Broadcast signal interference correction method, system, medium and equipment

Abstract

The invention discloses a broadcast signal interference correction method and system, a medium and equipment. A broadcast signal is collected in real time and converted into a digital signal, preprocessing is carried out to obtain a first audio signal, feature extraction is carried out according to the first audio signal to obtain audio features, and corresponding preprocessing is carried out on different types of audio features; integrating the audio features to generate an audio feature matrix, inputting the audio feature matrix into a preset fusion model to output interference analysis, receiving the interference analysis, obtaining a corresponding template in combination with a first confidence coefficient, and filling dynamic parameters to generate a correction strategy and the confidence coefficient; wherein the fusion model comprises a decision tree unit and a multi-modal fusion decision unit; and obtaining a correction index corrected by the correction strategy, generating an evaluation index in combination with interference analysis and the correction strategy, and performing adaptive feedback on the fusion model through the evaluation index. Intelligent detection and automatic correction of broadcast signal interference are realized, and the stability and signal quality of a broadcast system are improved.

Description

Technical Field

[0001] The present application belongs to the field of broadcast signal processing technology, and specifically relates to a broadcast signal interference correction method, system, medium and device. Background Art

[0002] With the rapid development of broadcasting technology, the quality and stability of broadcast signals have become key factors in ensuring the smooth dissemination of broadcast content. Traditional broadcast signal interference and correction methods rely primarily on hardware equipment and fixed algorithms. For example, these methods use spectrum analyzers, filters, equalizers, and other instruments to process signals. By monitoring signal parameters such as frequency and amplitude, they identify and suppress interfering signals, thereby improving broadcast quality. However, as the complexity of the broadcasting environment increases, traditional methods have certain limitations in the following areas:

[0003] Limited processing capacity: As broadcast systems expand in size, the real-time and accuracy requirements for signal processing increase. Traditional methods may experience delays or misjudgments when processing large amounts of data.

[0004] Lack of intelligent analysis: Traditional methods mainly rely on manually set parameters and rules, lack a deep understanding of signal characteristics and intelligent analysis, and are difficult to achieve adaptive optimization.

[0005] The present invention realizes intelligent analysis and adaptive correction of broadcast signals by introducing fusion model technology and utilizing advantages in data processing and pattern recognition. Summary of the Invention

[0006] This application proposes a broadcast signal interference correction method, system, medium and device to solve the problem that the above-mentioned traditional methods cannot meet user needs.

[0007] A first aspect of the present application provides a broadcast signal interference correction method, the method comprising:

[0008] Collecting broadcast signals in real time and converting them into digital signals, performing preprocessing to obtain a first audio signal, performing feature extraction based on the first audio signal to obtain audio features, and performing corresponding preprocessing on different types of audio features;

[0009] Integrate audio features to generate an audio feature matrix and input it into a preset fusion model to output an interference analysis, receive the interference analysis and combine it with the first confidence level to obtain a corresponding template, fill in dynamic parameters to generate a correction strategy and credibility; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit;

[0010] The correction index after correction by the correction strategy is obtained, and the evaluation index is generated by combining the interference analysis and the correction strategy. The fusion model is adaptively fed back through the evaluation index.

[0011] The above solution converts broadcast signals into digital signals through real-time acquisition, avoiding the attenuation and distortion problems during analog signal transmission and ensuring the integrity of the original signal characteristics. The interference analysis capability is enhanced by combining a fusion model with a decision tree unit and a multimodal fusion decision unit. Regularized feature splitting logic is used to quickly locate the interference center. The multimodal fusion decision unit is combined to compensate for the limitations of single analysis and accurately identify complex interference. Corrected signal indicators are obtained, and a closed-loop evaluation system is formed by combining interference analysis results with correction strategies. In the face of time-varying interference environments, real-time adjustments to the closed-loop feedback mechanism ensure the timeliness of interference analysis and correction strategies, improve the stability and signal quality of the broadcast system, and meet the needs of stable signal output from the broadcast system.

[0012] Furthermore, the feature extraction based on the first audio signal to obtain the audio feature is specifically:

[0013] Loading an audio file via an audio processing library to receive a first audio signal;

[0014] Extract audio features in parallel, the audio features including: unique identifier, basic audio file information, Mel-frequency cepstral coefficients, Mel-frequency spectrogram, MFCC coefficients, zero-crossing rate, and spectral centroid; wherein the basic audio file information includes: duration, sampling rate, number of channels, and format;

[0015] The audio signal is formatted as JSON for subsequent processing.

