Student classroom learning state evaluation method and system based on artificial intelligence

Abstract

The invention discloses a student classroom learning state evaluation method and system based on artificial intelligence, and belongs to the technical field of image processing, and the method comprises the steps: collecting student classroom learning data, enhancing student classroom learning images, constructing a student classroom learning state evaluation model, and generating an evaluation report. According to the scheme, the image is enhanced for the first time by calculating the neighborhood pixel screening coefficient, the local minimum divisible quantity and the pixel adjustment factor, the enhancement upper limit value and the enhancement lower limit value are obtained by calculating the mean value and the standard deviation of the image, and the image is enhanced for the second time, so that the image quality is improved; low-level features and high-level features are extracted respectively, a cosine similarity matrix is constructed through global features, a quantization matrix and a two-dimensional quantization counting matrix are calculated, statistical features are obtained, the low-level features are enhanced, the high-level features are enhanced through attention weights, a comprehensive feature sequence is constructed, and state scores are obtained; and the accuracy of evaluating the classroom learning state of the student is improved.

Description

Technical Field

[0001] The present invention belongs to the field of image processing technology, and specifically refers to a method and system for evaluating students' classroom learning status based on artificial intelligence. Background Art

[0002] The method of evaluating students' classroom learning status is a method that uses image processing and artificial intelligence technology to process and analyze students' classroom learning videos, identify key features that reflect students' learning status, and conduct real-time evaluation of students' classroom learning status, providing a scientific basis for educators, thereby improving students' learning outcomes, optimizing classroom management, and improving overall teaching quality. However, the existing methods of evaluating students' classroom learning status have problems such as image quality differences, noise interference, and rapid and complex movements that blur the image and lack details, resulting in low accuracy in evaluating students' classroom learning status; the existing methods of evaluating students' classroom learning status have problems such as difficulty in accurately capturing subtle dynamic changes and complex dynamic behaviors in the students' learning process, ignoring changes in details in the image, resulting in insufficient accuracy in evaluating students' classroom learning status. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for evaluating the student classroom learning status based on artificial intelligence. In view of the problem that the existing method for evaluating the student classroom learning status has image quality differences, noise interference and fast and complex movements that make the image blurred and lack of details, resulting in low accuracy in evaluating the student classroom learning status, this scheme normalizes each frame of the image to eliminate equipment and environmental differences and ensure input consistency; by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount and the pixel adjustment factor, the image is initially enhanced to remove noise and retain learning-related features, thereby enhancing the key details of the image; by calculating the mean and standard deviation of the image to obtain the enhancement upper limit and enhancement lower limit, the image is secondary enhanced to reasonably control the image enhancement range, avoid distortion and make the details clearer; and then gamma correction is performed to obtain a learning image enhancement sequence to ensure image brightness consistency, improve the stability of the evaluation model, significantly improve the quality of each frame of the image, and thereby improve the accuracy of the student learning status evaluation. High; In view of the problem that the existing methods for evaluating students' classroom learning status are difficult to accurately capture the subtle dynamic changes and complex dynamic behaviors in the students' learning process, and ignore the detailed changes in the image, resulting in insufficient accuracy in evaluating students' classroom learning status, this scheme extracts low-level features through standard convolution and uses deep separable convolution to extract high-level features, thereby enhancing the comprehensive reflection of students' learning status; by calculating the global features to construct the cosine similarity matrix, and then calculating the quantization matrix and the two-dimensional quantization count matrix, the statistical features are obtained, and the low-level features are enhanced, which can not only capture the basic behavioral characteristics, but also reflect the subtle changes in the learning status; by enhancing the high-level features through the attention weight, the sensitivity to the changes in details can be enhanced, and the recognition accuracy of the students' behavioral status can be improved; by constructing a comprehensive feature sequence and obtaining the status score through the long short-term memory network, the students' behavioral performance in the classroom environment can be comprehensively considered, and the students' learning status can be comprehensively reflected, providing a more refined classroom learning status score, and enhancing the accuracy and stability of the evaluation.

[0004] The technical solution adopted by the present invention is as follows: The method for evaluating the student classroom learning status based on artificial intelligence provided by the present invention comprises the following steps:

[0005] Step S1: Student classroom learning data collection;

[0006] Step S2: Students learn image enhancement in class;

[0007] Step S3: Constructing a model for evaluating students’ classroom learning status;

[0008] Step S4: Generate an evaluation report.

[0009] Furthermore, in step S1, the student classroom learning data collection is to collect historical student classroom learning data, and the historical student classroom learning data includes student classroom learning videos and status scores, and the status scores are used as data labels.

