Learning status evaluation method and system based on artificial intelligence

Abstract

The present invention provides an artificial intelligence-based learning state assessment method and system, which relate to the field of artificial intelligence technology. In the present invention, first, first learning state data and a first number of second learning state data of a target student during the learning process are obtained; second, a target learning state assessment network is used to perform learning state assessment processing on the first learning state data and the first number of second learning state data, outputting a corresponding first number of learning state assessment results; and then, based on the first number of learning state assessment results, a target learning state assessment result for the target student is determined. Based on the above method, the relatively poor reliability of learning state assessments in the prior art can be improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based learning status assessment method and system. Background Art

[0002] After students have learned, strengthening their learning for their weak links is one of the important means to ensure the quality of their learning. Among them, in traditional technical solutions, weak links in learning are generally determined based on students' feedback. This method is inefficient and unreliable. Alternatively, students' learning status is determined based on teachers' observations, that is, their weak links are determined. This also has the problem of low efficiency and low reliability (it is difficult to effectively cover the entire learning process of all students). In other words, in the existing technology, there is a problem of relatively poor reliability in learning status assessment (due to the poor reliability of learning status assessment, it will lead to the problem of wasted learning resources in the subsequent learning reinforcement process, or the problem of low effective utilization of learning resources). Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a learning status assessment method and system based on artificial intelligence to improve the problem of poor reliability of learning status assessment in the prior art.

[0004] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0005] A learning status assessment method based on artificial intelligence, comprising:

[0006] Acquiring first learning state data and a first number of second learning state data of a target student during a learning process, wherein the first learning state data is image data of the target student's learning behavior, and the second learning state data is image data of the target student's learning expression, and formation times corresponding to the first number of learning expressions reflected by the first number of second learning state data are different from each other, and a duration corresponding to the learning behavior reflected by the first learning state data and a duration corresponding to the first number of learning expressions reflected by the first number of second learning state data have the same starting point and end point;

[0007] Using a target learning state evaluation network, performing learning state evaluation processing on the first learning state data and the first number of second learning state data, and outputting a corresponding first number of learning state evaluation results, wherein, for any one of the first number of learning state evaluation results, a corresponding evaluation basis includes at least the first learning state data and one second learning state data, and the target learning state evaluation network is a neural network formed by training based on sample learning state data and corresponding learning state labels;

[0008] Based on the first number of learning status assessment results, the target learning status assessment results of the target student are determined, wherein the target learning status assessment results refer to the learning status assessment results in the first number of learning status assessment results whose learning status does not meet the predetermined reference learning status, and the target learning status assessment results serve as the basis for strengthening the learning of the target student.

[0009] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the step of using the target learning state assessment network to perform learning state assessment processing on the first learning state data and the first number of second learning state data, and outputting the corresponding first number of learning state assessment results includes:

[0010] Performing feature mining on the first learning state data to output a corresponding first state semantic feature, and outputting a corresponding first number of second state semantic features based on the first state semantic feature and the first number of second learning state data, wherein one second state semantic feature is obtained by mining based on the first state semantic feature and one second learning state data;

[0011] Determine the interference characteristics of the learning state;

[0012] The learning state interference feature is loaded so that a first feature transfer unit in a target learning state evaluation network obtains the learning state interference feature, and the first feature transfer unit processes the learning state interference feature to form a transferred learning state interference feature;

[0013] The transfer learning state interference feature is loaded so that a second feature transfer unit in the target learning state evaluation network obtains the transfer learning state interference feature, and a first number of feature extension links are determined using the second feature transfer unit, wherein the second feature transfer unit is used to input the first state semantic feature in a first feature extension process and input a second state semantic feature for each feature extension link in a second feature extension process, and the input data of the second feature extension process includes the output data of the first feature extension process;

[0014] Based on the transferred learning state interference feature, the first state semantic feature input by the first feature expansion process, and the first number of second state semantic features input by the second feature expansion process, the learning state evaluation results corresponding to each of the first number of feature expansion links are output.

[0015] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the step of performing feature mining on the first learning state data, outputting corresponding first state semantic features, and outputting corresponding first number of second state semantic features based on the first state semantic features and the first number of second learning state data, includes:

[0016] Performing a segmentation operation on the first learning state data to output a plurality of corresponding local learning state data, and performing a pixel expansion operation on each of the local learning state data based on an image size corresponding to the first learning state data to form a plurality of corresponding expanded learning state data, wherein the image size corresponding to each of the expanded learning state data is equal to the image size corresponding to the first learning state data;

[0017] performing a convolution operation on the first learning state data to form a corresponding learning state convolution feature, and performing a convolution operation on each of the local learning state data to form a local learning state convolution feature corresponding to each of the local learning state data, and fusing the learning state convolution feature with the local learning state convolution feature corresponding to each of the local learning state data to output a corresponding first state semantic feature;

[0018] performing a convolution operation on each of the first number of second learning state data, and outputting an initial learning state feature corresponding to each second learning state data;

[0019] For each second learning state data, the initial learning state feature corresponding to the second learning state data and the first state semantic feature are fused to output the second state semantic feature corresponding to the second learning state data.

[0020] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state evaluation method, the step of outputting the learning state evaluation results corresponding to each of the first number of feature extension links based on the transferred learning state interference feature, the first state semantic feature input by the first feature expansion process, and the first number of second state semantic features input by the second feature expansion process includes:

[0021] In the first feature expansion process, the first state semantic feature is inputted using the second feature transfer unit;

[0022] Performing a deep mining operation on the transferred learning state interference feature and the first state semantic feature inputted by the first feature expansion process, and outputting a learning state deep feature corresponding to the first feature expansion process;

[0023] Associating the learning state deep features with the first number of feature extension links;

[0024] In the second feature expansion process, a second state semantic feature is input for each feature expansion link;

[0025] In each feature expansion link, a feature expansion operation is performed on the learning state deep feature and the second state semantic feature input in the second feature expansion process, and a first number of learning state extended features corresponding to the second feature expansion process are output;

[0026] A learning state evaluation operation is performed according to the first number of learning state extension features, and learning state evaluation results corresponding to each of the first number of feature extension links are output.