[0016] Furthermore, the corresponding extraction method for different types of audio features is specifically as follows:

[0017] For non-interference audio features, lightweight extraction is performed;

[0018] For audio features affected by narrowband interference, frequency domain analysis is enhanced and outlier suppression is strengthened;

[0019] For audio features affected by dynamic interference, we strengthen the capture of temporal features and introduce dynamic filtering and attention mechanisms.

[0020] Furthermore, the preprocessing further includes:

[0021] Performing interpolation / dropping processing on invalid values ​​in audio features, wherein the invalid values ​​include missing values ​​and infinite values;

[0022] When the invalid value loss ratio exceeds the preset threshold, loss processing is performed;

[0023] When the loss ratio does not reach the preset threshold, linear interpolation is performed on invalid values;

[0024] Each audio feature is normalized using a linear function and correlation analysis is performed.

[0025] In a possible implementation method of the first aspect, integrating audio features to generate an audio feature matrix and inputting the matrix into a preset fusion model is specifically as follows:

[0026] Regularize the audio feature matrix through the decision tree unit to obtain the preliminary interference type and confidence level;

[0027] The decision tree unit adopts a preset CART decision tree, traverses the path from the root node to the leaf node, and reaches the leaf node to obtain the preliminary interference type;

[0028] The interference type label obtained by reaching the leaf node and the training sample of the leaf node are used as the confidence.

[0029] In a possible implementation method of the first aspect, the fusion model further includes:

[0030] Encode the modal data according to the multimodal fusion decision unit, align them to the same dimension through linear mapping and concatenate them to form a global vector;

[0031] The global vector is input into a preset fully connected network, and the output global fusion vector is combined with the confidence level output by the decision tree unit to perform weighted calculation to obtain a first confidence level. In a possible implementation method of the first aspect, the adaptive feedback of the fusion model through the evaluation index is specifically as follows:

[0032] Receive the analysis results of the fusion model, map them to the preset correction strategy template to identify the interference characteristics in the signal, generate dynamic parameter adjustment plans and output structured results;

[0033] The strategy template for each interference type is initially assigned an estimated initial success rate and mapped to the corresponding rating;

[0034] Through closed-loop feedback, the analysis results and correction effects are collected and comprehensively rated, and the initial success rate is adjusted using exponentially weighted average.

[0035] A second aspect of the present application provides a broadcast signal interference correction system, the system comprising: a signal processing module, a correction module, and a feedback module;

[0036] The signal processing module collects the broadcast signal in real time and converts it into a digital signal, performs preprocessing to obtain a first audio signal, performs feature extraction based on the first audio signal to obtain audio features, and performs corresponding preprocessing for different types of audio features;

[0037] The correction module integrates audio features to generate an audio feature matrix and inputs it into a preset fusion model to output an interference analysis, receives the interference analysis and combines it with the first confidence level to obtain a corresponding template, and fills in dynamic parameters to generate a correction strategy and credibility; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit;

[0038] The feedback module obtains the correction index after correction by the correction strategy, generates an evaluation index in combination with the interference analysis and the correction strategy, and performs adaptive feedback on the fusion model through the evaluation index.

[0039] A third aspect of the present application provides a storage medium storing computer-readable program code, which, when executed, implements the steps of the broadcast signal interference correction method described in any one of the embodiments of the present application.

[0040] A fourth aspect of the present application provides a terminal device, which includes: a terminal device including a processor and a memory, the memory storing a computer program, and the processor implementing the steps of the broadcast signal interference correction method described in any one of the embodiments of the present application when executing the computer program. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solution of the present application, the following is a brief introduction to the drawings required for use in the implementation. Obviously, the drawings described below are only some implementation methods of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0042] Figure 1 is a flowchart of a broadcast signal interference correction method provided by an embodiment of the present application;

[0043] Figure 2 Schematic diagram of the structure of the data input module provided in the embodiment of the present application;

[0044] Figure 3 It is a structural diagram of the fusion model analysis module provided in an embodiment of the present application;

[0045] Figure 4 Schematic diagram of the closed-loop feedback mechanism module provided in an embodiment of the present application; DETAILED DESCRIPTION