[0010] Further, in step S2, the student classroom learning image enhancement specifically includes the following steps:

[0011] Step S21: constructing a learning image sequence; extracting continuous frames from the student's classroom learning video to generate a learning image sequence;

[0012] Step S22: normalization: normalizing each frame of the image in the learning image sequence;

[0013] Step S23: initial image enhancement: for each normalized frame image in the learning image sequence, the following initial enhancement processing is performed in sequence:

[0014] Step S231: Design neighborhood pixel screening coefficients; calculate the neighborhood pixel screening coefficients at each pixel in the image based on the local maximum value, local minimum value, and pixel value variance in the neighborhood of each pixel point; the formula used is as follows:

[0015] ;

[0016] In the formula, is the neighborhood pixel screening coefficient at the pixel point (x, y), and are the local maximum and local minimum in a 5×5 neighborhood centered on the pixel point (x, y), It is the variance of all pixel values ​​in a 5×5 neighborhood centered on the pixel point (x, y), where x is the horizontal coordinate index of the pixel point in the image, and y is the vertical coordinate index of the pixel point in the image. is an exponential function;

[0017] Step S232: Calculate the local minimum divisible amount; take the product of the neighborhood pixel screening coefficient and the local minimum value as the local minimum divisible amount at each pixel point;

[0018] Step S233: Design a pixel calibration factor; calculate the pixel calibration factor at each pixel point according to the local minimum divisible amount and the local maximum value; the formula used is as follows:

[0019] ;

[0020] In the formula, is the pixel calibration factor at the pixel point (x, y), is the local minimum divisible value at the pixel point (x, y);

[0021] Step S234: initial enhancement: performing initial enhancement processing on each pixel point according to the pixel adjustment factor, the local minimum divisible amount and the local maximum value; the formula used is as follows:

[0022] ;

[0023] In the formula, is the normalized pixel value at the pixel point (x, y), yes Pixel value after initial enhancement;

[0024] Step S24: secondary image enhancement: for each frame of the image in the learning image sequence after the initial enhancement, the following secondary enhancement processing is performed in sequence:

[0025] Step S241: re-normalization: re-normalizing the image after the initial enhancement according to the minimum-maximum normalization method;

[0026] Step S242: Calculate the enhancement upper limit value and the enhancement lower limit value; calculate the enhancement upper limit value and the enhancement lower limit value at each pixel point in the image according to the mean and standard deviation of all pixel values ​​in the image after normalization again; the formula used is as follows:

[0027] ;

[0028] ;

[0029] In the formula, and are the enhancement upper limit and lower limit at the pixel point (x, y), yes According to the pixel value processed by the minimum-maximum normalization method, β is the adjustment factor, μ and σ are the mean and standard deviation of all pixel values ​​in the image after re-normalization;

[0030] Step S243: secondary enhancement; performing secondary enhancement processing on each pixel point according to the enhancement upper limit value and the enhancement lower limit value; the formula used is as follows:

[0031] ;

[0032] In the formula, is the pixel value after secondary enhancement at the pixel point (x, y);

[0033] Step S25: gamma correction; performing gamma correction processing on each frame of the image after secondary enhancement in the learning image sequence to obtain a learning image enhancement sequence.

[0034] Furthermore, in step S3, the construction of the student classroom learning status evaluation model specifically includes the following steps:

[0035] Step S31: feature extraction: for each frame of the image in the learning image enhancement sequence, the following feature extraction processes are performed in sequence:

[0036] Step S311: low-level feature extraction: input the image into the first convolutional layer for processing, and extract a low-level feature map;

[0037] Step S312: extracting high-level features; inputting the low-level feature map into the depth-separable convolutional layer, the second convolutional layer, the average pooling layer and the fully connected layer for processing, and extracting a high-level feature map;

[0038] Step S32: low-level feature enhancement: for the low-level feature map extracted from each frame of the image in the learning image enhancement sequence, the following enhancement processes are performed in sequence:

[0039] Step S321: Calculate global features; perform global average pooling on the low-level feature map to obtain global features Q;

[0040] Step S322: construct a cosine similarity matrix; construct a cosine similarity matrix E according to the cosine similarity between the element value at each position in the low-level feature map and the global feature, and quantize the cosine similarity to obtain F quantization levels ; Among them, S 1 , S f and S F are the 1st, fth and Fth quantization levels, where f is the index;

[0041] Step S323: construct a quantization matrix; perform quantization processing on the value of each position in the cosine similarity matrix E to obtain a quantization matrix H of size R×F; the formula used is as follows:

[0042] ;

[0043] In the formula, is the value of the cosine similarity matrix E at position (a, d), r is the index, is the value of the quantization matrix H at position (r, f), and R is the number of elements in the low-level feature map;

[0044] Step S324: construct a two-dimensional quantization count matrix; generate a two-dimensional quantization count matrix L according to the quantization matrix H, wherein the first dimension in the two-dimensional quantization count matrix represents each quantization level, and the second dimension represents the corresponding normalized count; the formula used is as follows:

[0045] ;

[0046] Where, L f is the normalized count corresponding to the fth quantization level, is a join operation;

[0047] Step S325: Calculate statistical features; process the two-dimensional quantized count matrix L through a multi-layer perceptron and connect it with the global feature Q to obtain the statistical feature Z;

[0048] Step S326: Calculate the enhanced low-level features; Calculate the enhanced low-level features according to the statistical features Z and the quantization matrix H; the formula used is as follows:

[0049] ;

[0050] Where V is the enhanced low-level feature, T is the matrix transposition operation, is a standard convolution of size 1×1, is the Softmax activation function;

[0051] Step S33: high-level feature enhancement; for each high-level feature map O extracted from the image enhancement sequence, enhancement processing is performed in sequence; the enhancement processing is to first calculate the attention weight W of the high-level feature map; then the high-level feature map is weighted summed by the attention weight to obtain the enhanced high-level feature P;

[0052] Step S34: constructing a comprehensive feature sequence; merging the enhanced low-level feature V and the enhanced high-level feature P through a connection operation to obtain a comprehensive feature M for each frame image, and integrating the comprehensive features of all frame images in the order of the learning image enhancement sequence to obtain a comprehensive feature sequence;

[0053] Step S35: Evaluation: Input the comprehensive feature sequence into the long short-term memory network for evaluation to obtain the status score of the student's classroom learning.

