Deep learning-based teaching video transmission bit rate optimization method

Abstract

The application relates to the field of video transmission technology, in particular to a teaching video transmission code rate optimization method based on deep learning, which comprises the following steps: dividing a teaching video into multiple video segments based on the image complexity and semantic complexity of each timestamp; assigning an initial code rate to each video segment to obtain multiple transmission schemes, and calculating the fitness function of each transmission scheme; iteratively updating the multiple transmission schemes by using a particle swarm optimization algorithm to obtain the transmission code rate optimization result of the teaching video. According to the technical scheme, the transmission code rate of each timestamp in the teaching video can be determined according to the video content, and the transmission efficiency of the teaching video is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of video transmission, in particular to a teaching video transmission code rate optimization method based on deep learning. BACKGROUND

[0002] With the rapid development of Internet technology, online education has also emerged, which often needs to transmit teaching videos from one end to the other end. In the process of transmitting teaching videos, the transmission code rate is an important indicator. Too low code rate will seriously reduce the quality of video playback, and too high code rate will increase the load of Internet hardware devices, and even cannot be achieved due to poor network environment. Therefore, dynamically adjusting the code rate in the process of transmitting teaching videos is a development trend of video transmission.

[0003] At present, the patent application file with publication number CN114007113A discloses a video code rate self-adaptive adjustment method and device, wherein the method comprises: collecting network environment parameters, the network environment parameters including transmission code rates at different transmission times and lengths of video cache time periods in the buffer; scoring the cache quality, transmission quality and code rate switching quality at time t based on the collected network environment parameters; wherein the cache quality is scored based on the length of the video cache time period in the buffer under different transmission code rates, the transmission quality is scored based on the video blocks transmitted under different transmission code rates within the first preset time period, and the code rate switching quality is scored based on the number and level of video code rate switching within the second preset time period; and adjusting the video code rate at time t+1 based on the scores of the cache quality, transmission quality and code rate switching quality.

[0004] The above method adjusts the video code rate at the next time based on the cache quality, transmission quality and code rate switching quality at each time in the video transmission process. However, the above transmission process ignores the influence of video content on the transmission code rate, cannot realize reasonable allocation of transmission resources according to video content, and leads to low efficiency of video transmission. SUMMARY

[0005] In order to solve the technical problem of low efficiency of video transmission, the present application provides a teaching video transmission code rate optimization method based on deep learning, which can determine the transmission code rate of each timestamp in the teaching video according to the video content, and improve the transmission efficiency of the teaching video.

[0006] In a first aspect, the application provides a deep learning-based teaching video transmission code rate optimization method, which comprises: dividing a teaching video into a plurality of video segments based on the image complexity and semantic complexity of each timestamp; assigning an initial code rate to each video segment to obtain a plurality of transmission schemes, and calculating the fitness function of each transmission scheme, which is positively correlated with the transmission quality of each video segment and negatively correlated with the volatility of the transmission quality sequence; iteratively updating the plurality of transmission schemes using a particle swarm optimization algorithm to obtain the transmission code rate optimization result of the teaching video; wherein the transmission quality is obtained by a prediction model, and the input of the prediction model is the network state and the image complexity, semantic complexity and code rate of the video segment, and the output is the transmission quality of the video segment.

[0007] By quantifying the image complexity and semantic complexity of the video content, allocating the transmission code rate according to the video content, and combining the particle swarm algorithm for global optimization, the transmission code rate optimization result is obtained; the transmission code rate of each timestamp in the teaching video can be determined according to the video content, the transmission efficiency of the teaching video is improved, and the transmission quality of the key teaching content with high complexity is guaranteed.

[0008] Preferably, the method for obtaining the image complexity of any timestamp comprises: obtaining the entropy value of the gray level co-occurrence matrix of the video image corresponding to the timestamp; interpolating the video images of the previous adjacent timestamp and the next adjacent timestamp of the timestamp to obtain a regression image, and taking the average pixel value of the difference image between the regression image and the video image as the regression complexity; and taking the sum of the normalized entropy value and the regression complexity as the image complexity of the timestamp.