[0046] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0047] Example 1

[0048] See also Figure 1 As shown, a broadcast signal interference correction method provided by an embodiment of the present application includes steps S101-S103 specifically as follows:

[0049] Step S101: collecting a broadcast signal in real time and converting it into a digital signal, performing preprocessing to obtain a first audio signal, performing feature extraction based on the first audio signal to obtain audio features, and performing corresponding preprocessing for different types of audio features;

[0050] Furthermore, the feature extraction based on the first audio signal to obtain the audio feature is specifically:

[0051] Loading an audio file via an audio processing library to receive a first audio signal;

[0052] Extract audio features in parallel, the audio features including: unique identifier, basic audio file information, Mel-frequency cepstral coefficients, Mel-frequency spectrogram, MFCC coefficients, zero-crossing rate, and spectral centroid; wherein the basic audio file information includes: duration, sampling rate, number of channels, and format;

[0053] The audio signal is formatted as JSON for subsequent processing.

[0054] In this example, standardized MP3 data is used as input, and the audio processing library is used to load the file. Possible loading errors are captured, and if loading fails, a data re-retrieval mechanism is triggered to re-pull the corresponding audio file and accompanying feature data from the source. A consistency check is performed on the audio features, ensuring the correct association between the MP3 file and the JSON feature data by matching the unique identifiers of the MP3 file and the JSON feature data. The integrity of the JSON field is verified by checking for missing key fields, marking them as invalid data if they are indeed key fields.

[0055] It should be noted that the feature data in a standardized JSON format is read and spectral anomalies are identified through Mel-spectrogram data. When a specific frequency band has a sustained high-energy peak, the audio feature data is determined to be subject to narrowband interference; when the spectral features fluctuate significantly over time, the audio data is determined to be subject to dynamic interference. If the MP3 sampling rate is not the preset target value, resampling is performed to convert multi-channel to mono, eliminate interference differences between channels, and unify the audio format. The feature arrays of audio of different durations are zero-padded or truncated to unify the time frame length to ensure consistent dimensionality when input into the fusion model.

[0056] Furthermore, the corresponding extraction method for different types of audio features is specifically:

[0057] For non-interference audio features, lightweight extraction is performed;

[0058] For audio features affected by narrowband interference, frequency domain analysis is enhanced and outlier suppression is strengthened;

[0059] For audio features affected by dynamic interference, we strengthen the capture of temporal features and introduce dynamic filtering and attention mechanisms.

[0060] Specifically, for the lightweight extraction of non-interference audio features, basic wavelet noise reduction is applied to retain the original details of the signal and avoid artifacts introduced by excessive processing; for narrowband interference audio features, high-resolution spectrum is performed to more accurately capture the frequency peaks of narrowband interference, accurately locate the interference frequency, eliminate pollution to the features while retaining the original audio information; in a specific embodiment, for dynamic interference audio features, the changing trend between adjacent frames is captured by differential features, the weight of each frame is calculated by sub-attention, and the focus is automatically on the interfered frame and irrelevant background noise is suppressed. By capturing rapidly changing interference patterns, accurate modeling and compensation of time-division signals are achieved. Through feature extraction strategies for different interference types, the system achieves a balance between computational efficiency and anti-interference capabilities, and is particularly suitable for broadcast signal processing scenarios with limited resources and complex interference environments.

[0061] Furthermore, the preprocessing further includes:

[0062] Performing interpolation / dropping processing on invalid values ​​in audio features, wherein the invalid values ​​include missing values ​​and infinite values;

[0063] When the invalid value loss ratio exceeds the preset threshold, loss processing is performed;

[0064] When the loss ratio does not reach the preset threshold, linear interpolation is performed on invalid values;

[0065] Each audio feature is normalized using a linear function and correlation analysis is performed.

[0066] In a specific embodiment, when the missing ratio of a row / column exceeds a preset threshold, it is directly deleted. A preset function is used to retain rows or columns with at least thresh non-null values. When the non-null values ​​in a row or column are less than this threshold, it is discarded. By setting axis = 0 (row) or axis = 1 (column), flexible control can be performed on the row or column. Preferably, when there is a lot of missing data and the sample discard rate needs to be reduced, samples with sufficient information are retained while rows / columns with poor information are eliminated; linear interpolation is performed on the remaining missing values, and the missing values ​​are treated as linear interpolation between their adjacent non-null values ​​at equal intervals.