[0027] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the step of performing a deep mining operation on the transferred learning state interference feature and the first state semantic feature input by the first feature expansion process, and outputting a learning state deep feature corresponding to the first feature expansion process, includes:

[0028] The transfer learning state interference feature is loaded so that the second feature transfer unit obtains the transfer learning state interference feature, wherein the second feature transfer unit includes a feature mining subunit, the feature mining subunit includes a feature compression module, a feature diffusion module and a feature expansion module, the input data of the feature diffusion module includes the output data of the feature compression module, the input data of the feature expansion module includes the output data of the feature diffusion module, and the feature compression module includes a first focusing submodule and a second focusing submodule;

[0029] Using the first focusing submodule, performing a focused mining operation on the transfer learning state interference feature, and outputting a corresponding first learning state focused feature;

[0030] Utilizing the second focusing submodule, the first learning state focusing feature is associated with the input first state semantic feature to perform a focusing mining operation, and the learning state deep feature corresponding to the first feature expansion process is output, wherein the number of dimensions of the transferred learning state interference feature is greater than the number of dimensions of the learning state deep feature, and the feature diffusion module is used to perform a diffusion operation on the learning state deep feature to form a first number of learning state deep features, and to perform a loading operation on the first number of learning state deep features to load them into a feature expansion module used to output a first number of learning state expansion features in the second feature expansion process.

[0031] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the step of using the second focusing submodule to perform an associated focusing mining operation on the first learning state focus feature and the input first state semantic feature, and outputting the learning state deep feature corresponding to the first feature expansion process, includes:

[0032] performing a multiplication operation on the first learning state focus feature and the first spatial mapping parameter in the second focusing submodule, and outputting a corresponding first learning state mapping feature;

[0033] performing a multiplication operation on the first learning state focus feature and the second spatial mapping parameter in the second focusing submodule, and outputting a corresponding second learning state mapping feature;

[0034] Performing a multiplication operation on the first state semantic feature and a third spatial mapping parameter in the second focusing submodule to output a corresponding third learned state mapping feature;

[0035] Performing a multiplication operation on the first learning state mapping feature and the second learning state mapping feature to output a corresponding focus association parameter, and performing an update operation on the focus association parameter according to the number of dimensions of the second learning state mapping feature to output a corresponding updated focus association parameter;

[0036] A multiplication operation is performed on the updated focus association parameter and the third learning state mapping feature to output a corresponding learning state depth feature.

[0037] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the first number of feature extension links includes any feature extension link, and the any feature extension link refers to any feature extension link;

[0038] The step of performing a feature expansion operation on the learning state deep feature and the second state semantic feature input in the second feature expansion process in each feature expansion link, and outputting a first number of learning state extended features corresponding to the second feature expansion process, includes:

[0039] Loading the learning state deep feature so that the feature extension module acquires the learning state deep feature, wherein the feature extension module includes a third focusing submodule and a fourth focusing submodule corresponding to the arbitrary feature extension link;

[0040] Using the third focusing submodule, performing a focused mining operation on the learning state deep features, and outputting corresponding second learning state focused features;

[0041] Using the fourth focusing submodule, performing an association focus mining operation on the second learning state focus feature and the second state semantic feature input by the arbitrary feature extension link, and outputting a corresponding learning state association focus feature;

[0042] Based on the learning state associated focus feature, the learning state extended feature corresponding to the arbitrary feature extension link is determined, wherein the number of dimensions of the transferred learning state interference feature is equal to the number of dimensions of the learning state extended feature.

[0043] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the second feature transfer unit includes a second number of feature mining sub-units connected in cascade, each feature mining sub-unit being configured to input the first state semantic feature in the first feature expansion process, and to input a second state semantic feature for each feature expansion link in the second feature expansion process;

[0044] The second number of feature mining subunits includes any feature mining subunit, and the any feature mining subunit refers to any feature mining subunit;

[0045] The step of outputting the learning state evaluation results corresponding to each of the first number of feature extension links based on the transferred learning state interference feature, the first state semantic feature input by the first feature extension process, and the first number of second state semantic features input by the second feature extension process includes:

[0046] Loading the feature to be processed so that the arbitrary feature mining subunit obtains the feature to be processed, and using the arbitrary feature mining subunit, performing a deep mining operation on the feature to be processed and the first state semantic feature input in the first feature expansion process, outputting a first number of learning state deep features corresponding to the arbitrary feature mining subunit, and performing an association operation on the first number of learning state deep features corresponding to the arbitrary feature mining subunit so as to be associated one-to-one with the first number of feature expansion links corresponding to the arbitrary feature mining subunit, wherein, when the arbitrary feature mining subunit belongs to the first feature mining subunit, the feature to be processed is the transferred learning state interference feature, and the first number of learning state deep features are all the same; when the arbitrary feature mining subunit does not belong to the first feature mining subunit, the feature to be processed includes the first number of learning state expansion features corresponding to the previous feature mining subunit of the arbitrary feature mining subunit, and one learning state deep feature is formed based on a learning state expansion feature corresponding to the previous feature mining subunit of the arbitrary feature mining subunit and the first state semantic feature, and the input data of the arbitrary feature mining subunit includes the output data of the previous feature mining subunit of the arbitrary feature mining subunit;

[0047] In the second feature expansion process, a second state semantic feature is input into each feature expansion link of the arbitrary feature mining subunit, and a feature expansion operation is performed on the learning state deep feature corresponding to the arbitrary feature mining subunit and the input second state semantic feature in each feature expansion link, and a first number of learning state expansion features corresponding to the arbitrary feature mining subunit is output;

[0048] When the arbitrary feature mining subunit is the last feature mining subunit, the learning state evaluation results corresponding to the first number of feature extension links of the arbitrary feature mining subunit are determined according to the first number of learning state extension features corresponding to the arbitrary feature mining subunit.

[0049] In some preferred embodiments, in the above-mentioned artificial intelligence-based learning state assessment method, the step of determining the learning state interference feature includes:

[0050] Determining historical learning status data, wherein the historical learning status data refers to historical image data about the learning behavior of the target student;

[0051] Loading the historical learning state data so that a feature mining unit in a target learning state evaluation network acquires the historical learning state data;

[0052] Using the feature mining unit, performing a first feature mining operation and a second feature mining operation on the historical learning state data, and outputting a first historical state feature and a second historical state feature corresponding to the historical learning state data;

[0053] Determining to filter the first historical state feature and the second historical state feature based on the first historical state feature and the second historical state feature corresponding to the historical learning state data;

[0054] A learning state interference feature is determined based on the screening of the first historical state feature and the screening of the second historical state feature.

[0055] An embodiment of the present invention also provides an artificial intelligence-based learning status assessment system, comprising a processor and a memory, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to implement the above-mentioned artificial intelligence-based learning status assessment method.