[0054] Furthermore, in step S4, the generation of the evaluation report is to collect real-time student classroom learning videos and perform student classroom learning image enhancement processing, and then input them into the student classroom learning status evaluation model for evaluation, and generate an evaluation report on the student classroom learning based on the output status score.

[0055] The student classroom learning status evaluation system based on artificial intelligence provided by the present invention includes a student classroom learning data acquisition module, a student classroom learning image enhancement module, a student classroom learning status evaluation model construction module and an evaluation report generation module;

[0056] The student classroom learning data collection module collects historical student classroom learning data and sends the data to the student classroom learning image enhancement module;

[0057] The student classroom learning image enhancement module performs normalization processing on each frame of the image, performs primary enhancement on the image by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount and the pixel adjustment factor, obtains the enhancement upper limit value and the enhancement lower limit value by calculating the mean and standard deviation of the image, performs secondary enhancement on the image, and then performs gamma correction processing to obtain a learning image enhancement sequence, and sends the data to the module for constructing a student classroom learning status evaluation model;

[0058] The module for constructing a student classroom learning status evaluation model extracts low-level features using standard convolution, extracts high-level features using depthwise separable convolution, constructs a cosine similarity matrix through global features, calculates a quantization matrix and a two-dimensional quantization count matrix, obtains statistical features, enhances low-level features, enhances high-level features through attention weights, constructs a comprehensive feature sequence, obtains a status score, and sends the data to a module for generating an evaluation report;

[0059] The evaluation report generation module collects real-time student classroom learning videos and performs student classroom learning image enhancement processing, and then inputs them into the student classroom learning status evaluation model for evaluation, and generates a student classroom learning evaluation report based on the output status score.

[0060] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0061] (1) In view of the problems in existing methods for evaluating students' classroom learning status, such as image quality differences, noise interference, and rapid and complex movements that make images blurred and lack details, resulting in low accuracy in evaluating students' classroom learning status, this scheme normalizes each frame of the image to eliminate equipment and environmental differences and ensure input consistency; by calculating the neighborhood pixel screening coefficient, the local minimum divisible value, and the pixel adjustment factor, the image is initially enhanced to remove noise and retain learning-related features, thereby enhancing the key details of the image; by calculating the mean and standard deviation of the image to obtain the enhancement upper and lower limits, the image is secondary enhanced to reasonably control the image enhancement range, avoid distortion, and make the details clearer; and then gamma correction is performed to obtain a learning image enhancement sequence to ensure image brightness consistency, improve the stability of the evaluation model, and significantly improve the quality of each frame of the image, thereby improving the accuracy of the student learning status evaluation.

[0062] (2) The existing methods for evaluating students’ classroom learning status have the problem of being unable to accurately capture the subtle dynamic changes and complex dynamic behaviors in the students’ learning process, and ignoring the detailed changes in the image, resulting in insufficient accuracy in evaluating students’ classroom learning status. This scheme extracts low-level features through standard convolution and extracts high-level features through deep separable convolution, thereby enhancing the comprehensive reflection of students’ learning status. By calculating the global features to construct the cosine similarity matrix, and then calculating the quantization matrix and the two-dimensional quantization count matrix to obtain statistical features, the low-level features are enhanced, which can not only capture the basic behavioral features, but also reflect the subtle changes in the learning status. By enhancing the high-level features through the attention weights, the sensitivity to the changes in details can be enhanced, thereby improving the recognition accuracy of students’ behavioral status. By constructing a comprehensive feature sequence and obtaining the status score through the long short-term memory network, the status score can be comprehensively considered, the students’ behavioral performance in the classroom environment can be comprehensively reflected, and a more refined classroom learning status score can be provided, thereby enhancing the accuracy and stability of the evaluation. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flow chart of a method for evaluating student classroom learning status based on artificial intelligence provided by the present invention;

[0064] Figure 2 A schematic diagram of a student classroom learning status assessment system based on artificial intelligence provided by the present invention;

[0065] Figure 3 is a schematic flow chart of step S2;

[0066] Figure 4 It is a schematic diagram of the process of step S3.

[0067] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0069] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0070] Example 1, see Figure 1 The present invention provides an artificial intelligence-based student classroom learning status assessment method, which comprises the following steps:

[0071] Step S1: Collecting student classroom learning data; Collecting historical student classroom learning data;

[0072] Step S2: Students learn image enhancement in class; normalize each frame of the image, perform primary enhancement on the image by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount and the pixel adjustment factor, calculate the mean and standard deviation of the image to obtain the enhancement upper limit and enhancement lower limit, perform secondary enhancement on the image, and then perform gamma correction to obtain a learning image enhancement sequence;

[0073] Step S3: Construct a student classroom learning status evaluation model; use standard convolution to extract low-level features, use depthwise separable convolution to extract high-level features, construct a cosine similarity matrix through global features, calculate the quantization matrix and the two-dimensional quantization count matrix, obtain statistical features, enhance low-level features, enhance high-level features through attention weights, construct a comprehensive feature sequence, and obtain a status score;

[0074] Step S4: Generate an evaluation report; collect real-time student classroom learning videos and perform student classroom learning image enhancement processing, and then input them into the student classroom learning status evaluation model for evaluation, and generate an evaluation report on the student classroom learning based on the output status score.