[0067] Each feature dimension is normalized by a linear function to ensure comparability of different dimensions, and the value of each feature is mapped to a specified minimum-maximum interval. The normalized feature data is further screened and shrunk, and correlation analysis is used to measure the redundancy of two features through the linear correlation coefficient between the features. When the absolute correlation between two features exceeds a preset threshold, the redundant information they carry is often very close, and one of them can be eliminated based on the actual situation. In a specific embodiment, the recursive feature elimination method is based on an external estimator (such as linear regression, decision tree) to score the importance of features, and recursively eliminates the least important features from the current feature set until the preset target number of features is reached.

[0068] For example, the Pearson correlation coefficient between all features is quickly calculated to obtain a symmetric matrix. The threshold is set to 0.8–0.95, which can be adjusted for different scenarios. For example, 0.8 can be used in general scenarios, and 0.95 is used for more stringent redundancy removal scenarios. In the upper triangle (or lower triangle) area of ​​the correlation matrix, find the feature pairs whose absolute correlation coefficient exceeds the threshold; if A and B are highly correlated, retain the one with a stronger correlation with the target variable and eliminate this redundant feature. After eliminating all highly correlated pairs, a candidate set without highly redundant features can be obtained, which helps to reduce multicollinearity, improve model interpretability and training efficiency.

[0069] This invention uses precise discarding and retention strategies to avoid contaminating model training with samples carrying a large number of missing values. By prioritizing the retention of audio clips with high time frame integrity for axis = 0 and the retention of key frame columns for axis = 1, sample utilization is improved compared to full discarding. Linear interpolation of adjacent frames is used to maintain the temporal continuity of spectral features, avoiding interference analysis breakpoints caused by direct deletion. Feature scaling is unified to prevent the model from favoring general features during gradient descent, enhancing the comparability of feature representation. Correlation analysis is used to identify highly correlated feature pairs, eliminating redundant features to reduce the number of model parameters and avoid overfitting. This improves the accuracy of subsequent interference analysis and system robustness, providing high-quality data support for subsequent fusion model decision-making.

[0070] Step S102: Integrate audio features to generate an audio feature matrix and input it into a preset fusion model to output an interference analysis, receive the interference analysis and combine it with the first confidence level to obtain a corresponding template, fill in dynamic parameters to generate a correction strategy and credibility; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit;

[0071] Furthermore, the audio features are integrated to generate an audio feature matrix and input into a preset fusion model, specifically:

[0072] Regularize the audio feature matrix through the decision tree unit to obtain the preliminary interference type and confidence level;

[0073] The decision tree unit adopts a preset CART decision tree, traverses the path from the root node to the leaf node, and reaches the leaf node to obtain the preliminary interference type;

[0074] The interference type label obtained by reaching the leaf node and the training sample of the leaf node are used as the confidence.

[0075] Specifically, based on correlation analysis and recursive feature elimination, the most discriminative feature vectors are screened out from numerous audio features, and the screened ones are arranged in time frame order to construct a two-dimensional feature matrix. In a specific embodiment, a preset CART (Classification and Regression Tree) decision tree is used, which is a binary tree structure. Each internal node corresponds to a feature splitting condition, and each leaf node corresponds to an interference type label. It supports Gini inequality or entropy splitting criteria. The splitting criteria is used to measure the "purity" or "uncertainty" of the node, and to guide how to select the optimal splitting features and thresholds, so as to better find the optimal split and reduce the risk of overfitting or underfitting; set decision tree related parameters, such as the maximum tree depth, the minimum number of leaf node samples, etc. The audio feature matrix data with labeled interference types is used for training. During the training process, the decision tree is split by continuously trying different features and thresholds, and the tree structure is optimized according to the selected splitting criteria.

[0076] The constructed audio feature matrix is ​​input row by row into the decision tree unit. For each row of data in the matrix (i.e., the feature vector for each time frame), starting from the root node of the decision tree, a judgment is made based on the feature splitting condition at the node. If the current feature value is less than or equal to the splitting threshold, the node enters the left child node; otherwise, the node enters the right child node, and the recursive traversal continues until a leaf node is reached. Upon reaching a leaf node, the label stored in the leaf node is the initially determined interference type, and the ratio of the number of samples of that interference type in the leaf node to the total number of samples in the leaf node is used as the confidence level.