[0056] The artificial intelligence-based learning status assessment method and system provided in the embodiments of the present invention can obtain the first learning status data and the first number of second learning status data of the target student during the learning process; secondly, using the target learning status assessment network, the first learning status data and the first number of second learning status data are subjected to learning status assessment processing, and the corresponding first number of learning status assessment results are output; then, based on the first number of learning status assessment results, the target learning status assessment result of the target student is determined. Based on the above method, since when the target learning state evaluation network is used to perform learning state evaluation processing, the evaluation basis corresponding to each learning state evaluation result obtained includes at least first learning state data and one second learning state data, so that the evaluation basis corresponding to each learning state evaluation result includes not only the expression information of the corresponding learning stage, but also the behavior information of the entire learning process. Therefore, the evaluation basis is rich and the evaluation basis is fully correlated, so that the obtained learning state evaluation result is more reliable; in addition, since the state data for the expression dimension will be subdivided, that is, the granularity is smaller and the accuracy is higher, and the state data for the behavior dimension is global, making the semantics richer, such a combination can also further improve the reliability of the evaluation, and thus improve the problem of relatively poor reliability of learning state evaluation in the existing technology (the improvement in reliability can ensure that the resources of subsequent reinforcement learning will not be wasted).

[0057] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1This is a structural block diagram of the artificial intelligence-based learning status assessment system provided by an embodiment of the present invention.

[0059] Figure 2 A flowchart of the steps of the artificial intelligence-based learning status assessment method provided in an embodiment of the present invention.

[0060] Figure 3 This is a flowchart of the internal processing of the first feature transfer unit provided in an embodiment of the present invention.

[0061] Figure 4 This is a flowchart of the internal processing of the second feature transfer unit provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0063] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort shall fall within the scope of protection of the present invention.

[0064] like Figure 1 As shown, an embodiment of the present invention provides a learning status assessment system based on artificial intelligence. The learning status assessment system based on artificial intelligence may include a memory and a processor. In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, they can be electrically connected to each other through one or more communication buses or signal lines. The memory may store at least one software function module (computer program) that can exist in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby realizing the learning status assessment method based on artificial intelligence provided by the embodiment of the present invention (as described later).

[0065] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), etc. The processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on a chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0066] and, Figure 1 The structure shown is for illustration only. The artificial intelligence-based learning status assessment system may also include Figure 1 More or fewer components than shown, or with Figure 1 The different configurations shown, for example, may include a communication unit for exchanging information with other devices. In a specific application example, the artificial intelligence-based learning status assessment system may be a server with data processing capabilities.

[0067] Combine Figure 2 , the embodiment of the present invention also provides a learning status assessment method based on artificial intelligence, which can be applied to the above-mentioned learning status assessment system based on artificial intelligence. Among them, the method steps defined in the process related to the learning status assessment method based on artificial intelligence can be implemented by the learning status assessment system based on artificial intelligence. Figure 2 The process shown is explained.

[0068] Step S110 , obtaining first learning status data and a first number of second learning status data of the target student during the learning process.

[0069] In an embodiment of the present invention, the artificial intelligence-based learning state assessment system can obtain the first learning state data and the first number of second learning state data of the target student during the learning process. The first learning state data is the image data of the target student's learning behavior (which may include multiple frames of images, such as those acquired through a connected image acquisition device, or may be acquired by acquiring images of the target student's entire body), and the second learning state data is the image data of the target student's learning expression (which may include multiple frames of images, or may be acquired by acquiring images of the target student's face). The formation time corresponding to the first number of learning expressions reflected by the first number of second learning state data is different from each other, and the duration corresponding to the learning behavior reflected by the first learning state data and the duration corresponding to the first number of learning expressions reflected by the first number of second learning state data have the same starting point and end point. That is, the first learning state data corresponds to the entire learning process, and the second learning state data corresponds to part of the learning process.

[0070] Step S120 : Using a target learning state evaluation network, perform learning state evaluation processing on the first learning state data and the first number of second learning state data, and output a corresponding first number of learning state evaluation results.

[0071] In an embodiment of the present invention, the artificial intelligence-based learning state evaluation system can use a target learning state evaluation network to perform learning state evaluation processing on the first learning state data and the first number of second learning state data, and output the corresponding first number of learning state evaluation results. Wherein, for any of the first number of learning state evaluation results, the corresponding evaluation basis includes at least the first learning state data and one second learning state data. For example, the first learning state data and the first second learning state data correspond to the first learning state evaluation result; the first learning state data and the second second learning state data correspond to the second learning state evaluation result; the first learning state data and the third second learning state data correspond to the third learning state evaluation result. In addition, the target learning state evaluation network is a neural network trained based on sample learning state data (including sample first learning state data and sample second learning state data) and corresponding learning state labels (i.e., actual learning states, which can be formed by expert evaluation). For example, the evaluation is performed based on the sample learning state data, and then the obtained evaluation state and learning state label are subjected to error calculation. Afterwards, the parameters of the neural network can be adjusted in the direction of reducing the error so that the error converges, thereby obtaining the target learning state evaluation network. Exemplarily, the learning state evaluation result can be a discrete distribution (such as excellent learning state, better learning state, good learning state, average learning state, poor learning state, poor learning state, extremely poor learning state, etc.), or it can be a continuous distribution (such as 0-10 or 0-100, the higher the score, the better the state). Correspondingly, for discrete distribution, the output function of the target learning state neural network can be a classification function such as softmax. For continuous distribution, the output function of the target learning state neural network can be some linear regression functions, etc., which can be configured according to actual needs.

[0072] Step S130: Determine a target learning status assessment result of the target student based on the first number of learning status assessment results.

[0073] In an embodiment of the present invention, the artificial intelligence-based learning state assessment system can determine the target learning state assessment result of the target student based on the first number of learning state assessment results. The target learning state assessment result refers to the learning state assessment result of the first number of learning state assessment results in which the learning state does not meet the predetermined reference learning state, and the target learning state assessment result is used as the basis for strengthening the learning of the target student. For example, the learning state assessment results of poor learning state, relatively poor learning state, and extremely poor learning state can be used as the target learning state assessment result, or the learning state assessment results with a score less than 5 or 50 can be used as the target learning state assessment result, so that the learning content during the time of the corresponding second learning state data can be strengthened (i.e., the content learned during the period of poor learning state is re-learned to achieve the reinforcement of weak links. For example, when the eyes are squinting, the learning state may be poor, and the content of this node may not be fully learned. However, since squinting does not necessarily mean that the attention is not focused, the assessment is also combined with the student's overall action (behavior) to make the assessment result more accurate).

[0074] Based on the above method, since when the target learning state evaluation network is used to perform learning state evaluation processing, the evaluation basis corresponding to each learning state evaluation result obtained includes at least first learning state data and one second learning state data, so that the evaluation basis corresponding to each learning state evaluation result includes not only the expression information of the corresponding learning stage, but also the behavior information of the entire learning process. Therefore, the evaluation basis is rich and the evaluation basis is fully correlated, so that the obtained learning state evaluation result is more reliable; in addition, since the state data for the expression dimension will be subdivided, that is, the granularity is smaller and the accuracy is higher, and the state data for the behavior dimension is global, making the semantics richer, such a combination can also further improve the reliability of the evaluation, and thus improve the problem of relatively poor reliability of learning state evaluation in the existing technology (the improvement in reliability can ensure that the resources of subsequent reinforcement learning will not be wasted).