[0009] The image complexity combines the entropy value of the gray level co-occurrence matrix and the interpolation regression residual to reflect the richness of image detail information and the difficulty of time series prediction, respectively, quantifies the importance of the image from the texture information entropy and the time series stability, realizes fine-grained image quality perception, and helps to accurately allocate transmission resources to the visual key area and improve the fidelity of video details.

[0010] Preferably, the method for obtaining the semantic complexity of any timestamp comprises: converting the audio information of the preset window of the timestamp into a frequency space, and taking the normalized information entropy of each frequency energy value in the frequency space as the semantic complexity.

[0011] By converting the voice signal into a frequency space and calculating the normalized information entropy, the information density of the voice content can be effectively evaluated. High information entropy corresponds to high content concentration and large semantic complexity, such as key explanation segments of formula explanation and case analysis, and the semantic complexity is constructed from the perspective of frequency energy distribution, so that the code rate allocation is more sensitive and accurate to the teaching voice content.

[0012] Preferably, the timestamp semantic complexity satisfy:

[0013] ; The number of frequency types, For the first The energy value of a certain frequency, It is the sum of the energy values ​​of all frequencies. This represents the maximum value of the information entropy.

[0014] Preferably, dividing the teaching video into multiple video segments based on the image complexity and semantic complexity of each timestamp includes: constructing a complexity sequence based on the maximum value of the image complexity and semantic complexity of each timestamp; performing ordered sample clustering on the complexity sequence to obtain multiple time periods, wherein the multiple time periods correspond to the multiple video segments.

[0015] By constructing a complexity sequence using the maximum value of image and semantic complexity, the "most difficult to preserve" feature at each time point is preserved, ensuring that video segmentation fully identifies abrupt change points. Combined with ordered sample clustering, the segmentation can be guaranteed to have temporal continuity and intra-segment complexity consistency, achieving adaptive video segmentation and improving the rationality and optimization efficiency of segmentation.

[0016] Preferably, obtaining multiple transmission schemes includes: acquiring adjustment coefficients. The adjustment coefficient is a random number between 0 and 1; the initial bitrate of each video segment is calculated based on the adjustment coefficient to obtain the adjustment coefficient. Corresponding transmission scheme, video segment initial bitrate for: ; and These are the minimum and maximum bitrate values, respectively. For adjustment coefficients, For video segments The maximum values ​​of image complexity and semantic complexity are determined; the adjustment coefficient is then adjusted. The value of is used to obtain multiple transmission schemes.

[0017] Preferably, the training process of the prediction model includes: collecting the bitrate and network status during the transmission of video segment samples, and obtaining the VMAF of the received video segment samples after transmission as a quality label; inputting the bitrate, network status, and image complexity and semantic complexity of the video segment samples into the prediction model to obtain the prediction result; iteratively updating the prediction model based on the mean squared error loss between the prediction result and the quality label until the number of iterations is greater than a preset number, or the mean squared error loss is less than a preset loss, to complete the training.

[0018] Preferably, the fitness function Satisfying the relation:

[0019] ; The number of video segments, For video segments The maximum value of image complexity and semantic complexity in the medium. It is the sum of the maximum values ​​of image complexity and semantic complexity across all video segments; For video segments Duration The total duration of the instructional videos. , and Video segments Video segment and video clips Transmission quality; For the change threshold, For step function, when hour, =1, otherwise, =0; For penalty weights.

[0020] It can be regarded as a weighted transmission quality of each video segment, with more attention paid to video segments with high complexity and long duration. The higher the weighted transmission quality, the better the transmission quality of each video segment after transmission according to the transmission plan. This is used to characterize the percentage of times that changes in transmission quality affect the viewing experience in teaching videos. The higher the percentage, the greater the volatility of the transmission quality sequence, the worse the user's viewing experience, and the smaller the value of the fitness function. By combining the transmission quality of each video segment and the temporal volatility of transmission quality, the advantages and disadvantages of the transmission scheme can be accurately quantified.

[0021] Preferably, the network status includes at least one of remaining bandwidth, network volatility, packet loss rate, and transmission delay.

[0022] Preferably, the particle swarm optimization algorithm is used to iteratively update the multiple transmission schemes to obtain the transmission bitrate optimization result of the teaching video, including: in response to the fact that the increase in the fitness function of the optimal position of the population in the continuous multiple iterations is less than the preset increment, the optimal position of the population in the last iteration is taken as the transmission bitrate optimization result.