[0077] Furthermore, the fusion model also includes:

[0078] Encode the modal data according to the multimodal fusion decision unit, align them to the same dimension through linear mapping and concatenate them to form a global vector;

[0079] The global vector is input into a preset fully connected network, and the output global fusion vector is weighted and combined with the confidence level output by the decision tree unit to obtain a first confidence level. In a specific embodiment, the input data is divided into three categories based on features: device status data, Mel-frequency cepstral coefficients, and Mel-spectrogram amplitude values. The device status data includes real-time operating parameters of the broadcast receiving device; the Mel-frequency cepstral coefficients are frequency domain features extracted from the audio signal; and the Mel-spectrogram amplitude values ​​reflect the spectral complement of the audio signal at the Mel-scale entropy, preserving time-frequency domain details. Preferably, a single-layer bidirectional LSTM network is independently constructed for each type of modal data, wherein each LSTM unit outputs an 8-dimensional hidden state to balance expressiveness and computational complexity. The single-layer network avoids overfitting while supporting single-sample real-time processing, making it suitable for streaming input of broadcast signals. Forward propagation is used to process from the beginning to the end of the sequence, capturing contextual information of past time steps; backward propagation is used to process from the end to the beginning of the sequence, capturing contextual information of future time steps, and ultimately outputting a hidden state tensor. The two directional hidden states of the bidirectional LSTM are concatenated in the feature dimension to obtain a 16-dimensional vector.

[0080] The features of different modalities are mapped to a unified semantic space through affine transformation to solve the problem of inconsistent feature dimensions, where the expression is Among them, h i w is the hidden state vector after RNN encodes the input sequence of the i-th modality (device state, MFCC, Mel spectrum, etc.); i Represents the linear mapping weight matrix, which is used to project into a unified k-dimensional space; b i represents a bias vector of length k; The length of the k-th aligned feature vector after affine transformation represents the expression of the modality in the unified semantic space. All modality alignment vectors are concatenated along the axis to obtain the global fusion feature. In this embodiment, the length of the global fusion feature is 3k, which is used as the input of the downstream decision maker.

[0081] It should be noted that the global fusion feature generated by the multimodal fusion decision unit is input into the preset fully connected network, and weighted calculation is performed with the confidence output by the decision tree module to obtain the final confidence; the final confidence calculation is specifically expressed as follows: p * =σ(α·p+(1-α)·g(β)), where g is the fusion network output and α∈[0,1] is the manual or learned weight; is a preset probability output function used to nonlinearly map any real number to (0, 1); p is the estimated probability that the returned sample belongs to category c, which is calculated based on the proportion of samples of this category in the leaf node; β is the global fusion feature, and g(β) is the probability or score obtained by calculating the global fusion feature vector β as input through the fully connected network; α∈[0,1] is used to balance the relative importance of the outputs of the two sub-models. When α→1, the decision tree is more trusted; when α→0, the MMFM network is more trusted; p * The value obtained after mapping is used to determine the decision threshold. The fused confidence and the original decision tree label are combined and mapped to the structured result.

[0082] The fusion model analyzes the preprocessed MP3 feature data to identify interference features in the signal, such as multipath interference, noise interference, frequency offset, dynamic interference, etc. After receiving the model analysis results (such as "multipath interference" and a confidence level of 0.92), it will automatically map them to the corresponding correction template to generate directly implementable frequency, power, filtering or dynamic parameter adjustment schemes and output structured results.

[0083] This application uses bidirectional LSTM to simultaneously process past and future contexts to improve the ability to track dynamic interference; based on the collaborative mechanism of the fusion model, decision-making complementary logic is performed. When the decision tree unit outputs a high-confidence result, a regularized judgment is adopted. When it is placed at a low confidence level, multimodal features are integrated through a multimodal fusion decision unit. By capturing nonlinear relationships, the recognition accuracy of complex interference is improved, providing more robust technical support for real-time interference analysis of broadcast signals.

[0084] Step S103: obtaining a correction index after correction by the correction strategy, generating an evaluation index in combination with the interference analysis and the correction strategy, and performing adaptive feedback on the fusion model through the evaluation index.

