A source-channel joint rate allocation system for deep image compression

Abstract

The present invention relates to a source-channel joint rate allocation system for deep image compression in a mixed context. The system includes a context importance analysis module, a packetization module, a signaling lossless compression module, and a rate allocation module. The present invention implements allocation through the rate allocation module, and finds the optimal rate allocation strategy for the data packets based on the context importance of the potential representation generated by the deep image encoder, that is, selecting different channel coding rates for data packets with different context importance, so as to maximize the image reconstruction quality and classification accuracy in a wireless communication environment with limited channel bandwidth and various noise interferences.

Description

Technical Field

[0001] The present invention relates to the technical field of channel resource allocation, and in particular to a source-channel joint rate allocation system for deep image compression in a mixed context. Background Art

[0002] Due to the limited available bandwidth of wireless communication channels and the impact of various noise and multipath phenomena, such as slow and fast fading, wireless channels have significantly higher bit error rates than wired environments, limiting the application of multimedia services in wireless communications. Furthermore, practical communication systems must constrain code length complexity and latency. Traditional source-channel separation coding cannot achieve high performance while maintaining these constraints. Therefore, multimedia data transmission in practical communication systems typically utilizes joint source-channel coding. This method combines source and channel coding, achieving better transmission performance under practical channel conditions.

[0003] The design of joint source-channel rate allocation based on signal source characteristics is a research hotspot in modern communication systems. With the widespread application of deep learning in image compression, joint source-channel rate allocation for the code streams generated by deep image compression has become a widely studied topic.

[0004] Unequal Error Protection (UEP) is an effective technique for allocating source and channel bit rates, as opposed to equal error protection (EEP). When communication network resources are limited, equal error protection schemes can result in insufficient protection of important information, severely degrading decoding quality. Unequal error protection, on the other hand, employs targeted channel protection mechanisms of varying strength based on the importance of different parts of the bitstream to the target context. This involves allocating different channel coding rates based on the importance of the bitstream. This allows for more efficient allocation of communication resources and achieves superior performance compared to equal protection schemes. Unequal Error Protection (UEP) is a form of joint source and channel coding. Although UEP reduces noise immunity for non-critical bitstreams, it improves the overall system error tolerance. However, finding an optimal UEP scheme for mobile multimedia services is a challenging problem, dependent on many factors, including the source rate-distortion function, the specific communication channel scenario, the available bandwidth, and the available channel coding rate.

[0005] According to research, existing studies on joint source-channel unequal error protection for image transmission all consider code streams that have forward and backward dependencies and are transmitted in descending order of importance, such as JPEG2000, SPIHT, HEVC, and other source coding methods that use the multi-resolution characteristics of wavelet transform for progressive image coding and transmission. In addition, the optimization objectives all consider traditional image reconstruction indicators, such as peak signal-to-noise ratio, or subjective perceptual quality indicators, such as structural similarity.

[0006] The latent representation codestreams generated by deep learning-based image encoders have different structure and characteristics than those generated by traditional image encoders. Deep image compression coding methods based on semantic masks are a prime example. These methods generate latent representation codestreams that only have progressive semantic importance and no contextual dependencies.

[0007] Based on extensive research, the inventors learned that there is currently no source-channel joint rate allocation scheme for deep image coding in mixed contexts such as image reconstruction quality and classification accuracy. However, in the era of intelligent communications, the target context is often not a simple single context, but a mixed context of multiple targets. Deep image coding methods with significant performance gains have been increasingly widely used, and performing corresponding unequal error protection design is an indispensable step in applying deep image coding to actual communication systems. Therefore, based on the gap in this research field and its important research significance and application value, the present invention proposes a source-channel joint rate allocation method for deep image compression in a mixed context. Summary of the Invention

[0008] The purpose of the present invention is to provide a source-channel joint rate allocation system for deep image compression in a mixed context. Based on the contextual importance of the potential representation generated by deep image coding and its unique code stream characteristics, a corresponding rate allocation scheme under unequal error protection is designed to maximize the reconstruction quality and classification accuracy of the reconstructed image at the receiving end under the condition of limited total transmission rate.