[0075] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, the student classroom learning data collection is to collect historical student classroom learning data. The historical student classroom learning data includes student classroom learning videos and status scores. The status scores are used as data labels.

[0076] Example 3, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S2, the student classroom learning image enhancement specifically includes the following steps:

[0077] Step S21: construct a learning image sequence; extract 30 frames of images per second to ensure that every important moment in the student's learning process is captured and avoid missing key information due to discontinuous data; extract continuous frames at a rate of 30 frames per second from the student's classroom learning video to generate a learning image sequence;

[0078] Step S22: Normalization; The pixel values ​​in the video image may be different under the influence of different video devices, different light, etc. Through normalization processing, the influence of device and environment differences on image quality evaluation is eliminated, making the subsequent image enhancement processing more reliable; each frame image in the learning image sequence is normalized, and the pixel value of the image is scaled from [0, 255] to [0, 1]; the formula used is as follows:

[0079] ;

[0080] In the formula, and are the original pixel value and the normalized pixel value at the pixel point (x, y), respectively. x and y are the horizontal and vertical coordinate indexes of the pixel point in the image, respectively.

[0081] Step S23: primary image enhancement; in the student classroom learning video, the student's attention and behavior may vary greatly in the local area. Through the primary image enhancement, the local features of the image are strengthened and the noise interference is effectively suppressed, so that the subsequent learning state recognition is more accurate; for each normalized frame image in the learning image sequence, the following primary enhancement processing is performed in sequence:

[0082] Step S231: Design neighborhood pixel screening coefficients; images in students’ classroom learning videos often cause local image distortion due to background noise, light changes or students’ rapid movements. These distortions will affect the correct identification of students’ learning behaviors. By calculating the neighborhood pixel screening coefficients, the noise in the local area can be effectively removed while retaining the image features related to students’ learning behaviors; the neighborhood pixel screening coefficients at each pixel in the image are calculated based on the local maximum, local minimum and pixel value variance in the neighborhood of each pixel point; the formula used is as follows:

[0083] ;

[0084] In the formula, is the neighborhood pixel screening coefficient at the pixel point (x, y), and are the local maximum and local minimum in a 5×5 neighborhood centered on the pixel point (x, y), , , is the normalized pixel value at pixel point (i, j), is the variance of all pixel values ​​in a 5×5 neighborhood centered on the pixel point (x, y), where x and i are the horizontal coordinate indexes of the pixel points in the image, and y and j are the vertical coordinate indexes of the pixel points in the image. , and They are exponential function, maximum function and minimum function respectively;

[0085] Step S232: Calculate the local minimum divisible amount; each frame of the student classroom learning video may have uneven brightness and contrast. By calculating the local minimum divisible amount, the effectiveness and contribution of each pixel can be clarified, and inefficient information can be suppressed; the product of the neighborhood pixel screening coefficient and the local minimum value is used as the local minimum divisible amount at each pixel point; the formula used is as follows:

[0086] ;

[0087] In the formula, is the local minimum divisible value at the pixel point (x, y);

[0088] Step S233: Design pixel calibration factors; the brightness and contrast of different video frames are inconsistent, which will lead to the loss of some details and affect the judgment of students' learning status. Through the pixel calibration factors, the brightness and contrast can be adjusted in detail in the local area of ​​the image, making the key behavioral characteristics of students more obvious; the pixel calibration factor at each pixel point is calculated according to the local minimum divisible amount and the local maximum value; the formula used is as follows:

[0089] ;

[0090] In the formula, is the pixel calibration factor at the pixel point (x, y), is the local minimum divisible value at the pixel point (x, y);

[0091] Step S234: initial enhancement; the initial enhancement of the image avoids the evaluation deviation caused by light or shooting angle by highlighting the key features; each pixel is initially enhanced according to the pixel adjustment factor, the local minimum divisible amount and the local maximum value; the formula used is as follows:

[0092] ;

[0093] In the formula, is the normalized pixel value at the pixel point (x, y), yes Pixel value after initial enhancement;

[0094] Step S24: secondary image enhancement; some images may be difficult to distinguish details due to low brightness or contrast, which affects the analysis of learning status. The secondary enhancement process not only improves the contrast of the image, but also makes the details of each pixel clearer through the calculation of the mean and standard deviation, thereby helping the evaluation system to identify the students' subtle learning behaviors; for each frame of the image in the learning image sequence after the initial enhancement, the following secondary enhancement processes are performed in sequence:

[0095] Step S241: re-normalization; the image after the initial enhancement may have fluctuations in pixel values. Re-normalization can remap the pixel values ​​of the image to a uniform range, eliminate the extreme value changes caused by the initial enhancement, balance the details of the image, and ensure the stability of subsequent processing steps; the image after the initial enhancement is normalized again according to the minimum-maximum normalization method; the formula used is as follows:

[0096] ;

[0097] In the formula, yes According to the pixel value processed by the minimum-maximum normalization method, and are the minimum pixel value and maximum pixel value of the image after the initial enhancement;