[0075] What needs to be explained about the above-mentioned step S120 is that the specific method of performing learning status evaluation processing on the first learning status data and the first number of second learning status data is not restricted and can be selected according to actual needs. For example, in a specific application example, in order to make the learning status evaluation processing more reliable and avoid similar input data (all the same students and all have the same first learning status data) causing the output data to be fitted as too similar, the above-mentioned step S120 can further include step S121, step S122, step S123, step S124 and step S125. The specific content of each step is described below.

[0076] Step S121 , performing feature mining on the first learning state data, outputting corresponding first state semantic features, and outputting corresponding first number of second state semantic features based on the first state semantic features and the first number of second learning state data.

[0077] In an embodiment of the present invention, the first learning state data can be feature mined to output a corresponding first state semantic feature, and based on the first state semantic feature and the first number of second learning state data, a corresponding first number of second state semantic features can be output. A second state semantic feature is mined based on the first state semantic feature and one of the second learning state data. In other words, a second state semantic feature carries the semantic information of the first state semantic feature and the semantic information of the second learning state data.

[0078] Step S122: determining the learning state interference characteristics.

[0079] In an embodiment of the present invention, a learning state interference feature may be determined. Exemplarily, the learning state interference feature may be randomly generated or generated based on historical learning state data.

[0080] Step S123, loading the learning state interference feature so that the first feature transfer unit in the target learning state evaluation network obtains the learning state interference feature, and uses the first feature transfer unit to process the learning state interference feature to form a transferred learning state interference feature.

[0081] In an embodiment of the present invention, after determining the learning state interference feature, the learning state interference feature can be loaded so that the first feature transfer unit in the target learning state evaluation network obtains the learning state interference feature, and the first feature transfer unit is used to process the learning state interference feature to form a transferred learning state interference feature.

[0082] Step S124 : loading the transfer learning state interference feature so that the second feature transfer unit in the target learning state evaluation network acquires the transfer learning state interference feature, and using the second feature transfer unit to determine a first number of feature extension links.

[0083] In an embodiment of the present invention, after obtaining the transfer learning state interference feature, the transfer learning state interference feature can be loaded so that the second feature transfer unit in the target learning state evaluation network obtains the transfer learning state interference feature and uses the second feature transfer unit to determine a first number of feature extension links. The second feature transfer unit is configured to input the first state semantic feature in the first feature extension process and input a second state semantic feature for each feature extension link in the second feature extension process. The input data of the second feature extension process includes the output data of the first feature extension process. That is, the second feature transfer unit includes two stages of processing, namely the first feature extension process and the second feature extension process, and the data processed by the second feature extension process includes the data (features) processed by the first feature extension process. In addition, the first number of feature extension links can refer to the first number of deep feature mining and evaluation output units in the second feature transfer unit, so that the first number of deep feature mining and evaluation output units can be used to perform deep mining on the transfer learning state interference feature, the first state semantic feature, and the first number of second state semantic features, and perform evaluation based on the mined deep features, thereby obtaining learning state evaluation results corresponding to each of the first number of feature extension links.

[0084] Step S125: Based on the transferred learning state interference feature, the first state semantic feature input by the first feature expansion process, and the first number of second state semantic features input by the second feature expansion process, output the learning state evaluation results corresponding to each of the first number of feature expansion links.

[0085] In an embodiment of the present application, based on the transferred learning state interference feature, the first state semantic feature input by the first feature expansion process, and the first number of second state semantic features input by the second feature expansion process, the learning state evaluation results corresponding to each of the first number of feature expansion links can be output. In other words, after the transferred learning state interference feature, the first state semantic feature input by the first feature expansion process, and the first number of second state semantic features input by the second feature expansion process are loaded into the second feature transfer unit, the second feature transfer unit can be used to process these loaded features, thereby realizing learning state evaluation and obtaining the corresponding learning state evaluation result.

[0086] In detail, in a specific application example, in order to further improve the representation ability of the obtained semantic features, the above step S121 may further include the following sub-steps:

[0087] First, the first learning state data may be segmented to output corresponding multiple local learning state data, and based on the image size corresponding to the first learning state data, pixel expansion operations may be performed on each of the local learning state data to form corresponding multiple expanded learning state data, wherein the image size corresponding to each of the expanded learning state data is equal to the image size corresponding to the first learning state data; illustratively, the image corresponding to the first learning state data may be segmented according to a predetermined image size (e.g., 108*192, etc.) to obtain multiple segmented sub-images, i.e., multiple local learning state data; then, pixel expansion (e.g., interpolation, etc.) may be performed on the segmented sub-images to obtain multiple expanded learning state data, such that the pixel distribution (e.g., 1080x1920, etc.) of the segmented sub-images after pixel expansion is the same as the pixel distribution of the image before segmentation;

[0088] Secondly, a convolution operation can be performed on the first learning state data to form a corresponding learning state convolution feature, and a convolution operation can be performed on each of the local learning state data respectively to form a local learning state convolution feature corresponding to each of the local learning state data, and the learning state convolution feature and the local learning state convolution feature corresponding to each of the local learning state data are fused to output the corresponding first state semantic feature; exemplarily, a convolution operation (and a pooling operation, etc.) can be performed on the first learning state data through a convolution unit (including one or more convolution kernels, etc.) in the target learning state evaluation network to form a corresponding learning state convolution feature, and A convolution operation is performed on each of the local learning state data to form a corresponding local learning state convolution feature. Then, the learning state convolution feature and each local learning state convolution feature can be concatenated (Concat, where the sizes of the features are consistent) to achieve fusion, thereby obtaining a first state semantic feature. Based on this, the first state semantic feature can include both global behavioral semantics and local behavioral semantics, achieving global and local attention, thereby improving the representation capability. It should be noted that, in other examples, the first learning state data can also be semantically segmented, such as segmented according to body parts, so that each body part can be further focused on.

[0089] Then, a convolution operation can be performed on each of the first number of second learning state data, and an initial learning state feature corresponding to each second learning state data can be output. Exemplarily, a convolution operation can also be performed on the second learning state data by a convolution unit in the target learning state evaluation network (which can be the same as or different from the aforementioned convolution unit) to output the corresponding initial learning state feature.

[0090] Finally, for each second learning state data, the initial learning state feature corresponding to the second learning state data and the first state semantic feature can be fused to output the second state semantic feature corresponding to the second learning state data; exemplarily, the initial learning state feature and the first state semantic feature can be connected to obtain the second state semantic feature, and the connected feature can be convolved to obtain the second state semantic feature; or, in other examples, the initial learning state feature and the first state semantic feature can be weightedly superimposed, etc., wherein the weight coefficient of the initial learning state feature can be larger to achieve differentiation between the second state semantic features, or focus on different expression information.