[0085] Furthermore, the adaptive feedback of the fusion model through the evaluation index is specifically as follows:

[0086] Receive the analysis results of the fusion model, map them to the preset correction strategy template to identify the interference characteristics in the signal, generate dynamic parameter adjustment plans and output structured results;

[0087] The strategy template for each interference type is initially assigned an estimated initial success rate and mapped to the corresponding rating;

[0088] Through closed-loop feedback, the analysis results and correction effects are collected and comprehensively rated, and the initial success rate is adjusted using exponentially weighted average.

[0089] It should be noted that correction strategy templates are pre-set for each interference type based on historical data. These templates include: correction operation category (such as frequency adjustment and power control), default parameters (configurable operating parameters such as frequency range and gain value), initial success rate (based on historical data and expert experience), and discrete rating (success rate is mapped to four levels: high / medium / low / invalid). The database is loaded through JSON parsing to maintain memory mapping and support dynamic addition, deletion, modification, and query.

[0090] Receive the fusion model output and retrieve the corresponding strategy from the template library, and dynamically adjust the default parameters based on real-time measurement data; obtain key indicators of the recognition process and after strategy execution to evaluate the accuracy of the recognition results; after performing filtering or power control, measure key indicators before and after, such as signal-to-noise ratio, bit error rate, etc. For digital communications, measure the impact of the correction strategy on the bit error rate; for real-time systems, record the data transmission performance before and after the strategy is applied.

[0091] The template credibility is dynamically adjusted through exponential weighted average, and discrete ratings are remapped according to the updated credibility. If the recognition rating is too low for multiple consecutive times, it will trigger decision tree retraining or screening strategy adjustment; templates with continuously declining credibility will be automatically downgraded to backup strategies and trigger manual review. When the template credibility falls below the preset threshold, the strategy will be automatically marked and an alarm will be pushed. Manual adjustments can be made to template parameters or reset credibility through the visual panel.

[0092] The beneficial effects of the present invention include: reducing feature loss through feature classification loading, eliminating redundant features through RFE and correlation analysis, and improving feature recognition accuracy; improving the recognition accuracy of complex interference through parallel processing of decision tree units and multimodal fusion units in the fusion model; and improving the credibility of policy templates by combining exponentially weighted average dynamic optimization and driving continuous evolution through closed-loop feedback. Through an adaptive feedback mechanism, a continuously evolving intelligent anti-interference system is constructed, improving the stability and signal quality of the broadcast system and meeting the requirements of modern broadcast systems for stable signal output.

[0093] Example 2

[0094] Please refer to Figure 2-4 , a broadcast signal interference correction system provided by an embodiment of the present application includes: a data input module, a fusion model analysis module and a closed-loop feedback mechanism module;

[0095] The data input module collects broadcast signals in real time and converts them into digital signals, performs preprocessing to obtain a first audio signal, performs feature extraction based on the first audio signal to obtain audio features, and performs corresponding preprocessing for different types of audio features;

[0096] The fusion model analysis module integrates audio features to generate an audio feature matrix and inputs the matrix into a preset fusion model to output an interference analysis, receives the interference analysis, combines it with the first confidence level to obtain a corresponding template, and fills in dynamic parameters to generate a correction strategy and a confidence level; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit;

[0097] The closed-loop feedback mechanism module obtains the correction index after correction by the correction strategy, generates an evaluation index in combination with the interference analysis and the correction strategy, and performs adaptive feedback on the fusion model through the evaluation index.

[0098] The broadcast signal interference correction system described above can implement the broadcast signal interference correction method of the method embodiment described above. The optional options in the method embodiment described above also apply to this embodiment and will not be described in detail here. The remaining contents of the embodiments of this application can refer to the contents of the method embodiment described above. In certain preferred embodiments, no further description will be given.

[0099] Example 3

[0100] Accordingly, the present application also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the broadcast signal interference correction method described in any one of the above embodiments.

[0101] Exemplarily, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program in the terminal device.

[0102] The terminal device may be a computing device such as a desktop computer, a notebook computer, a PDA, a cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0103] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0104] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0105] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by the processor, it can implement the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device that can carry the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc.

[0106] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. A broadcast signal interference correction method, characterized in that: include: Collecting broadcast signals in real time and converting them into digital signals, performing preprocessing to obtain a first audio signal, performing feature extraction based on the first audio signal to obtain audio features, and performing corresponding preprocessing on different types of audio features; Integrate audio features to generate an audio feature matrix and input it into a preset fusion model to output an interference analysis, receive the interference analysis and combine it with the first confidence level to obtain a corresponding template, fill in dynamic parameters to generate a correction strategy and credibility; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit; The correction index after correction by the correction strategy is obtained, and the evaluation index is generated by combining the interference analysis and the correction strategy. The fusion model is adaptively fed back through the evaluation index.

