of [0,1];

[0082] Normalize it as follows:

[0083]

[0084] Where: is the parameter P of the i-th dung beetle in the τ-th iteration m or P e or P f Particle position; α i and b i are the minimum and maximum values ​​of the random walk variable of the i-th dimension respectively; and are the minimum and maximum values ​​of the random walk variable of the i-th dimension at the τ-th iteration, respectively.

[0085] In step S35, the horizontal cross hypothesis SM i1 and SM i2 is the parent individual after pairing, and its offspring is and Obtained by the following formula:

[0086]

[0087] Among them SM i1j and SM i2j Respectively represent SM i1 and SM i2 The jth dimension of It's SM i1j and SM i2j The horizontal crossover in the jth dimension generates the parameter P m , P e , P f In the j-th dimension of the offspring, r1 and r2 are random numbers uniformly distributed in the range (0,1), and λ1 and λ2 are random numbers uniformly distributed in the range (-1,1); the generated parameter P m , P e , P f The offspring are compared with their parents respectively, and the individuals with smaller objective function values ​​are retained;

[0088] The vertical cross is as follows:

[0089]

[0090] Among them, SM ij1 and SM ij2 Respectively represent SM j1 and SM j2 The i-th dimension of It's SM j1 and SM j2 , in the i-th dimension of the vertical crossover to generate offspring, r is a random number uniformly distributed in the range of (0,1), and the generated parameter P m , P e , P f The offspring are compared with their parents, and the individuals with smaller target values ​​are kept;

[0091] Set the threshold function pv:

[0092] pv=2*(1-(τ / T) 1.5 ) / 3;

[0093] τ is the current iteration number, T is the maximum iteration number, when r≥pv, the vertical and horizontal cross strategy is adopted; when r<pv, the vertical and horizontal cross strategy is skipped in this iteration.

[0094] In step S35, the direction following strategy steps are as follows:

[0095] S35-1. After all thief dung beetles have updated their positions, they adopt a direction-following strategy to update their positions again;

[0096] S35-2. Before stealing, the thief dung beetle needs to observe the ball-pushing dung beetle, randomly select a ball-pushing dung beetle based on their respective observations, and follow it to facilitate the next step of stealing;

[0097] S35-3. Assuming that the global optimal position is the most suitable position for observation, the position update formula using the direction following strategy is as follows:

[0098]

[0099] δ=pv*(||X b +Ψ(U maX -U lim )||)

[0100] In the formula, X b is the global optimal position, R τ The thief dung beetle randomly selects the location of the target pusher dung beetle; D (R τ -X b ) represents the unit vector of the straight line direction from the optimal position to the randomly selected position of the ball-pushing dung beetle; Ψ is a random number with a value of [0,1], δ is the moving step length, U max , U lim They are the upper and lower limits of the optimized position respectively.

[0101] S36: Update the number of iterations τ=τ+1, and return to Step 32.

[0102] S4. Feed the optimized parameters back to the deep learning algorithm so that it can continuously adjust the weight coefficient of each layer, continuously reduce the error between the true value and the predicted value, and output the predicted value P of the deep learning algorithm. m , P e , P f is substituted into step S2, and the score P is calculated according to the objective function.

[0103] Figure 2In the above, it is assumed that A represents the global optimal position vector, B represents the position vector of a randomly selected ball-pushing dung beetle, and BA is recorded as C. From the synthesis and decomposition of motion, it can be seen that if δ≥0, then A+δ×C represents a new position vector after moving δ units from A along the direction C. From the probability theory, it can be known that the probability of ball-pushing dung beetles being distributed in all directions around the optimal position is equal. Therefore, the thief dung beetle group adopts this strategy to move from the global optimal position A along the direction C with a certain step length to achieve full development of the area with the most hope of finding the optimal solution (i.e., the dotted circle area in the figure). A+δ×C is the new position vector obtained by the direction following strategy. The new position obtained will be compared with the previously obtained position and the best one will be retained.

[0104] In step S4, the loss function of the deep learning algorithm uses the mean square error MSE, and the error between the true value and the predicted value is calculated as follows:

[0105]

[0106]

[0107] Where W1-W4 is the weight of each layer, b1-b4 is the bias of each layer; N is the number of samples, is the predicted value, y i is the true value, The feature vector value for each layer input.

[0108] from Figure 3 It can be found that the curve of the deep learning algorithm (neural network) adjusted by the improved DBO algorithm is steeper and converges faster. After 30 iterations, the fitness value tends to be stable and reaches about 0.10, meeting the accuracy requirements. However, using only the neural network BP (comparative experiment) requires more iterations and the convergence speed is also limited.