[0009] To achieve the above object, the technical solution adopted by the present invention is:

[0010] A source-channel joint rate allocation system for deep image compression in mixed contexts, comprising a context importance analysis module, a packetization module, a signaling lossless compression module, and a rate allocation module;

[0011] The context importance analysis module includes a reconstruction importance analysis unit, a semantic importance analysis unit and a context importance calculation unit;

[0012] The semantic importance analysis unit is used to input the original image, analyze the importance of each pixel of the original image to the reconstructed image classification result, and then map it to the corresponding plane position of the potential representation in the latent space generated by the deep encoder, thereby outputting the semantic importance size φ of the potential representation at each spatial position class ;

[0013] The reconstruction importance analysis unit is used to input the potential representation output by the depth encoder, analyze and output the importance of each plane position in the potential representation generated by depth image compression to image reconstruction rec ;

[0014] The context importance calculation unit is used to input the reconstruction importance of the potential representation output by the reconstruction importance analysis unit and the semantic importance of the potential representation output by the semantic importance analysis unit, calculate and output the context importance of the potential representation; context importance refers to assigning different weights α and β to the reconstruction importance and semantic importance according to the different requirements of the specific target context for image reconstruction quality and image classification accuracy, and weighted calculation to obtain the final context importance size φ of the potential representation lat , that is, φ lat =α·φ rec +β·φ class , where 0≤α≤1, 0≤β≤1, and α+β=1;

[0015] The signaling lossless compression module is used to input the auxiliary signaling q1 output by the depth encoder and the auxiliary signaling q2 output by the subpackaging module, perform lossless compression on them, and output the compressed auxiliary signaling code stream;

[0016] The packetization module is used to input the potential representations output by the deep encoder, sort the potential representations in descending order of contextual importance, and then equally divide the potential representations into N data packets. The number of potential representations in each data packet is Y. Since the potential representations are quantized using a fixed-length method, the code length of each data packet is also the same, denoted as K. A data packet is the minimum unit for subsequent channel coding protection. At the same time, the packetization module will generate a corresponding auxiliary signaling q2 to inform the code rate allocation module which potential representations are contained in each data packet, to assist it in performing the corresponding channel code rate allocation.

[0017] The channel code rate r is defined as the ratio of the number of information bits to the total number of bits after channel coding. The optional set of channel code rates is represented by r set ={r1,r2,...,r Q}, where r1>r2>…>r Q , the bit error rate corresponding to each channel code rate is expressed as And p1>p2>…>pQ ;

[0018] The code rate allocation module is used to input the compressed auxiliary signaling code stream output by the signaling lossless compression module and the potential representation data packet output by the packetization module, as well as the context importance auxiliary information from the context importance analysis module, and find an optimal channel code rate allocation strategy Ψ under a given channel total transmission rate R to guide the encoding of the channel encoder;

[0019] Specifically, first, the auxiliary signaling q1 and q2 are subjected to the highest level of channel protection, and the minimum channel code rate in the available channel code rate set is selected for encoding. The code rate occupied by the auxiliary signaling code stream after channel encoding is recorded as R'; secondly, based on the auxiliary signaling q2 output from the packetization module, the potential representation contained in each data packet is obtained, and the context importance analysis module is used to obtain the size of the context importance corresponding to each potential representation. Then, the context importance of each data packet is obtained by averaging the context importance of the potential representations in the same data packet. Therefore, Next, the channel protection level of each data packet is selected according to the importance of the data packet's context. The greater the context importance, the higher the channel protection level should be. A lower channel code rate is selected and more redundant bits are applied for protection. Since the wireless channel determines the total available code rate of the entire code stream is R, the upper limit of the code length after channel coding of the potential representation code stream is The channel rate allocation strategy Ψ for the entire potential representation code stream is expressed as in Indicates the channel code rate used by each data packet, The corresponding bit error rate BER is

[0020] Each potential representation data packet is channel coded according to the assigned channel code rate, and then transmitted through the noisy channel to reach the receiver, so that the comprehensive performance φ(Ψ) of the reconstruction quality and classification accuracy of the reconstructed image at the receiver is optimized.