2. The broadcast signal interference correction method according to claim 1, wherein: The feature extraction based on the first audio signal to obtain audio features is specifically as follows: Loading an audio file via an audio processing library to receive a first audio signal; Extract audio features in parallel, the audio features including: unique identifier, basic audio file information, Mel-frequency cepstral coefficients, Mel-frequency spectrogram, MFCC coefficients, zero-crossing rate, and spectral centroid; wherein the basic audio file information includes: duration, sampling rate, number of channels, and format; The audio signal is formatted as JSON for subsequent processing.

3. The broadcast signal interference correction method according to claim 2, wherein: The corresponding extraction method for different types of audio features is specifically as follows: For non-interference audio features, lightweight extraction is performed; For audio features affected by narrowband interference, frequency domain analysis is enhanced and outlier suppression is strengthened; For audio features affected by dynamic interference, we strengthen the capture of temporal features and introduce dynamic filtering and attention mechanisms.

4. The broadcast signal interference correction method according to claim 3, wherein: The preprocessing further includes: Performing interpolation / dropping processing on invalid values ​​in audio features, wherein the invalid values ​​include missing values ​​and infinite values; When the invalid value loss ratio exceeds the preset threshold, loss processing is performed; When the loss ratio does not reach the preset threshold, linear interpolation is performed on invalid values; Each audio feature is normalized using a linear function and correlation analysis is performed.

5. The broadcast signal interference correction method according to claim 1, wherein: The audio feature matrix is ​​generated by integrating the audio features and inputting the matrix into the preset fusion model, specifically: Regularize the audio feature matrix through the decision tree unit to obtain the preliminary interference type and confidence level; The decision tree unit adopts a preset CART decision tree, traverses the path from the root node to the leaf node, and reaches the leaf node to obtain the preliminary interference type; The interference type label obtained by reaching the leaf node and the training sample of the leaf node are used as the confidence.

6. The broadcast signal interference correction method according to claim 5, characterized in that: The fusion model also includes: Encode the modal data according to the multimodal fusion decision unit, align them to the same dimension through linear mapping and concatenate them to form a global vector; The global vector is input into a preset fully connected network, and the output global fusion vector is combined with the confidence output by the decision tree unit for weighted calculation to obtain a first confidence.

7. The broadcast signal interference correction method according to claim 1, characterized in that: The adaptive feedback of the fusion model through the evaluation indicators is specifically as follows: Receive the analysis results of the fusion model, map them to the preset correction strategy template to identify the interference characteristics in the signal, generate dynamic parameter adjustment plans and output structured results; The strategy template for each interference type is initially assigned an estimated initial success rate and mapped to the corresponding rating; Through closed-loop feedback, the analysis results and correction effects are collected and comprehensively rated, and the initial success rate is adjusted using exponentially weighted average.

8. A broadcast signal interference correction system, characterized in that: include: Data input module, fusion model analysis module and closed-loop feedback mechanism module; The data input module collects broadcast signals in real time and converts them into digital signals, performs preprocessing to obtain a first audio signal, performs feature extraction based on the first audio signal to obtain audio features, and performs corresponding preprocessing for different types of audio features; The fusion model analysis module integrates audio features to generate an audio feature matrix and inputs the matrix into a preset fusion model to output an interference analysis, receives the interference analysis, combines it with the first confidence level to obtain a corresponding template, and fills in dynamic parameters to generate a correction strategy and a confidence level; wherein the fusion model includes a decision tree unit and a multimodal fusion decision unit; The closed-loop feedback mechanism module obtains the correction index after correction by the correction strategy, generates an evaluation index in combination with the interference analysis and the correction strategy, and performs adaptive feedback on the fusion model through the evaluation index.

9. A storage medium, characterized in that: The storage medium stores computer-readable program codes, and when the computer-readable program codes are executed, the steps of a broadcast signal interference correction method according to any one of claims 1 to 7 are implemented.

10. A terminal device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the broadcast signal interference correction method according to any one of claims 1 to 7 are implemented.

Images