[0098] Step S242: Calculate the enhancement upper limit and the enhancement lower limit; in the process of image enhancement, some pixel values ​​may be over-enhanced or over-compressed. By calculating the enhancement upper limit and the enhancement lower limit, it can be ensured that the image enhancement effect is within a reasonable range, and the image will not be distorted due to over-enhancement, nor will key information be lost due to insufficient enhancement, which can more accurately reflect the subtle changes in the student's learning state; according to the mean and standard deviation of all pixel values ​​in the image after re-normalization, the enhancement upper limit and the enhancement lower limit at each pixel point in the image are calculated; the formula used is as follows:

[0099] ;

[0100] ;

[0101] In the formula, and are the enhancement upper limit and lower limit at the pixel point (x, y), yes According to the pixel value processed by the minimum-maximum normalization method, β is the adjustment factor in the range of (0, 1), μ and σ are the mean and standard deviation of all pixel values ​​in the image after re-normalization;

[0102] Step S243: secondary enhancement; secondary enhancement ensures that the details of the image remain in a natural and balanced state after enhancement, which helps to enhance the accurate recognition of students' learning status, especially the accurate capture of small behaviors and emotional changes; secondary enhancement processing is performed on each pixel according to the enhancement upper limit value and the enhancement lower limit value; the formula used is as follows:

[0103] ;

[0104] In the formula, is the pixel value after secondary enhancement at the pixel point (x, y);

[0105] Step S25: gamma correction; gamma correction adjusts the brightness of the image to an appropriate range, so that images from different video sources and environmental conditions can be processed uniformly, improving the stability and consistency of the model analysis; gamma correction is performed on each frame of the image after secondary enhancement in the learning image sequence to obtain a learning image enhancement sequence; the formula used is as follows:

[0106] ;

[0107] In the formula, is the pixel value after gamma correction at the pixel point (x, y), γ is the gamma value, is the maximum pixel value of the image after secondary enhancement.

[0108] By performing the above operations, in order to solve the problem that the existing methods for evaluating students' classroom learning status have image quality differences, noise interference, and fast and complex movements that make the image blurred and lack details, resulting in low accuracy in evaluating students' classroom learning status, this scheme normalizes each frame of the image to eliminate equipment and environmental differences and ensure input consistency; by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount, and the pixel adjustment factor, the image is initially enhanced to remove noise and retain learning-related features, thereby enhancing the key details of the image; by calculating the mean and standard deviation of the image to obtain the enhancement upper limit and enhancement lower limit, the image is secondary enhanced to reasonably control the image enhancement range, avoid distortion, and make the details clearer; and then gamma correction is performed to obtain a learning image enhancement sequence to ensure image brightness consistency, improve the stability of the evaluation model, and significantly improve the quality of each frame of the image, thereby improving the accuracy of the student learning status evaluation.

[0109] Example 4, see Figure 1 and Figure 4 Based on the above embodiment, in step S3, building a student classroom learning status evaluation model specifically includes the following steps:

[0110] Step S31: feature extraction: for each frame of the image in the learning image enhancement sequence, the following feature extraction processes are performed in sequence:

[0111] Step S311: low-level feature extraction; the learning status of students in the classroom is reflected by details such as expressions and postures. Traditional image processing methods are difficult to accurately capture these subtle changes. By extracting low-level features through the convolution layer, the basic behavioral characteristics of students can be effectively captured from the image; the image is input to the first convolution layer for processing, and a low-level feature map is extracted; the first convolution layer is composed of a standard convolution of size 3×3;

[0112] Step S312: high-level feature extraction; although low-level features can provide basic information, they cannot capture the more complex dynamic behaviors of students in the learning process. By extracting high-level features through the deep separable convolution layer, more complex behavior patterns can be captured, reflecting the comprehensive performance of students in the classroom; the low-level feature map is input into the deep separable convolution layer, the second convolution layer, the average pooling layer and the fully connected layer for processing, and the high-level feature map is extracted; the deep separable convolution layer is composed of a deep separable convolution with a size of 3×3 and an expansion factor of 1, a deep separable convolution with a size of 3×3 and an expansion factor of 6, a deep separable convolution with a size of 5×5 and an expansion factor of 6, two deep separable convolutions with a size of 3×3 and an expansion factor of 6, a deep separable convolution with a size of 5×5 and an expansion factor of 6, and a deep separable convolution with a size of 3×3 and an expansion factor of 6. The second convolution layer is composed of a standard convolution with a size of 1×1;

[0113] Step S32: low-level feature enhancement; low-level features may not fully express the complex behavioral characteristics related to the student's learning status, especially in dynamic scenes. Low-level feature enhancement improves the sensitivity of low-level features to the evaluation of the student's learning status; for the low-level feature map extracted from each frame of the learning image enhancement sequence, the following enhancement processes are performed in sequence:

[0114] Step S321: Calculate global features; individual local features are often easily affected by noise and are difficult to reflect the overall learning status of students. Especially when dealing with complex classroom environments, global features can be calculated by global average pooling, which can better capture the global information of the image and avoid over-reliance on local features, so that the model can obtain a more comprehensive and stable learning status evaluation; perform global average pooling on low-level feature maps to obtain global features Q; the formula used is as follows:

[0115] ;

[0116] Where A and D are the width and height of the low-level feature map, a and d are the row index and column index, respectively. is the element value of the low-level feature map at position (a, d);