[0021] The following matrix X is used to represent the channel code rate distribution of the entire potential representation code stream:

[0022] r1r2···r Q

[0023]

[0024] Among them, x ij The value is 0 or 1. When x ij =1, indicating the i-th data packet D i The jth channel code rate r is selected j , when x ij=0, it means the i-th data packet D i The jth channel code rate r is not selected j ; Since each data packet only selects one channel code rate, there is And because the total length of the potential representation code stream after channel coding cannot exceed the available code rate Therefore Therefore the constraints are: The solution to the optimal strategy for channel rate allocation is expressed as the optimization problem described by formula (1):

[0025]

[0026] The genetic algorithm in the heuristic algorithm is used to find the optimal channel code rate allocation strategy, which specifically includes the following steps:

[0027] Step S1, individual encoding; for each data packet D i The possible channel rate choices are represented by encoding a Q-bit binary sequence, where Q is the total number of possible channel rates. The combination of the binary representations of the rate choices for each of the N packets in the stream gives one possible rate allocation strategy for the entire potential representation of the stream, i.e., an individual in the population.

[0028] Step S2: Randomly generate a certain number of individuals as the initial population, representing a set of initial candidate bit rate allocation schemes, and set the maximum evolutionary generation GEN, algorithm convergence condition, selection strategy, crossover probability, and mutation probability;

[0029] Step S3: Design a fitness function for calculating the individual fitness value; the optimization objective function is the fitness function f(Ψ), therefore,

[0030]

[0031] Step S4, setting the initialization generation G=1, indicating the first generation population;

[0032] Step S5: Calculate the fitness of each individual in the population using the fitness function, and select crossover individuals based on the fitness;

[0033] Step S6: Perform a crossover operation on the selected individuals to generate new individuals, replace the individuals with poor fitness in the G generation population, and keep the population size unchanged;

[0034] Step S7: randomly replace the selected individuals in the population according to the population mutation probability, modify some genes, and generate new individuals;

[0035] Step S8: Generate the next generation population through the above steps S6 to S8, G=G+1;

[0036] Step S9: Determine whether G is greater than GEN or whether the fitness of the optimal individual meets the convergence condition. If so, the algorithm terminates and exits the loop. The individual with the highest fitness in the population is used as the solution to the original problem, that is, the bit rate allocation strategy is obtained; if not, repeat steps S5-S9.

[0037] After adopting the above scheme, the present invention finds the optimal bit rate allocation strategy for the data packets based on the contextual importance of the potential representation generated by the deep image encoder, that is, selects different channel coding rates for data packets with different contextual importance, and maximizes the image reconstruction quality and classification accuracy in a wireless communication environment with limited channel bandwidth and various noise interferences. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the overall framework of the source-channel joint rate allocation system for deep image compression in a mixed context proposed by the present invention;

[0039] Figure 2 A schematic diagram of channel code rate allocation based on unequal error protection for data packets divided by a potential representation code stream generated for deep image coding according to an embodiment of the present invention;

[0040] Figure 3 A schematic diagram of a process for finding an optimal bit rate allocation strategy using a genetic algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0042] The present invention proposes a source-channel joint rate allocation system for deep image compression in a mixed context. Figure 1 As shown, the system includes a context importance analysis module, a packetization module, a signaling lossless compression module and a bit rate allocation module.

[0043] Among them, the context importance analysis module includes a reconstruction importance analysis unit, a semantic importance analysis unit and a context importance calculation unit.

[0044] The semantic importance analysis unit is used to input the original image, analyze the importance of each pixel of the original image to the reconstructed image classification result, and then map it to the corresponding plane position of the potential representation in the latent space generated by the deep encoder, thereby outputting the semantic importance size φ of the potential representation at each spatial position class Semantic importance refers to its importance to the classification accuracy of the final reconstructed image. This example uses the VGG16 neural network to classify images and employs class activation mapping (CAM) technology to calculate the contribution of each pixel to the image classification accuracy. Since convolutional neural networks have the characteristic of keeping spatial positions relatively unchanged, the importance of pixels is mapped into the latent space, thereby obtaining the semantic importance of the potential representation of each plane position.

[0045] Reconstruction importance analysis unit: used to input the potential representation output by the depth encoder, analyze and output the importance of each plane position in the potential representation generated by the depth image compression to the image reconstruction rec . This example uses channel complexity as the evaluation criterion for reconstruction importance. Assume that the latent space generated by the deep image encoder has M channels of potential representation vectors, that is, each plane position has M potential representation values. Calculate the variance of the M potential representation values ​​at each plane position of the latent representation vector. The larger the variance, the greater the entropy of the position. The greater the entropy, the greater the amount of information. Therefore, the potential representation of the plane position contributes more to the reconstruction quality of the image.