[0023] The technical solution of this application has the following beneficial technical effects:

[0024] By combining the image and semantic complexity of different timestamps in the teaching videos, the videos are adaptively segmented, and the bitrate is optimized on a per-segment basis. Multiple transmission schemes are obtained by assigning initial bitrates to each video segment, and the segments are propagated based on these initial bitrates. A predictive model is used to accurately estimate the transmission quality of each video segment. A fitness function is constructed by comprehensively considering transmission quality and quality fluctuations to evaluate the merits of the transmission schemes. Finally, a particle swarm optimization algorithm is used to iteratively search for the optimal transmission scheme. This system can adaptively segment video content based on its content, ensuring the transmission quality of key teaching content (such as formula derivations, PPT presentations, and blackboard writing) while balancing the overall playback experience and avoiding abrupt changes in image or audio quality. This significantly improves the transmission efficiency and viewing experience of teaching videos. Attached Figure Description

[0025] Figure 1 This is a flowchart of a deep learning-based method for optimizing the transmission bitrate of teaching videos according to an embodiment of this application.

[0026] Figure 2 This is a structural diagram of the prediction model according to an embodiment of this application. Detailed Implementation

[0027] According to the first aspect of this application, this application provides a method for optimizing the transmission bitrate of teaching videos based on deep learning. Figure 1 This is a flowchart of a deep learning-based method for optimizing the transmission bitrate of teaching videos according to an embodiment of this application. Figure 1 As shown, the deep learning-based teaching video transmission bitrate optimization method includes steps S101 to S103, which are described in detail below.

[0028] S101 divides the teaching video into multiple video segments based on the image complexity and semantic complexity of each timestamp.

[0029] In one embodiment, the content of each timestamp in a teaching video differs. For the lecturing part of the teaching video, the semantic features of its audio information are particularly important, while for the demonstration part of the teaching video, the image features of its video images are particularly important.

[0030] The semantic complexity of voice information and the image complexity of video images will differ for different timestamps. The greater the image complexity, the more detailed features there are in the video image (such as text display, animation simulation, etc.). In order to ensure that the detailed features of the video image can be preserved after video transmission, a larger transmission bitrate needs to be allocated to the video segment in which the video image is located. Similarly, the greater the semantic complexity, the richer the semantics in the voice information (such as concept explanation, formula derivation, case analysis, etc.). In order to ensure that the semantic features of the voice information can be preserved after video transmission, a larger transmission bitrate also needs to be allocated to the video segment in which the voice information is located.

[0031] Therefore, to ensure that the receiving end can retain the complete video content after transmission, the teaching video is adaptively segmented according to the image complexity and semantic complexity of each timestamp, resulting in multiple video segments. The image complexity and semantic complexity of each timestamp within a video segment are basically the same. In subsequent transmission, the transmission bitrate is optimized on a per-video-segment basis, thereby improving transmission efficiency while preserving the complete video content.

[0032] Specifically, the method for obtaining the image complexity of any timestamp is as follows: obtain the entropy value of the gray-level co-occurrence matrix of the video image corresponding to the timestamp; interpolate the video images of the previous and next adjacent timestamps of the timestamp to obtain a regression image, and use the average pixel value of the difference map between the regression image and the video image as the regression complexity; and use the sum of the normalized entropy value and the regression complexity as the image complexity of the timestamp.

[0033] Understandably, the entropy value of the gray-level co-occurrence matrix indicates the complexity of the pixel value distribution in a video image. A larger entropy value indicates more information contained in the video image, making it more complex. The regression image is obtained by interpolating video images from the previous and next adjacent timestamps when the video image corresponding to the timestamp is missing. If the average pixel value of the regression image and the video image difference map is 0, it means that the video images from the previous and next adjacent timestamps can accurately capture the video image of the timestamp, and the regression complexity of that timestamp is also 0. The normalization method uses Min-Max normalization.

[0034] Specifically, the method for obtaining the semantic complexity of any timestamp is as follows: convert the audio information of the preset window of the timestamp into a frequency space, and use the normalized information entropy of each frequency energy value in the frequency space as the semantic complexity.

[0035] The preset window includes multiple timestamps on both sides adjacent to the timestamp; timestamp semantic complexity satisfy: ; The number of frequency types, For the first The energy value of a certain frequency, It is the sum of the energy values ​​of all frequencies. This represents the maximum value of the information entropy.