[0091] In detail, in a specific application example, in order to avoid excessive influence of the referenced learning state interference feature on subsequent feature processing, the above step S122 may further include:

[0092] First, historical learning state data can be determined, wherein the historical learning state data refers to the historical image data of the target student's learning behavior. It should be noted that when the target student does not have historical image data, the historical image data of other students can also be used;

[0093] Secondly, the historical learning state data can be loaded so that the feature mining unit in the target learning state evaluation network can obtain the historical learning state data; that is, the historical learning state data can be input into the feature mining unit;

[0094] Then, the feature mining unit can be used to perform a first feature mining operation and a second feature mining operation on the historical learning state data, respectively, to output a first historical state feature and a second historical state feature corresponding to the historical learning state data; illustratively, the first feature mining operation can be performed on the historical learning state data by performing a convolution operation and a mean pooling operation to obtain the first historical state feature; the first feature mining operation can be performed on the historical learning state data by performing a convolution operation and a maximum pooling operation (the pooling step size and window size are the same as the pooling step size and window size of the mean pooling operation) to obtain the second historical state feature;

[0095] Afterwards, based on the first historical state feature and the second historical state feature corresponding to the historical learning state data, the first historical state feature and the second historical state feature can be determined to be screened. For example, the parameters in the first historical state feature and the second historical state feature can be randomly sampled respectively, so that the first historical state feature and the second historical state feature can be obtained. In this way, the difference between the first historical state feature and the second historical state feature can be guaranteed, and the semantic information in the historical learning state data can also be guaranteed.

[0096] Finally, the learning state interference feature can be determined based on the first historical state feature and the second historical state feature. For example, any one of the first historical state feature and the second historical state feature can be used as the learning state interference feature, or the average of the first historical state feature and the second historical state feature can be used as the learning state interference feature. In this way, on the basis of adding interference features to avoid overfitting, the problem of poor feature expression caused by excessive interference can be avoided.

[0097] Specifically, in a specific application example, in order to avoid poor interference characteristics of the referenced learning state interference feature (such as excessive similarity between the historical learning state data and the first learning state data), the above step S123 may further include:

[0098] First, the first feature transfer unit may be used to determine an arbitrary interference feature. Exemplarily, the arbitrary interference feature may be randomly generated and include at least one non-zero parameter among multiple parameters. In addition, the size of the arbitrary interference feature may be greater than or equal to the size of the learning state interference feature.

[0099] Secondly, the first feature transfer unit can be used to perform a fusion operation of any local interference feature on the learning state interference feature in the second number of processing nodes, and output the corresponding transferred learning state interference feature, wherein the arbitrary local interference feature belongs to a partial feature of the arbitrary interference feature, and the arbitrary local interference features fused in the second number of processing nodes are different from each other, and the processing node carries a fusion weight coefficient for the fusion operation (the fusion weight coefficient can be used as a network parameter of the first feature transfer unit and is formed in the corresponding training process); exemplarily, combined with Figure 3In the first processing node, sampling can be performed from the arbitrary interference feature to obtain a first arbitrary local interference feature, and then, based on the corresponding fusion weight coefficient, the first arbitrary local interference feature and the learning state interference feature can be weighted summed to obtain a first fused interference feature; in the second processing node, sampling can be performed from the arbitrary interference feature (the sampling method is different in different processing nodes) to obtain a second arbitrary local interference feature, and then, based on the corresponding fusion weight coefficient, the second arbitrary local interference feature and the first fused interference feature can be weighted summed to obtain a second fused interference feature. In this way, the last fused interference feature corresponding to the last processing node can be obtained and used as the transferred learning state interference feature.

[0100] In detail, in a specific application example, in order to ensure that features with better representation capabilities can be mined so that a more reliable learning state evaluation can be performed based on the features, the above-mentioned step S125 can further include step S125a, step S125b, step S125c, step S125d, step S125e and step S125f. The specific content of each step is described below.

[0101] Step S125a: In the first feature expansion process, the first state semantic feature is inputted using the second feature transfer unit.

[0102] In an embodiment of the present invention, during the first feature expansion process, the second feature transfer unit may be used to input the first state semantic feature. That is, during the first feature expansion process, the first state semantic feature may be first input into the second feature transfer unit.

[0103] Step S125b: performing a deep mining operation on the transferred learning state interference feature and the first state semantic feature input by the first feature expansion process, and outputting a learning state deep feature corresponding to the first feature expansion process.

[0104] In an embodiment of the present invention, during the first feature expansion process, since the transferred learning state interference feature is previously input into the second feature transfer unit, the transferred learning state interference feature and the first state semantic feature input during the first feature expansion process can be subjected to a deep mining operation to output the learning state depth feature corresponding to the first feature expansion process. In this way, the learning state depth feature can represent the depth information, and the depth information has both the semantic information of the first state semantic feature and the semantic information of the transferred learning state interference feature, thereby enriching the depth information (such as fusing historical semantics as interference).

[0105] Step S125c: Associating the learning state deep features with the first number of feature extension links.

[0106] In an embodiment of the present invention, after outputting the learning state deep feature, the learning state deep feature can be associated with the first number of feature extension links, that is, each of the feature extension links can obtain the learning state deep feature.

[0107] Step S125d: In the second feature expansion process, a second state semantic feature is input for each feature expansion link.

[0108] In this embodiment of the present invention, after the output and association of the learned state deep features are completed, a second feature expansion process can be entered. In this second feature expansion process, a second state semantic feature can be input for each feature expansion link. In other words, since there are a first number of feature expansion links and a first number of second state semantic features, the second state semantic features can be input in a one-to-one correspondence with the feature expansion links.

[0109] Step S125e: In each feature expansion link, a feature expansion operation is performed on the learning state deep feature and the second state semantic feature input in the second feature expansion process, and a first number of learning state expanded features corresponding to the second feature expansion process are output.

[0110] In an embodiment of the present invention, still in the second feature expansion process, after realizing the input of the second state semantic feature, the learning state deep feature and the second state semantic feature input in the second feature expansion process can be subjected to feature expansion operation in each feature expansion link, and a first number of learning state expansion features corresponding to the second feature expansion process can be output; exemplarily, in the first feature expansion link, the learning state deep feature and the first second state semantic feature can be subjected to feature expansion operation, and the first learning state expansion feature can be output; in the second feature expansion link, the learning state deep feature and the second second state semantic feature can be subjected to feature expansion operation, and the second learning state expansion feature can be output; in the third feature expansion link, the learning state deep feature and the third second state semantic feature can be subjected to feature expansion operation, and the third learning state expansion feature can be output; based on this, the learning state expansion feature corresponding to each feature expansion link can be obtained, and for the first number of feature expansion links, the first number of learning state expansion features can be obtained in the second feature expansion process.