[0046] The context importance calculation unit is used to input the reconstruction importance of the potential representation output by the reconstruction importance analysis unit and the semantic importance of the potential representation output by the semantic importance analysis unit, calculate and output the context importance of the potential representation. Context importance refers to assigning different weights α and β to the reconstruction importance and semantic importance according to the different requirements of the specific target context for image reconstruction quality and image classification accuracy, and the weighted calculation is used to obtain the final context importance of the potential representation φ. lat , that is, φ lat =α·φ rec +β·φ class , where 0≤α≤1, 0≤β≤1, and α+β=1. Depending on the target context, different combinations of α and β are possible. It should be noted that although the values ​​of α and β are variable, they are not variables in the designed rate allocation algorithm. When the target context is determined, their values ​​are unique and fixed.

[0047] The signaling lossless compression module is used to input the auxiliary signaling q1 output by the depth encoder and the auxiliary signaling q2 output by the subpacketization module, perform lossless compression on them, and output the compressed auxiliary signaling code stream.

[0048] The packetization module inputs the latent representations output by the deep encoder, sorts them in descending order of contextual importance, and then equally divides the latent representations into N data packets, each containing Y number of latent representations. Because the latent representations are quantized using a fixed-length method, the code length of each data packet is also the same, denoted as K. A data packet is the minimum unit for subsequent channel coding protection. Simultaneously, the packetization module generates a corresponding auxiliary signaling q2 to inform the rate allocation module which latent representations are contained in each data packet, assisting it in making appropriate channel rate allocations.

[0049] The channel code rate r is defined as the ratio of the number of information bits to the total number of bits after channel coding. The optional set of channel code rates is represented by r set ={r1,r2,...,r Q}, where r1>r2>…>r Q , then the bit error rate BER (Bit Error Rate) corresponding to each channel code rate can be expressed as And p1>p2>…>p Q .

[0050] The code rate allocation module is used to input the compressed signaling code stream output by the signaling lossless compression module and the potential representation data packet output by the subpacketization module, as well as the context importance auxiliary information from the context importance analysis module, and derive the code rate allocation scheme based on this information. Specifically, first, the auxiliary signaling code stream input by the signaling lossless compression module plays a key role in the final optimization goal, and has a short length and low overhead. Therefore, the highest level of channel protection is performed on it (q1, q2), that is, the minimum channel code rate in the available channel code rate set is selected for encoding. The code rate occupied by the auxiliary signaling code stream after channel coding is recorded as R'. Secondly, according to the auxiliary signaling q2 output from the subpacketization module, it can be known which potential representations are specifically contained in each data packet. Through the context importance analysis module, the size of the context importance corresponding to each potential representation can be known. Then, the context importance size of each data packet is obtained by averaging the context importance of the potential representations in the same data packet. Therefore, Next, the channel protection level of each data packet is selected based on the context importance of the data packet. The greater the context importance, the higher the channel protection level should be, that is, a lower channel code rate should be selected and more redundant bits should be applied for protection. The process of applying corresponding channel redundant bits for protection to each data packet is as follows: Figure 2As shown. Since the wireless channel determines the total available code rate of the entire code stream is R, the upper limit of the code length after channel coding of the potential representation code stream is The channel rate allocation strategy Ψ for the entire potential representation code stream can be expressed as in Indicates the channel code rate used by each data packet, The corresponding bit error rate BER is

[0051] The goal of the code rate allocation module in the present invention is to find an optimal channel code rate allocation strategy Ψ under a given total channel transmission rate R. Each potential representation data packet is channel-coded according to the allocated channel code rate and then transmitted through a noisy channel to the receiving end, so that the comprehensive performance φ(Ψ) of the reconstruction quality and classification accuracy of the reconstructed image at the receiving end is optimized.