[0117] Step S322: construct a cosine similarity matrix; the features of different regions have different effects on the student's learning state. Traditional methods tend to ignore the importance of local features, which makes it difficult for the model to capture subtle changes in the learning state. Through cosine similarity, the importance of each region can be quantified, which helps to focus on the most critical areas for evaluating the student's learning state, thereby improving accuracy and robustness; according to the cosine similarity between the element value at each position in the low-level feature map and the global feature, a cosine similarity matrix E of size A×D is constructed, and the value of the cosine similarity matrix at position (a, d) is the cosine similarity between the element value at position (a, d) of the low-level feature map and the global feature, and the cosine similarity is quantized to obtain F quantization levels ; The formula used is as follows:

[0118] ;

[0119] In the formula, S 1 , S f and S F are the 1st, fth, and Fth quantization levels, f is the index, E min and E max are the minimum and maximum values ​​of cosine similarity respectively;

[0120] Step S323: construct a quantization matrix; the feature values ​​in the image vary in a complex manner, and directly using these values ​​may lead to unstable or inconsistent results, affecting the accuracy of the evaluation. The process of quantizing the cosine similarity matrix converts it into discrete values, so that the model can handle more accurate feature expressions and reduce the impact of differences between different images; quantize the value of each position in the cosine similarity matrix E to obtain a quantization matrix H of size R×F, where R=A×D; the formula used is as follows:

[0121] ;

[0122] In the formula, is the value of the cosine similarity matrix E at position (a, d), r is the index, is the value of the quantization matrix H at position (r, f), and R is the number of elements in the low-level feature map;

[0123] Step S324: construct a two-dimensional quantization count matrix; traditional methods may not be able to effectively handle the relative importance and distribution differences between different features, resulting in information loss or misjudgment. Through the two-dimensional quantization count matrix, the quantized features can be counted to obtain the distribution and frequency of the features, so that the model can better understand the distribution of the features, which helps to comprehensively consider the different behavioral performances of students in the classroom and provide a comprehensive evaluation basis; generate a two-dimensional quantization count matrix L based on the quantization matrix H, the first dimension in the two-dimensional quantization count matrix represents each quantization level, and the second dimension represents the corresponding normalized count; the formula used is as follows:

[0124] ;

[0125] Where, L f is the normalized count corresponding to the fth quantization level, is a join operation;

[0126] Step S325: Calculate statistical features; the multilayer perceptron can effectively combine the two-dimensional quantized count matrix with the global features to generate more abstract and comprehensive statistical features; the two-dimensional quantized count matrix L is processed by the multilayer perceptron and then connected with the global feature Q to obtain the statistical feature Z; the formula used is as follows:

[0127] ;

[0128] In the formula, is a multi-layer perceptron;

[0129] Step S326: Calculate the enhanced low-level features; the enhanced low-level features can more accurately reflect the student's learning status; calculate the enhanced low-level features according to the statistical features Z and the quantization matrix H; the formula used is as follows:

[0130] ;

[0131] Where V is the enhanced low-level feature, T is the matrix transposition operation, is a standard convolution of size 1×1, is the Softmax activation function;

[0132] Step S33: Advanced feature enhancement; Advanced features may sometimes lack sensitivity to certain specific details, especially when student behaviors are more complex or subtle, which are difficult to be accurately captured by traditional methods. By introducing the attention mechanism, attention can be automatically focused on the feature areas that are most valuable to the evaluation results, avoiding interference from irrelevant information, especially in a classroom environment. The subtle changes of students can be effectively identified through the attention mechanism; For each frame of the image in the learning image enhancement sequence, the advanced feature map O extracted is enhanced in turn; the enhancement process is to first calculate the attention weight of the advanced feature map ; Then the high-level feature map is weighted summed by attention weights to obtain enhanced high-level features ; where d k is the characteristic dimension;

[0133] Step S34: construct a comprehensive feature sequence; using low-level features or high-level features alone may not fully reflect the students' learning status, especially in terms of the complexity and diversity of classroom behavior. Combining low-level features with high-level features can more comprehensively express the students' learning status, integrating information at different levels from details to the whole, so that the model can more accurately evaluate the students' learning status; the enhanced low-level features V and the enhanced high-level features P are merged through a connection operation to obtain the comprehensive features of each frame image , and integrate the comprehensive features of all frame images according to the order in the learning image enhancement sequence to obtain a comprehensive feature sequence;

[0134] Step S35: Evaluation: Input the comprehensive feature sequence into the long short-term memory network for evaluation to obtain the status score of the student's classroom learning.

[0135] By performing the above operations, the existing methods for evaluating students' classroom learning status have the problem of being difficult to accurately capture the subtle dynamic changes and complex dynamic behaviors in the students' learning process, ignoring the detailed changes in the image, resulting in insufficient accuracy in evaluating students' classroom learning status. This scheme extracts low-level features through standard convolution and uses deep separable convolution to extract high-level features, thereby enhancing the comprehensive reflection of students' learning status; constructs a cosine similarity matrix by calculating global features, and then calculates a quantization matrix and a two-dimensional quantization count matrix to obtain statistical features, and enhances low-level features, which can not only capture basic behavioral features, but also reflect subtle changes in learning status; enhances high-level features through attention weights, which can enhance sensitivity to changes in details and improve the recognition accuracy of students' behavioral status; constructs a comprehensive feature sequence, and obtains a status score through a long short-term memory network, which can comprehensively consider students' behavioral performance in the classroom environment, comprehensively reflect students' learning status, provide more refined classroom learning status scores, and enhance the accuracy and stability of the evaluation.