[0111] Step S125f: performing a learning state evaluation operation based on the first number of learning state extension features, and outputting learning state evaluation results corresponding to each of the first number of feature extension links.

[0112] In an embodiment of the present invention, after obtaining a first number of learning state extended features, a learning state evaluation operation may be performed based on the first number of learning state extended features, and learning state evaluation results corresponding to each of the first number of feature extension links may be output. For example, for the first learning state extended feature, a learning state evaluation operation may be performed on the learning state extended feature, and the corresponding first learning state evaluation result may be output; for the second learning state extended feature, a learning state evaluation operation may be performed on the learning state extended feature, and the corresponding second learning state evaluation result may be output. In addition, it should be noted that since there may be a correlation between the expressions at different stages of the learning process, for each learning state extended feature, the learning state extended feature can be first subjected to an associated focus mining operation based on each other learning state extended feature to obtain multiple associated focus mining results corresponding to the learning state extended feature. Then, the multiple associated focus mining results can be weighted summed (wherein, the closer the learning stage corresponding to other learning state extended features is to the learning stage corresponding to the learning state extended feature, the larger the corresponding weighting coefficient is) to obtain the weighted summation result corresponding to the learning state extended feature. Finally, the weighted summation result is pooled and fully connected. Finally, the processed result is classified and output based on the corresponding output function to obtain the corresponding learning state evaluation result.

[0113] In detail, in a specific application example, in order to ensure the accuracy of deep mining operations so that the obtained learning state deep features have better semantic representation capabilities, the above-mentioned step S125b can further include step b1, step b2 and step b3, and the contents of each step are described as follows.

[0114] Step b1: loading the transfer learning state interference feature so that the second feature transfer unit acquires the transfer learning state interference feature.

[0115] In an embodiment of the present invention, the transfer learning state interference feature can be loaded so that the second feature transfer unit obtains the transfer learning state interference feature. The second feature transfer unit includes a feature mining subunit, and the feature mining subunit includes a feature compression module, a feature diffusion module, and a feature expansion module. The input data of the feature diffusion module includes the output data of the feature compression module, and the input data of the feature expansion module includes the output data of the feature diffusion module. That is, the feature compression module, the feature diffusion module, and the feature expansion module are cascaded, and the output data of the previous module serves as the input data of the next module. The feature compression module includes a first focusing submodule and a second focusing submodule.

[0116] Step b2: using the first focusing submodule, performing a focused mining operation on the transfer learning state interference feature, and outputting a corresponding first learning state focused feature.

[0117] In an embodiment of the present invention, after the transfer learning state interference feature is input into the first focusing submodule, the first focusing submodule can be used to perform a focused mining operation on the transfer learning state interference feature, outputting a corresponding first learning state focused feature. The specific processing method of the focused mining operation can be referred to the explanation of the associated focused mining operation below, that is, performing an associated focused mining operation on the transfer learning state interference feature and the transfer learning state interference feature, and outputting a corresponding first learning state focused feature.

[0118] Step b3: using the second focusing submodule, performing an associated focusing mining operation on the first learning state focusing feature and the input first state semantic feature, and outputting a learning state deep feature corresponding to the first feature expansion process.

[0119] In an embodiment of the present invention, after obtaining the first learning state focus feature, the second focus submodule can be used to perform an association focus mining operation on the first learning state focus feature and the input first state semantic feature, and output the learning state deep feature corresponding to the first feature expansion process. Wherein, the number of dimensions of the transferred learning state interference feature is greater than the number of dimensions of the learning state deep feature, and the feature diffusion module is used to perform a diffusion operation on the learning state deep feature to form a first number of learning state deep features, and to perform a loading operation on the first number of learning state deep features to load them into the feature expansion module used to output the first number of learning state expansion features in the second feature expansion process, that is, the feature expansion module is used to perform a feature expansion operation on the first number of learning state deep features, thereby outputting the first number of learning state expansion features. Exemplarily, the first number of learning state deep features may be completely identical or may not be completely identical.

[0120] In detail, in a specific application example, the specific manner of performing the association focus mining operation in the above step b3 is not limited. For example, in order to fully integrate the first learning state focus feature with the first state semantic feature through the association focus mining operation to ensure that the obtained learning state deep feature has better semantic representation ability, the above step b3 may further include the following sub-steps:

[0121] First, the first learning state focus feature and the first spatial mapping parameter in the second focusing submodule can be multiplied to output the corresponding first learning state mapping feature, wherein the first spatial mapping parameter serves as a network parameter of the second focusing submodule and can be formed in a corresponding training process;

[0122] Secondly, the first learning state focus feature and the second spatial mapping parameter in the second focusing submodule can be multiplied to output a corresponding second learning state mapping feature, wherein the second spatial mapping parameter serves as a network parameter of the second focusing submodule and can be formed in a corresponding training process;

[0123] Then, the first state semantic feature can be multiplied by the third spatial mapping parameter in the second focusing submodule to output a corresponding third learned state mapping feature, wherein the third spatial mapping parameter serves as a network parameter of the second focusing submodule and can be formed in a corresponding training process;

[0124] Afterwards, a multiplication operation may be performed on the first learning state mapping feature and the second learning state mapping feature to output a corresponding focus association parameter, and an update operation may be performed on the focus association parameter based on the number of dimensions of the second learning state mapping feature to output a corresponding updated focus association parameter. For example, the second learning state mapping feature may be divided by the number of dimensions, or by a positive correlation coefficient of the number of dimensions, such as the square of the positive correlation parameter being equal to the number of dimensions.

[0125] Finally, the updated focus association parameter and the third learning state mapping feature may be multiplied to output a corresponding learning state depth feature.