[0052] The following matrix X is used to represent the channel code rate distribution of the entire potential representation code stream:

[0053] r1r2···r Q

[0054]

[0055] Among them, x ij The value is 0 or 1. When x ij =1, indicating the i-th data packet D i The jth channel code rate r is selected j , when x ij =0, it means the i-th data packet D i The jth channel code rate r is not selected j ; Since each data packet only selects one channel code rate, there is And because the total length of the potential representation code stream after channel coding cannot exceed the available code rate Therefore Therefore the constraints are: The solution to the optimal strategy for channel rate allocation is expressed as the optimization problem described by formula (1):

[0056]

[0057] In this example, the channel model selects a Gaussian white noise continuous channel with limited bandwidth and limited average power. Therefore, in the present invention, the bit error rate in r is the corresponding channel coding rate, erfc is the complementary error function, SNR is the signal-to-noise ratio. In this embodiment, the solution to the optimal strategy for channel code rate allocation can be expressed as the optimization problem described by formula (2).

[0058]

[0059] The present invention uses the genetic algorithm in the heuristic algorithm to find the optimal channel code rate allocation strategy. The specific process is as follows: Figure 3 As shown, the following steps are included:

[0060] Step S1, individual encoding. Since each data packet D i Only one channel coding rate will be selected, so for each data packet D i The possible channel code rate selections can be encoded as a Q-bit binary sequence, where Q is the total number of selectable channel code rates. For example, when Q = 4, if data packet D1 selects channel code rate r1 for encoding, then D1's channel code rate selection can be encoded as 1000. Similarly, if channel code rate r2 is selected for encoding, D1's code rate selection can be represented as 0100, and so on. Finally, combining the binary code representations of the code rate selections of the N data packets of the code stream is one of the possible code rate allocation strategies for the entire potential representation of the code stream, that is, an individual in the population;

[0061] Step S2: Randomly generate a certain number of individuals as the initial population, representing a set of initial candidate bit rate allocation schemes, and set the maximum evolutionary generation GEN, algorithm convergence condition, selection strategy, and mutation probability;

[0062] Step S3: Design a fitness function for calculating individual fitness values. The optimization objective function of the present invention is the fitness function f(Ψ), so,

[0063]

[0064] Step S4, setting the initialization generation G=1, indicating the first generation population;

[0065] Step S5: Calculate the fitness of each individual in the population using the fitness function, and select crossover individuals based on the fitness;

[0066] Step S6: Perform a crossover operation on the selected individuals to generate new individuals, replace the individuals with poor fitness in the G generation population, and keep the population size unchanged;

[0067] Step S7: randomly replace the selected individuals in the population according to the population mutation probability, modify some genes, and generate new individuals;

[0068] Step S8: Generate the next generation population through the above steps S6 to S8, G=G+1;

[0069] Step S9: Determine whether G is greater than GEN or whether the fitness of the optimal individual meets the convergence condition. If so, the algorithm terminates, exits the loop, and takes the individual with the highest fitness in the population as the solution to the original problem (bitrate allocation strategy); if not, repeat steps S5-S9.

[0070] In summary, the present invention seeks the optimal bit rate allocation strategy for the potential representation data packets generated by the depth image encoder based on the contextual importance of the packets, that is, selecting different channel coding rates for data packets with different contextual importance, so as to maximize the image reconstruction quality and classification accuracy in a wireless communication environment with limited channel bandwidth and various noise interferences.