[0136] Example 5, see Figure 2 , this embodiment is based on the above embodiment, and the student classroom learning status evaluation system based on artificial intelligence provided by the present invention includes a student classroom learning data acquisition module, a student classroom learning image enhancement module, a student classroom learning status evaluation model construction module and an evaluation report generation module;

[0137] The student classroom learning data collection module collects historical student classroom learning data and sends the data to the student classroom learning image enhancement module;

[0138] The student classroom learning image enhancement module performs normalization processing on each frame of the image, performs primary enhancement on the image by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount and the pixel adjustment factor, obtains the enhancement upper limit value and the enhancement lower limit value by calculating the mean and standard deviation of the image, performs secondary enhancement on the image, and then performs gamma correction processing to obtain a learning image enhancement sequence, and sends the data to the module for constructing a student classroom learning status evaluation model;

[0139] The module for constructing a student classroom learning status evaluation model extracts low-level features using standard convolution, extracts high-level features using depthwise separable convolution, constructs a cosine similarity matrix through global features, calculates a quantization matrix and a two-dimensional quantization count matrix, obtains statistical features, enhances low-level features, enhances high-level features through attention weights, constructs a comprehensive feature sequence, obtains a status score, and sends the data to a module for generating an evaluation report;

[0140] The evaluation report generation module collects real-time student classroom learning videos and performs student classroom learning image enhancement processing, and then inputs them into the student classroom learning status evaluation model for evaluation, and generates a student classroom learning evaluation report based on the output status score.

[0141] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0142] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0143] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for evaluating students' classroom learning status based on artificial intelligence, characterized by: The method comprises the following steps: Step S1: Collecting student classroom learning data; Collecting historical student classroom learning data; Step S2: Students learn image enhancement in class; Step S3: Constructing a model for evaluating students’ classroom learning status; Step S4: Generate an evaluation report; collect real-time student classroom learning videos and perform student classroom learning image enhancement processing, and then input them into the student classroom learning status evaluation model for evaluation, and generate a student classroom learning evaluation report based on the output status score; Step S2 includes step S231: designing a neighborhood pixel screening coefficient; calculating the neighborhood pixel screening coefficient at each pixel point in the image according to the local maximum value, local minimum value and pixel value variance in the neighborhood of each pixel point; the formula used is as follows: ； In the formula, is the neighborhood pixel screening coefficient at the pixel point (x, y), and are the local maximum and local minimum in a 5×5 neighborhood centered on the pixel point (x, y), It is the variance of all pixel values ​​in a 5×5 neighborhood centered on the pixel point (x, y), where x is the horizontal coordinate index of the pixel point in the image, and y is the vertical coordinate index of the pixel point in the image. is an exponential function.

2. The method for evaluating students' classroom learning status based on artificial intelligence according to claim 1 is characterized by: In step S2, the student classroom learning image enhancement specifically includes the following steps: Step S21: constructing a learning image sequence; extracting continuous frames from the student's classroom learning video to generate a learning image sequence; Step S22: normalization: normalizing each frame of the image in the learning image sequence; Step S23: initial image enhancement; Step S24: secondary image enhancement: for each frame of the image in the learning image sequence after the initial enhancement, the following secondary enhancement processing is performed in sequence: Step S241: re-normalization: re-normalizing the image after the initial enhancement according to the minimum-maximum normalization method; Step S242: Calculate the enhancement upper limit value and the enhancement lower limit value; calculate the enhancement upper limit value and the enhancement lower limit value at each pixel point in the image according to the mean and standard deviation of all pixel values ​​in the image after normalization again; the formula used is as follows: ； ； In the formula, and are the enhancement upper limit and lower limit at the pixel point (x, y), yes According to the pixel value processed by the minimum-maximum normalization method, yes The pixel value after the initial enhancement, is the normalized pixel value at the pixel point (x, y), β is the adjustment factor, μ and σ are the mean and standard deviation of all pixel values ​​in the image after normalization; Step S243: secondary enhancement; performing secondary enhancement processing on each pixel point according to the enhancement upper limit value and the enhancement lower limit value; the formula used is as follows: ； In the formula, is the pixel value after secondary enhancement at the pixel point (x, y); Step S25: gamma correction; performing gamma correction processing on each frame of the image after secondary enhancement in the learning image sequence to obtain a learning image enhancement sequence.

3. The method for evaluating students' classroom learning status based on artificial intelligence according to claim 2 is characterized by: In step S23, the image is enhanced for the first time; for each normalized frame image in the learning image sequence, the following initial enhancement processing is performed in sequence: Step S231: designing neighborhood pixel screening coefficients; Step S232: Calculate the local minimum divisible amount; take the product of the neighborhood pixel screening coefficient and the local minimum value as the local minimum divisible amount at each pixel point; Step S233: Design a pixel calibration factor; calculate the pixel calibration factor at each pixel point according to the local minimum divisible amount and the local maximum value; the formula used is as follows: ； In the formula, is the pixel calibration factor at the pixel point (x, y), is the local minimum divisible value at the pixel point (x, y); Step S234: initial enhancement: performing initial enhancement processing on each pixel point according to the pixel adjustment factor, the local minimum divisible amount and the local maximum value; the formula used is as follows: ； In the formula, is the normalized pixel value at the pixel point (x, y), yes Pixel value after initial enhancement.