[0126] In detail, in a specific application example, in order to ensure the reliability of the feature expansion operation and to improve the semantic representation ability of the obtained learning state extended feature, the above-mentioned step S125e may further include the following sub-steps (i.e., the specific processing process of the above-mentioned feature expansion module, the first number of feature expansion links includes any feature expansion link, and the arbitrary feature expansion link refers to any feature expansion link):

[0127] First, the learning state deep feature can be loaded so that the feature extension module obtains the learning state deep feature, wherein the feature extension module includes the third focusing submodule and the fourth focusing submodule corresponding to the arbitrary feature extension link, that is, the learning state deep feature can be input into the feature extension module;

[0128] Secondly, the third focusing submodule can be used to perform a focused mining operation (as described above) on the learning state deep features to output corresponding second learning state focused features;

[0129] Then, the fourth focusing submodule may be used to perform an association focus mining operation (as described above) on the second learning state focus feature and the second state semantic feature input by the arbitrary feature extension link, and output a corresponding learning state association focus feature;

[0130] Finally, based on the learning state associated focused feature, the learning state extended feature corresponding to the arbitrary feature extension link can be determined, wherein the number of dimensions of the transmitted learning state interference feature is equal to the number of dimensions of the learning state extended feature; exemplarily, the learning state associated focused feature can be used as the learning state extended feature, or the learning state associated focused feature can be further processed to obtain the corresponding learning state extended feature. For example, the learning state associated focused feature and the second state semantic feature can be superimposed to obtain the corresponding learning state associated focused feature, or the superimposed result can be processed by convolution, pooling, full connection, etc.

[0131] In detail, in a specific application example, the second feature transfer unit includes a second number of feature mining sub-units connected in cascade, each of which is used to input the first state semantic feature in the first feature expansion process and input a second state semantic feature for each feature expansion link in the second feature expansion process. The second number of feature mining sub-units includes any feature mining sub-unit, and the any feature mining sub-unit refers to any feature mining sub-unit. Based on this, the above-mentioned step S125 may also include:

[0132] First, the features to be processed can be loaded so that the arbitrary feature mining subunit obtains the features to be processed, and the arbitrary feature mining subunit is used to perform a deep mining operation on the features to be processed and the first state semantic features input in the first feature expansion process (refer to the relevant description in the previous text), and output a first number of learning state deep features corresponding to the arbitrary feature mining subunit, and perform an association operation on the first number of learning state deep features corresponding to the arbitrary feature mining subunit, so that the first number of feature expansion links corresponding to the arbitrary feature mining subunit are associated one by one, wherein, combined with Figure 4 , when the arbitrary feature mining subunit belongs to the first feature mining subunit, the feature to be processed is the transferred learning state interference feature, and the first number of learning state deep features are the same; when the arbitrary feature mining subunit does not belong to the first feature mining subunit, the feature to be processed includes the first number of learning state extended features corresponding to the previous feature mining subunit of the arbitrary feature mining subunit, and a learning state deep feature is formed based on a learning state extended feature corresponding to the previous feature mining subunit of the arbitrary feature mining subunit and the first state semantic feature, and the input data of the arbitrary feature mining subunit includes the output data of the previous feature mining subunit of the arbitrary feature mining subunit;

[0133] Then, in the second feature expansion process, a second-state semantic feature can be input into each feature expansion link of the arbitrary feature mining subunit, and a feature expansion operation (refer to the relevant description above) is performed on the learning state deep feature corresponding to the arbitrary feature mining subunit and the input second-state semantic feature in each feature expansion link, and a first number of learning state expansion features corresponding to the arbitrary feature mining subunit is output;

[0134] Finally, when the arbitrary feature mining sub-unit belongs to the last feature mining sub-unit, the learning state evaluation results corresponding to the first number of feature extension links of the arbitrary feature mining sub-unit are determined based on the first number of learning state extension features corresponding to the arbitrary feature mining sub-unit; that is, the learning state evaluation operation can be performed on the first number of learning state extension features corresponding to the last feature mining sub-unit to obtain the learning state evaluation results corresponding to the first number of feature extension links.

[0135] Finally, it should be noted that the various features mentioned above can be expressed in the form of vectors, etc.

[0136] To sum up, the artificial intelligence-based learning status assessment method and system provided by the present invention can obtain the first learning status data and the first number of second learning status data of the target student during the learning process; secondly, using the target learning status assessment network, the first learning status data and the first number of second learning status data are subjected to learning status assessment processing, and the corresponding first number of learning status assessment results are output; then, based on the first number of learning status assessment results, the target learning status assessment results of the target student are determined. Based on the above method, since when the target learning state evaluation network is used to perform learning state evaluation processing, the evaluation basis corresponding to each learning state evaluation result obtained includes at least first learning state data and one second learning state data, so that the evaluation basis corresponding to each learning state evaluation result includes not only the expression information of the corresponding learning stage, but also the behavior information of the entire learning process. Therefore, the evaluation basis is rich and the evaluation basis is fully correlated, so that the obtained learning state evaluation result is more reliable; in addition, since the state data for the expression dimension will be subdivided, that is, the granularity is smaller and the accuracy is higher, and the state data for the behavior dimension is global, making the semantics richer, such a combination can also further improve the reliability of the evaluation, and thus improve the problem of relatively poor reliability of learning state evaluation in the existing technology (the improvement in reliability can ensure that the resources of subsequent reinforcement learning will not be wasted).

[0137] In the several embodiments provided in the embodiments of the present invention, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device and method embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the devices, methods, and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, which contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the boxes can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, as well as the combination of boxes in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified functions or actions, or can be implemented using a combination of dedicated hardware and computer instructions.

[0138] In addition, the functional modules in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0139] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk. It should be noted that, in this article, the terms "include", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements that are not explicitly listed, or also includes elements inherent to such a process, method, article or device. Without further constraints, an element defined by the phrase "comprises a..." does not preclude the existence of additional identical elements in the process, method, article or apparatus that includes the element.