[0071] The above description is merely an embodiment of the present invention and does not limit the technical scope of the present invention. Therefore, any minor modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A source-channel joint rate allocation system for deep image compression in a mixed context, characterized by: The system includes a context importance analysis module, a packetization module, a signaling lossless compression module, and a bit rate allocation module; The context importance analysis module includes a reconstruction importance analysis unit, a semantic importance analysis unit and a context importance calculation unit; The semantic importance analysis unit is used to input the original image, analyze the importance of each pixel of the original image to the reconstructed image classification result, and then map it to the corresponding plane position of the potential representation in the latent space generated by the deep encoder, thereby outputting the semantic importance size φ of the potential representation at each spatial position class ; The reconstruction importance analysis unit is used to input the potential representation output by the depth encoder, analyze and output the importance of each plane position in the potential representation generated by depth image compression to image reconstruction rec ; The context importance calculation unit is used to assign different weights α and β to the reconstruction importance and semantic importance, and the weighted calculation is used to obtain the final potential representation context importance size φ lat , that is, φ lat =α·φ rec +β·φ class , where 0≤α≤1, 0≤β≤1, and α+β=1; The signaling lossless compression module is used to input the auxiliary signaling q1 output by the depth encoder and the auxiliary signaling q2 output by the subpackaging module, perform lossless compression on them, and output the compressed auxiliary signaling code stream; The packetization module is used to generate a corresponding auxiliary signaling q2 to inform the code rate allocation module which potential representations each data packet contains, so as to assist it in performing corresponding channel code rate allocation; The code rate allocation module is used to find an optimal channel code rate allocation strategy Ψ; Specifically, first, the auxiliary signaling q1 and q2 are subjected to the highest level of channel protection, and the minimum channel code rate in the available channel code rate set is selected for encoding. The code rate occupied by the auxiliary signaling code stream after channel encoding is recorded as R'; secondly, based on the auxiliary signaling q2 output from the packetization module, the potential representation contained in each data packet is obtained, and the context importance analysis module is used to obtain the size of the context importance corresponding to each potential representation. Then, the context importance of each data packet is obtained by averaging the context importance of the potential representations in the same data packet. Therefore, Where Y is the number of potential representations in the data packet. Next, the channel protection level of each data packet is selected according to the importance of the data packet's context. The greater the context importance, the higher the channel protection level should be. A lower channel code rate should be selected and more redundant bits should be applied for protection. Since the wireless channel determines the total available code rate of the entire code stream is R, the upper limit of the code length after channel coding of the potential representation code stream is The channel rate allocation strategy Ψ for the entire potential representation code stream is expressed as in Indicates the channel code rate used by each data packet, The corresponding bit error rate BER is N is the number of packets into which the potential representation is equally divided; Each potential representation data packet is channel coded according to the assigned channel code rate, and then transmitted through the noisy channel to reach the receiver, so that the comprehensive performance φ(Ψ) of the reconstruction quality and classification accuracy of the reconstructed image at the receiver is optimized.

2. The source-channel joint rate allocation system for deep image compression in a mixed context according to claim 1, characterized in that: The following matrix X is used to represent the channel code rate distribution of the entire potential representation code stream: r1 r2…r Q Among them, x ij The value is 0 or 1. When x ij =1, indicating the i-th data packet D i The jth channel code rate r is selected j , when x ij =0, it means the i-th data packet D i The jth channel code rate r is not selected j ; Since each data packet only selects one channel code rate, there is And because the total length of the potential representation code stream after channel coding cannot exceed the available code rate Therefore Where K is the code length of each data packet; therefore, the constraints are: The solution to the optimal strategy for channel rate allocation is expressed as the optimization problem described by formula (1):

3. The source-channel joint rate allocation system for deep image compression in a mixed context according to claim 2, characterized in that: The genetic algorithm in the heuristic algorithm is used to find the optimal channel code rate allocation strategy, which specifically includes the following steps: Step S1, individual encoding; for each data packet D i The possible channel rate choices are represented by encoding a Q-bit binary sequence, where Q is the total number of possible channel rates. The combination of the binary representations of the rate choices for each of the N packets in the stream gives one possible rate allocation strategy for the entire potential representation of the stream, i.e., an individual in the population. Step S2: Randomly generate a certain number of individuals as the initial population, representing a set of initial candidate bit rate allocation schemes, and set the maximum evolutionary generation GEN, algorithm convergence condition, selection strategy, crossover probability, and mutation probability; Step S3: Design a fitness function for calculating the individual fitness value; the optimization objective function is the fitness function f(Ψ), therefore, Step S4, setting the initialization generation G=1, indicating the first generation population; Step S5: Calculate the fitness of each individual in the population using the fitness function, and select crossover individuals based on the fitness; Step S6: Perform a crossover operation on the selected individuals to generate new individuals, replace the individuals with poor fitness in the G generation population, and keep the population size unchanged; Step S7: randomly replace the selected individuals in the population according to the population mutation probability, modify some genes, and generate new individuals; Step S8: Generate the next generation population through the above steps S6 to S8, G=G+1; Step S9: Determine whether G is greater than GEN or whether the fitness of the optimal individual meets the convergence condition. If so, the algorithm terminates and exits the loop. The individual with the highest fitness in the population is used as the solution to the original problem, that is, the bit rate allocation strategy is obtained; if not, repeat steps S5-S9.

Images