[0109] A teaching quality evaluation system; comprising a data acquisition module, a data processing module, a data calculation module and an output module;

[0110] The data acquisition module is used to obtain video information during the lecture through devices such as cameras and microphones;

[0111] The data processing module uses a deep learning method to process video information to obtain teaching matching, learning concentration, and the number of teacher-student interactions;

[0112] The data calculation module: The data calculation module: constructs the objective function, and uses the improved DBO algorithm to search for the best solution, and feeds the search result back to the deep learning algorithm so that it continuously adjusts the weight coefficient of each layer to obtain the final score P;

[0113] The output module: outputs the score P visually.

[0114] For the recognition of learning concentration, convolutional neural network technology based on deep learning is used, and head posture data is obtained by using dimensionality reduction and adaptive gradient methods; head posture refers to the azimuth of a person's head in three-dimensional space, and the rotation angles of the head around the X, Y, and Z axes are recorded as pitch, yaw, and roll, respectively; when the student's face is facing the teacher or the teaching carrier, the class concentration is high, and the influence of the student's position in the classroom must also be considered. At this time, the yaw angle is 0°~20°; when the yaw angle exceeds 60°, it is considered that the student is looking around and the class concentration is average; if the student sleeps at the desk, the pitch angle will exceed 60°, and the class concentration is seriously insufficient.

[0115] A computer device, characterized in that it includes at least one processor and a computer-readable medium storing a computer program, wherein when the computer program is read and executed by the processor, the method described in the present application is implemented.

[0116] A computer-readable medium, characterized in that the computer-readable medium stores a computer program, and when the computer program is read and executed by a processor, the method described in the present application is implemented.

[0117] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus dynamic random access memory (Rambus DRAM, referred to as RDRAM), and interface dynamic random access memory (DRDRAM).

[0118] The above embodiments are only exemplary embodiments of the present invention and are not intended to limit the present invention. The protection scope of the present invention is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present invention within the essence and protection scope of the present invention, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present invention.

Claims

1. A teaching quality evaluation method, characterized in that: The following steps are involved: S1. Upload the courseware content to the database and annotate it with keywords; S2. Collect video information during lectures and construct the objective function P; P=aF(P m )+kF(P e )+cF(P f ); Where a, k and c are weight coefficients; P m Teaching matching degree: the teaching content is compared with the uploaded teaching courseware content to obtain the teaching matching degree; F(P m ) is P m The standardized function of P e For learning concentration: whether the student faces the teacher or the teaching medium; F(P e ) is P e The standardized function of P f Teacher-student interaction activity: the ratio of the number of teacher-student interactions to the number of questions asked; F(P f ) is P f The standardized function of S3. Using the improved DBO algorithm to m , P e , P f Conduct optimization search; S4. Feed the optimized parameters back to the deep learning algorithm so that it can continuously adjust the weight coefficient of each layer, continuously reduce the error between the true value and the predicted value, and output the predicted value P of the deep learning algorithm. m , P e , P f is substituted into step S2, and the score P is calculated according to the objective function.

2. A teaching quality evaluation method according to claim 1, characterized in that: The rules for selecting values ​​for a, k, and c are as follows: If P m ≥0.8, then a=0.7; if 0.8 <P m ≤0.5, then a=0.2, if P m <0.5, then a=0.1; If the student faces toward the teacher or the teaching medium, it is considered that the class concentration is high, k = 0.7; if the student faces away from the teacher or the teaching medium, k = 0.3; if the student sleeps at the desk, k = 0; If the number of teacher-student interactions is greater than the number of questions asked, then c = 0.7; if the number of teacher-student interactions is equal to the number of questions asked, then c = 0.2; if the number of teacher-student interactions is less than the number of questions asked, then c = 0.

1.

3. A teaching quality evaluation method according to claim 1, characterized in that: In step S3, the improved dung beetle algorithm identification steps are: S31: Population initialization: initial population size, sampling time, number of iterations, and parameter value range; S32: Calculate the fitness function P(P m , P e , P f )value; S33: Record and save the best results to determine the population range for the next generation; S34: Determine the iteration condition: determine whether the number of iterations or the identification accuracy has been reached. If the termination condition is met, output the optimal parameters and end; otherwise, proceed to the next step; S35: Update the next generation through the process of rolling, dancing, foraging, stealing and reproduction. Add random walk strategy and cross-border strategy to the rolling behavior for perturbation; add direction following strategy to stealing to improve the optimization accuracy. S36: Update the number of iterations τ = τ + 1, and return to Step 32.