[0140] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A learning status assessment method based on artificial intelligence, characterized in that: include: Step S110: Acquire first learning state data and a first number of second learning state data of a target student during a learning process, wherein the first learning state data is image data of the target student's learning behavior, and the second learning state data is image data of the target student's learning expression, and formation times corresponding to the first number of learning expressions reflected by the first number of second learning state data are different from each other, and the duration corresponding to the learning behavior reflected by the first learning state data and the duration corresponding to the first number of learning expressions reflected by the first number of second learning state data have the same starting point and end point; Step S120: Using a target learning state evaluation network, perform learning state evaluation processing on the first learning state data and the first number of second learning state data, and output a corresponding first number of learning state evaluation results, wherein, for any one of the first number of learning state evaluation results, the corresponding evaluation basis includes at least the first learning state data and one second learning state data, and the target learning state evaluation network is a neural network formed by training based on sample learning state data and corresponding learning state labels; Step S130: Determine a target learning status assessment result for the target student based on the first number of learning status assessment results, wherein the target learning status assessment result refers to a learning status assessment result in the first number of learning status assessment results whose learning status does not meet a predetermined reference learning status, and the target learning status assessment result serves as a basis for learning reinforcement for the target student; Wherein, step S120 includes: Step S121: Perform feature mining on the first learning state data to output a corresponding first state semantic feature, and output a corresponding first number of second state semantic features based on the first state semantic feature and the first number of second learning state data, wherein one second state semantic feature is mined based on the first state semantic feature and one second learning state data; Step S122, determining the learning state interference characteristics; Step S123: Loading the learning state interference feature so that the first feature transfer unit in the target learning state evaluation network obtains the learning state interference feature, and using the first feature transfer unit to process the learning state interference feature to form a transferred learning state interference feature, wherein the processing method is to fuse the learning state interference feature with the randomly generated arbitrary interference feature, and the fusion operation is a weighted summation; Step S124: loading the transfer learning state interference feature so that the second feature transfer unit in the target learning state evaluation network obtains the transfer learning state interference feature, and using the second feature transfer unit, determining a first number of feature extension links, wherein the second feature transfer unit is used to input the first state semantic feature in the first feature extension process, and input a second state semantic feature for each feature extension link in the second feature extension process, and the input data of the second feature extension process includes the output data of the first feature extension process; Step S125: outputting learning state evaluation results corresponding to each of the first number of feature extension links based on the transferred learning state interference feature, the first state semantic feature input by the first feature extension process, and the first number of second state semantic features input by the second feature extension process; Wherein, step S122 includes: Determining historical learning status data, wherein the historical learning status data refers to historical image data about the learning behavior of the target student; Loading the historical learning state data so that a feature mining unit in a target learning state evaluation network acquires the historical learning state data; Using the feature mining unit, performing a first feature mining operation and a second feature mining operation on the historical learning state data, respectively, and outputting a first historical state feature and a second historical state feature corresponding to the historical learning state data, wherein the first feature mining operation includes performing a convolution operation and a mean pooling operation on the historical learning state data to obtain the first historical state feature; and the second feature mining operation includes performing a convolution operation and a maximum pooling operation on the historical learning state data to obtain the second historical state feature; Based on the first historical state feature and the second historical state feature corresponding to the historical learning state data, determining to select the first historical state feature and the second historical state feature, wherein the screening process is to randomly sample parameters in the first historical state feature and the second historical state feature to obtain the first historical state feature and the second historical state feature; Determining a learning state interference feature based on the first filtered historical state feature and the second filtered historical state feature, wherein the determining process includes using either the first filtered historical state feature or the second filtered historical state feature as the learning state interference feature, or using an average of the first filtered historical state feature and the second filtered historical state feature as the learning state interference feature; Wherein, step S125 includes: Step S125a: In the first feature expansion process, the first state semantic feature is inputted using the second feature transfer unit; Step S125b, performing a deep mining operation on the transferred learning state interference feature and the first state semantic feature input by the first feature expansion process, and outputting a learning state deep feature corresponding to the first feature expansion process; Step S125c, associating the learning state deep features with the first number of feature extension links; Step S125d: In the second feature expansion process, a second state semantic feature is input for each feature expansion link; Step S125e: In each feature extension link, perform a feature extension operation on the learning state deep feature and the second state semantic feature input in the second feature extension process, and output a first number of learning state extended features corresponding to the second feature extension process; Step S125f, performing a learning state evaluation operation based on the first number of learning state extension features, and outputting learning state evaluation results corresponding to each of the first number of feature extension links; Wherein, step S125b includes: Step b1: loading the transfer learning state interference feature so that the second feature transfer unit acquires the transfer learning state interference feature, wherein the second feature transfer unit includes a feature mining subunit, the feature mining subunit includes a feature compression module, a feature diffusion module and a feature expansion module, the input data of the feature diffusion module includes the output data of the feature compression module, the input data of the feature expansion module includes the output data of the feature diffusion module, and the feature compression module includes a first focusing submodule and a second focusing submodule; Step b2: using the first focusing submodule to perform a focused mining operation on the transfer learning state interference feature, and outputting a corresponding first learning state focused feature; Step b3: using the second focusing submodule, performing an association focusing mining operation on the first learning state focus feature and the input first state semantic feature, and outputting a learning state deep feature corresponding to the first feature expansion process, wherein the number of dimensions of the transferred learning state interference feature is greater than the number of dimensions of the learning state deep feature, and the feature diffusion module is used to perform a diffusion operation on the learning state deep feature to form a first number of learning state deep features, and perform a loading operation on the first number of learning state deep features to load them into a feature expansion module used to output a first number of learning state expansion features in the second feature expansion process; Wherein, step S125e includes: Step e1: Loading the learning state deep feature so that the feature extension module acquires the learning state deep feature, wherein the first number of feature extension links includes any feature extension link, the arbitrary feature extension link refers to any feature extension link, and the feature extension module includes a third focusing sub-module and a fourth focusing sub-module corresponding to the arbitrary feature extension link; Step e2: using the third focusing submodule to perform a focused mining operation on the learning state deep features, and outputting corresponding second learning state focused features; Step e3: Using the fourth focusing submodule, perform an association focusing mining operation on the second learning state focused feature and the second state semantic feature input by the arbitrary feature extension link, and output a corresponding learning state associated focused feature; Step e4: determining the learning state extension feature corresponding to the arbitrary feature extension link based on the learning state associated focus feature, wherein the number of dimensions of the transferred learning state interference feature is equal to the number of dimensions of the learning state extension feature.

2. The method for evaluating learning status based on artificial intelligence according to claim 1, wherein: Step S121 includes: Performing a segmentation operation on the first learning state data to output a plurality of corresponding local learning state data, and performing a pixel expansion operation on each of the local learning state data based on an image size corresponding to the first learning state data to form a plurality of corresponding expanded learning state data, wherein the image size corresponding to each of the expanded learning state data is equal to the image size corresponding to the first learning state data, and the pixel expansion operation includes interpolation; performing a convolution operation on the first learning state data to form a corresponding learning state convolution feature, and performing a convolution operation on each of the local learning state data to form a local learning state convolution feature corresponding to each of the local learning state data, and fusing the learning state convolution feature with the local learning state convolution feature corresponding to each of the local learning state data to output a corresponding first state semantic feature, wherein the fusing process includes concatenating the learning state convolution feature with each of the local learning state convolution features; performing a convolution operation on each of the first number of second learning state data, and outputting an initial learning state feature corresponding to each second learning state data; For each second learning state data, the initial learning state feature corresponding to the second learning state data and the first state semantic feature are fused, and the second state semantic feature corresponding to the second learning state data is output, wherein the fusion process includes weighted superposition of the initial learning state feature and the first state semantic feature, or connecting the initial learning state feature and the first state semantic feature to obtain the second state semantic feature, or performing a convolution operation on the feature after the initial learning state feature and the first state semantic feature are connected to obtain the second state semantic feature.

3. A learning status assessment system based on artificial intelligence, characterized in that: It includes a processor and a memory, the memory is used to store a computer program, and the processor is used to execute the computer program to implement the artificial intelligence-based learning status assessment method described in any one of claims 1-2.

Images