4. The method for evaluating students' classroom learning status based on artificial intelligence according to claim 1 is characterized by: In step S3, the construction of the student classroom learning status evaluation model specifically includes the following steps: Step S31: feature extraction: for each frame of the image in the learning image enhancement sequence, the following feature extraction processes are performed in sequence: Step S311: low-level feature extraction: input the image into the first convolutional layer for processing, and extract a low-level feature map; Step S312: extracting high-level features; inputting the low-level feature map into the depth-separable convolutional layer, the second convolutional layer, the average pooling layer and the fully connected layer for processing, and extracting a high-level feature map; Step S32: low-level feature enhancement; Step S33: high-level feature enhancement; for each high-level feature map O extracted from the image enhancement sequence, enhancement processing is performed in sequence; the enhancement processing is to first calculate the attention weight W of the high-level feature map; then the high-level feature map is weighted summed by the attention weight to obtain the enhanced high-level feature P; Step S34: constructing a comprehensive feature sequence; merging the enhanced low-level feature V and the enhanced high-level feature P through a connection operation to obtain a comprehensive feature M for each frame image, and integrating the comprehensive features of all frame images in the order of the learning image enhancement sequence to obtain a comprehensive feature sequence; Step S35: Evaluation: Input the comprehensive feature sequence into the long short-term memory network for evaluation to obtain the status score of the student's classroom learning.

5. The method for evaluating students' classroom learning status based on artificial intelligence according to claim 4 is characterized by: In step S32, the low-level features are enhanced; For the low-level feature map extracted from each frame of the learning image enhancement sequence, the following enhancement processes are performed in sequence: Step S321: Calculate global features; Perform global average pooling on the low-level feature map to obtain the global feature Q; Step S322: construct a cosine similarity matrix; construct a cosine similarity matrix E according to the cosine similarity between the element value at each position in the low-level feature map and the global feature, and quantize the cosine similarity to obtain F quantization levels ; Among them, S1, S f and S F are the 1st, fth and Fth quantization levels, where f is the index; Step S323: construct a quantization matrix; perform quantization processing on the value of each position in the cosine similarity matrix E to obtain a quantization matrix H of size R×F; the formula used is as follows: ； In the formula, is the value of the cosine similarity matrix E at position (a, d), r is the index, is the value of the quantization matrix H at position (r, f), and R is the number of elements in the low-level feature map; Step S324: construct a two-dimensional quantization count matrix; generate a two-dimensional quantization count matrix L according to the quantization matrix H, wherein the first dimension in the two-dimensional quantization count matrix represents each quantization level, and the second dimension represents the corresponding normalized count; the formula used is as follows: ； Where, L f is the normalized count corresponding to the fth quantization level, is a join operation; Step S325: Calculate statistical features; process the two-dimensional quantized count matrix L through a multi-layer perceptron and connect it with the global feature Q to obtain the statistical feature Z; Step S326: Calculate the enhanced low-level features; Calculate the enhanced low-level features according to the statistical features Z and the quantization matrix H; the formula used is as follows: ； Where V is the enhanced low-level feature, T is the matrix transposition operation, is a standard convolution of size 1×1, is the Softmax activation function.

6. The method for evaluating students' classroom learning status based on artificial intelligence according to claim 1 is characterized by: In step S4, the student classroom learning data collection is to collect historical student classroom learning data, and the historical student classroom learning data includes student classroom learning videos and status scores, and the status scores are used as data labels.

7. An artificial intelligence-based student classroom learning status evaluation system, used to implement an artificial intelligence-based student classroom learning status evaluation method as described in any one of claims 1 to 6, characterized in that: It includes a student classroom learning data collection module, a student classroom learning image enhancement module, a student classroom learning status evaluation model building module and an evaluation report generation module; The student classroom learning data collection module collects historical student classroom learning data and sends the data to the student classroom learning image enhancement module; The student classroom learning image enhancement module performs normalization processing on each frame of the image, performs primary enhancement on the image by calculating the neighborhood pixel screening coefficient, the local minimum divisible amount and the pixel adjustment factor, obtains the enhancement upper limit value and the enhancement lower limit value by calculating the mean and standard deviation of the image, performs secondary enhancement on the image, and then performs gamma correction processing to obtain a learning image enhancement sequence, and sends the data to the module for constructing a student classroom learning status evaluation model; The module for constructing a student classroom learning status evaluation model extracts low-level features using standard convolution, extracts high-level features using depthwise separable convolution, constructs a cosine similarity matrix through global features, calculates a quantization matrix and a two-dimensional quantization count matrix, obtains statistical features, enhances low-level features, enhances high-level features through attention weights, constructs a comprehensive feature sequence, obtains a status score, and sends the data to a module for generating an evaluation report; The evaluation report generation module collects real-time student classroom learning videos and performs student classroom learning image enhancement processing, and then inputs them into the student classroom learning status evaluation model for evaluation, and generates a student classroom learning evaluation report based on the output status score.

Images