[0036] In one embodiment, after obtaining the image complexity and semantic complexity of each timestamp, dividing the teaching video into multiple video segments based on the image complexity and semantic complexity of each timestamp includes: constructing a complexity sequence based on the maximum value of the image complexity and semantic complexity of each timestamp; performing ordered sample clustering on the complexity sequence to obtain multiple time periods, wherein the multiple time periods correspond to the multiple video segments.

[0037] The ordered sample clustering is a clustering algorithm for time-series data. It divides the time-series data by finding multiple split points, resulting in smaller differences within the same data segment and larger differences between adjacent data segments. Ordered sample clustering is a well-known technique to those skilled in the art and will not be elaborated further here. The number of video segments can be obtained using the elbow method.

[0038] In this way, the teaching videos are adaptively divided based on the image complexity and semantic complexity of each timestamp, resulting in multiple video segments. The complexity within each video segment is basically the same, while the complexity between adjacent video segments differs significantly, providing a data foundation for subsequent optimization of the transmission bitrate.

[0039] S102, assign an initial bitrate to each video segment to obtain multiple transmission schemes, and calculate the fitness function of each transmission scheme. The fitness function is positively correlated with the transmission quality of each video segment and negatively correlated with the volatility of the transmission quality sequence.

[0040] In one embodiment, after obtaining multiple video segments, an initial bitrate is assigned to each video segment to obtain a transmission scheme. Thus, multiple transmission schemes can be obtained. The initial bitrate is any one of a bitrate range; in this embodiment, the bitrate range is... .

[0041] In another embodiment, within the bitrate range, a higher transmission bitrate results in higher quality teaching videos after transmission. To ensure good quality for video segments with greater complexity, important information from the teaching videos is retained after transmission. Initial bitrates are assigned to each video segment based on its complexity, resulting in multiple transmission schemes. This method generates multiple transmission schemes that not only guarantee higher bitrates for video segments with greater complexity but also improve the optimization efficiency of subsequent particle swarm optimization algorithms to some extent.

[0042] Specifically, obtaining multiple transmission schemes includes: acquiring adjustment coefficients. The adjustment coefficient is a random number between 0 and 1; the initial bitrate of each video segment is calculated based on the adjustment coefficient to obtain the adjustment coefficient. Corresponding transmission scheme, video segment initial bitrate for:

[0043] ; and These are the minimum and maximum bitrate values, respectively. For adjustment coefficients, For video segments The maximum values ​​of image complexity and semantic complexity are determined; the adjustment coefficient is then adjusted. The value of is used to obtain multiple transmission schemes.

[0044] The bit rate range is: Then the minimum bitrate is The maximum bitrate is .

[0045] After obtaining multiple transmission schemes, the fitness function of each transmission scheme is calculated. The fitness function is used to evaluate the merits of the transmission schemes to ensure that the particle swarm optimization algorithm can search for the transmission scheme with the largest fitness function value.

[0046] On the one hand, the purpose of optimizing the transmission bitrate is to ensure the playback quality of the teaching video after transmission. The playback quality of the teaching video after transmission is considered the transmission quality. Therefore, the higher the transmission quality of each video segment, the better the transmission scheme. Thus, the fitness function is positively correlated with the transmission quality of each video segment. On the other hand, if there is a large difference in transmission quality between adjacent video segments, sudden changes in image and audio quality may occur, affecting the user's viewing experience. Therefore, the greater the fluctuation in transmission quality between adjacent video segments, the worse the transmission scheme. Thus, the fitness function is negatively correlated with the volatility of the transmission quality sequence.

[0047] In one embodiment, accurate prediction of transmission quality directly affects the calculation of the fitness function, and thus the optimization result of the particle swarm optimization algorithm. To accurately predict the transmission quality of each video segment, a prediction model needs to be constructed. The input of the prediction model is the network state, as well as the image complexity, semantic complexity, and bitrate of the video segment. The output is the transmission quality of the video segment at that bitrate, where the transmission quality is the VMAF of the video segment after transmission is complete.

[0048] Understandably, VMAF is a video quality assessment tool launched by Netflix. It is an objective evaluation indicator of video quality and is common knowledge to those skilled in the art, so it will not be elaborated here.