4. A teaching quality evaluation method according to claim 3, characterized in that: In step S35, a random walk strategy is used to disturb the dung beetle's ball rolling behavior. The formula of the random walk strategy is as follows: X(m)={0,c ussum [2r(m1)-1],c ussum [2r(m2)-1],…,c ussum [2r(m n )-1]; Among them, X(m) is the number of steps of random walk; C ussum is the cumulative number of walking steps; m is the current number of walking steps; m n is the number of steps for the nth walk, r(m) is a random function with a value of [0,1]; Normalize it as follows: Where: is the parameter P of the i-th dung beetle in the τ-th iteration m or P e or P f Particle position; α i and b i are the minimum and maximum values ​​of the random walk variable of the i-th dimension respectively; and are the minimum and maximum values ​​of the random walk variable of the i-th dimension at the τ-th iteration, respectively.

5. A teaching quality evaluation method according to claim 3, characterized in that: In step S35, the horizontal cross hypothesis SM i1 and SM i2 is the parent individual after pairing, and its offspring is and Obtained by the following formula: Among them SM i1j and SM i2j Respectively represent SM i1 and SM i2 The jth dimension of It's SM i1j and SM i2j The horizontal crossover in the jth dimension generates the parameter P m , P e , P f In the j-th dimension of the offspring, r1 and r2 are random numbers uniformly distributed in the range (0,1), and λ1 and λ2 are random numbers uniformly distributed in the range (-1,1); the generated parameter P m , P e , P f The offspring are compared with their parents respectively, and the individuals with smaller objective function values ​​are retained; The vertical cross is as follows: Among them, SM ij1 and SM ij2 Respectively represent SM j1 and SM j2 The i-th dimension of It's SM j1 and SM j2 , in the i-th dimension of the vertical crossover to generate offspring, r is a random number uniformly distributed in the range of (0,1), and the generated parameter P m , P e , P f The offspring are compared with their parents, and the individuals with smaller target values ​​are kept; Set the threshold function pv: pv=2*(1-(τ / T) 1.5 ) / 3; τ is the current iteration number, T is the maximum iteration number, when r≥pv, the vertical and horizontal cross strategy is adopted; when r<pv, the vertical and horizontal cross strategy is skipped in this iteration.

6. A teaching quality evaluation method according to claim 5, characterized in that: In step S35, the direction following strategy steps are as follows: S35-1. After all thief dung beetles have updated their positions, they adopt a direction-following strategy to update their positions again; S35-2. Before stealing, the thief dung beetle needs to observe the ball-pushing dung beetle, randomly select a ball-pushing dung beetle based on their respective observations, and follow it to facilitate the next step of stealing; S35-3. Assuming that the global optimal position is the most suitable position for observation, the position update formula using the direction following strategy is as follows: Where, X b is the global optimal position, R τ The thief dung beetle randomly selects the location of the target pusher dung beetle; D (R τ -X b ) represents the unit vector of the straight line direction from the optimal position to the randomly selected position of the ball-pushing dung beetle; Ψ is a random number with a value of [0,1], δ is the moving step length, U max , U lim They are the upper and lower limits of the optimized position respectively.

7. A teaching quality evaluation method according to claim 1, characterized in that: In step S4, the loss function of the deep learning algorithm uses the mean square error MSE, and the error between the true value and the predicted value is calculated as follows: Where W1-W4 is the weight of each layer, b1-b4 is the bias of each layer; N is the number of samples, is the predicted value, y i is the true value, The feature vector value for each layer input.

8. A system for evaluating teaching quality according to any one of claims 1 to 7; characterized in that: It includes a data acquisition module, a data processing module, a data calculation module and an output module; The data acquisition module is used to obtain video information during the lecture through devices such as cameras and microphones; The data processing module uses a deep learning algorithm to process video information to obtain teaching matching, learning concentration, and the number of teacher-student interactions; The data calculation module: constructs the objective function, and uses the improved DBO algorithm to search for the best solution, and feeds the search result back to the deep learning algorithm so that it can continuously adjust the weight coefficient of each layer to obtain the final score P; the output module: outputs the score P visually.

9. The system according to claim 7, characterized in that For the recognition of learning concentration, convolutional neural network technology based on deep learning is used, and head posture data is obtained by using dimensionality reduction and adaptive gradient methods; head posture refers to the azimuth of a person's head in three-dimensional space, and the rotation angles of the head around the X, Y, and Z axes are recorded as pitch, yaw, and roll, respectively; when the student's face is facing the teacher or the teaching carrier, the classroom concentration is high, and the influence of the student's position in the classroom must also be considered. At this time, the yaw angle is 0°~20°; when the yaw angle exceeds 60°, it is considered that the student is looking around, away from the teacher or the teaching carrier, and the classroom concentration is average; if the student sleeps at the desk, the pitch angle will exceed 60°, and the classroom concentration is seriously insufficient.

Images