[0049] The prediction model employs a fully connected neural network or a multilayer perceptron. For example, please refer to [link to example]. Figure 2 This is a structural diagram of a prediction model according to an embodiment of this application. The prediction model includes an input layer, two hidden layers, and an output layer. The input layer is used to receive the image complexity, semantic complexity, and bitrate (corresponding to...) of the video segment. Figure 2 shown , and and network status Network status in this example The remaining bandwidth is used. The hidden layer is used to transform the dimension of the input data of the input layer to obtain a 3-row, 1-column feature vector; the output layer maps the feature vector to the transmission quality VMAF of the video segment.

[0050] It should be noted that, in other embodiments, the network state also includes at least one of network volatility, packet loss rate, and transmission latency. To avoid inconsistencies in data units, standardization is required before inputting the network state, as well as the image complexity, semantic complexity, and bitrate of the video segment into the prediction model.

[0051] The training process of the prediction model includes: collecting the bitrate and network status during the transmission of video segment samples, and obtaining the VMAF of the received video segment samples after transmission as a quality label; inputting the bitrate, network status, and image complexity and semantic complexity of the video segment samples into the prediction model to obtain the prediction result; iteratively updating the prediction model based on the mean squared error loss between the prediction result and the quality label until the number of iterations is greater than the preset number, or the mean squared error loss is less than the preset loss, and then completing the training.

[0052] The preset number of attempts is 100; the preset loss is 0.01.

[0053] In this way, the trained prediction model can learn the mapping relationship between transmission quality and input information (i.e., network state and the image complexity, semantic complexity and bit rate of video segments), enabling the prediction model to accurately output the transmission quality of each video segment.

[0054] In one embodiment, after obtaining the transmission quality of each video segment, the transmission quality of each video segment is arranged according to the chronological order of the video segments to obtain a transmission quality sequence. Then, a fitness function can be calculated. Satisfying the relation:

[0055] ; The number of video segments, For video segments The maximum value of image complexity and semantic complexity in the medium. It is the sum of the maximum values ​​of image complexity and semantic complexity across all video segments; For video segments Duration The total duration of the instructional videos. , and Video segments Video segment and video clips Transmission quality; For the change threshold, For step function, when hour, =1, otherwise, =0; For penalty weights.

[0056] in, The maximum value at which changes in transmission quality will not affect the viewing experience; this can be set based on experience. In this embodiment, the change threshold is... The value is 6, meaning that when the transmission quality change of adjacent video segments is greater than 6, it is considered to have an impact.

[0057] Understandably, It can be regarded as a weighted transmission quality of each video segment, with more attention paid to video segments with high complexity and long duration. The higher the weighted transmission quality, the better the transmission quality of each video segment after transmission according to the transmission plan. Used to reflect video segments and video clips Whether the change in transmission quality exceeds a threshold is checked. If the change exceeds the threshold, it indicates that the change in transmission quality has affected the viewing experience. In this case, [the system will...]. The value is recorded as 1. Conversely, if the change in transmission quality is not greater than the change threshold, it means that the change in transmission quality has not affected the viewing experience. The value of is denoted as 1; This is used to characterize the percentage of times changes in transmission quality in instructional videos affect the viewing experience. A higher percentage indicates greater volatility in the transmission quality sequence, a worse viewing experience for the user, and a smaller value for the fitness function. Penalty weight. In this application embodiment, the penalty weight is used to balance the impact of the volatility of the transmission quality sequence on the fitness function. The value is 1.

[0058] In this way, multiple transmission schemes are obtained, along with the fitness function for each transmission scheme. The fitness function is used to reflect the advantages and disadvantages of each transmission scheme.

[0059] S103, the particle swarm optimization algorithm is used to iteratively update the multiple transmission schemes to obtain the optimized transmission bitrate result of the teaching video.

[0060] In one embodiment, the multiple transmission schemes correspond to multiple particles in a particle swarm optimization algorithm, and the number of transmission schemes is 50, meaning the number of particles in the particle swarm optimization algorithm is 50. The particle swarm optimization algorithm iteratively updates the positions of each particle. In response to the fact that the increase in the fitness function of the optimal position of the population is less than a preset increment during multiple iterations, the optimal position of the population in the last iteration is taken as the transmission bitrate optimization result. The teaching video is transmitted according to the transmission bitrate optimization result to ensure the transmission quality of the teaching video.

[0061] It should be noted that, for those skilled in the art, various modifications and improvements can be made without departing from the concept of this application, and these all fall within the scope of protection of this application. Therefore, the scope of protection of this patent application shall be determined by the appended claims.

Claims

1. A method for optimizing the transmission bitrate of teaching videos based on deep learning, characterized in that, The optimization method includes: dividing the teaching video into multiple video segments based on the image complexity and semantic complexity of each timestamp; assigning an initial bitrate to each video segment to obtain multiple transmission schemes; and calculating the fitness function of each transmission scheme, including: ; The number of video segments, For video segments The maximum value of image complexity and semantic complexity in the medium. It is the sum of the maximum values ​​of image complexity and semantic complexity across all video segments; For video segments Duration The total duration of the instructional videos. , and Video segments Video segment and video clips Transmission quality; For the change threshold, For step function, when hour, =1, otherwise, =0; As a penalty weight, the fitness function is positively correlated with the transmission quality of each video segment and negatively correlated with the volatility of the transmission quality sequence; the multiple transmission schemes are iteratively updated using a particle swarm optimization algorithm to obtain the optimized transmission bitrate of the teaching video; The transmission quality is obtained by a prediction model, whose inputs are the network state, the image complexity, semantic complexity, and bitrate of the video segment, and whose output is the transmission quality of the video segment.

2. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The method for obtaining the complexity of an arbitrary timestamp image is as follows: Obtain the entropy value of the gray-level co-occurrence matrix of the video image corresponding to the timestamp; interpolate the video images of the previous and next adjacent timestamps to obtain a regression image, and use the average pixel value of the difference map between the regression image and the video image as the regression complexity; use the sum of the normalized entropy value and the regression complexity as the image complexity of the timestamp.

3. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The method for obtaining the semantic complexity of arbitrary timestamps is as follows: The audio information of the timestamp preset window is converted into a frequency space, and the normalized information entropy of each frequency energy value in the frequency space is used as the semantic complexity.

4. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 3, characterized in that, Timestamp semantic complexity satisfy: ; The number of frequency types, For the first The energy value of a certain frequency, It is the sum of the energy values ​​of all frequencies. This represents the maximum value of the information entropy.

5. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The teaching videos are divided into multiple video segments based on the image complexity and semantic complexity of each timestamp, including: Construct a complexity sequence based on the maximum value of image complexity and semantic complexity for each timestamp; The complexity sequence is subjected to ordered sample clustering to obtain multiple time periods, which correspond to multiple video segments.

6. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, Several transmission schemes were obtained, including: Obtain the adjustment coefficient The adjustment coefficient is a random number between 0 and 1; The initial bitrate of each video segment is calculated based on the adjustment coefficient, thus obtaining the adjustment coefficient. Corresponding transmission scheme, video segment initial bitrate for: ; and These are the minimum and maximum bitrate values, respectively. For adjustment coefficients, For video segments The maximum values ​​of image complexity and semantic complexity are determined; the adjustment coefficient is then adjusted. The value of is used to obtain multiple transmission schemes.

7. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The training process of the prediction model includes: The bit rate and network status during the transmission of video segment samples are collected, and the VMAF of the received video segment samples after the transmission is completed is obtained as a quality label. The prediction model is input with the bitrate, network status, and image and semantic complexity of the video segment samples to obtain the prediction results. The prediction model is iteratively updated based on the mean squared error loss between the prediction results and the quality labels until the number of iterations is greater than the preset number, or the mean squared error loss is less than the preset loss, at which point the training is complete.

8. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The network status includes at least one of remaining bandwidth, network volatility, packet loss rate, and transmission delay.

9. The method for optimizing the transmission bitrate of teaching videos based on deep learning according to claim 1, characterized in that, The multiple transmission schemes are iteratively updated using the particle swarm optimization algorithm to obtain the transmission bitrate optimization result of the teaching video, including: in response to the fact that the increase in the fitness function of the population optimal position in the continuous multiple iterations is less than the preset increment, the population optimal position in the last iteration is taken as the transmission bitrate optimization result.

Images