118 results about "Wiener filter" patented technology

The invention relates to a cloud edge cooperative computing framework and processing method for multi-modal data stream fusion processing, and the method comprises the following steps: S1, carrying out the noise suppression based on an original data stream collected by an edge computing node through employing an improved Wiener filtering algorithm, achieving the signal denoising through the adaptive threshold wavelet transformation, and obtaining a cloud edge data stream; and a timestamp alignment technology is utilized to solve the problem of time delay difference of multi-modal data, and a space-time alignment purified data stream is generated. Through combination of the improved Wiener filtering algorithm and the adaptive threshold wavelet transform, the noise suppression efficiency of the original data stream is significantly improved, the timestamp alignment technology effectively solves the time delay difference of the multi-modal data, the generation of the space-time alignment purified data stream ensures that the subsequent processing has a unified time sequence benchmark, and the efficiency of noise suppression of the original data stream is improved. The space-time attention fusion network adopts a collaborative architecture effect of a bidirectional gating circulation unit and a lightweight 3D convolutional network.

The invention discloses a target identification method for multi-sensor data fusion, particularly relates to the technical field of data fusion, and comprises a dynamic weight fusion module, a double-branch neural network and an abnormal sensing compensation system. Data are synchronously collected through a visible light camera, a thermal infrared imager and a millimeter wave radar, and a standardized feature map is generated through sensor specificity preprocessing; the dynamic weight fusion module generates an adaptive weight matrix based on the real-time confidence score and the environmental parameters, emphasizes infrared data when illumination suddenly changes, and improves intelligent distribution of radar weights in a rain and fog environment; the combined feature map after weight fusion is input into a double-branch neural network, a channel self-calibration branch suppresses interference noise, and a target category and coordinates are output after residual connection optimization; when the recognition confidence is insufficient, the abnormal compensation system starts visible light Wiener filtering restoration, generative adversarial network infrared compensation and radar multi-frame accumulation algorithms, and the system reliability is ensured when a single sensor fails.

Disclosed are a channel estimation method, apparatus and device, and a storage medium, which relate to the technical field of communications, comprising: acquiring a channel parameter and a channel coefficient matrix of a pilot symbol based on received signal data, the channel parameter being a channel model parameter of non-ground network communication; determining a first wiener filter tap coefficient from a preset typical wiener filter tap coefficient set according to the position combination of the channel parameter and the pilot symbol; and determining a time domain channel estimation result of the received signal data according to the first wiener filter tap coefficient and a channel coefficient matrix of a pilot symbol. According to the technical scheme, the channel estimation process in a non-ground network communication scene is simplified, and the real-time performance of time domain channel estimation is improved on the premise of ensuring the accuracy of a channel estimation result.

The invention relates to the technical field of image processing, and discloses an optical image noise reduction method and system of a composite imaging assembly. The method comprises the following steps: carrying out geometric correction on visible light, infrared and laser sensor data through a multi-sensor time synchronization mechanism to obtain a noise coupling matrix; generating annular and checkerboard type mask modes based on the noise coupling matrix; performing cross-band feature interaction through the conditional mask convolution block to obtain a multi-scale dense connection feature map; a double-blind-spot self-supervised learning algorithm is adopted to optimize cross-band coordination loss, and a noise reduction prediction result is obtained; and finally, edge protection processing is carried out through a block-based Wiener filter, and a final noise reduction image is obtained. The problem of noise reduction failure caused by multi-sensor noise coupling and pixel strong correlation in the composite imaging assembly is solved, and the imaging quality and target recognition precision of the composite imaging assembly in a complex environment are improved.

The invention relates to the technical field of voice communication, in particular to a voice signal processing method and device, equipment, a storage medium and a computer program product. The method comprises the following steps: performing real-time noise classification on a received original voice signal based on a preset deep neural network noise recognition model; according to the noise classification result, selecting a corresponding combined noise reduction model to perform multi-stage noise reduction processing on the original voice signal to obtain a noise-reduced voice signal, the combined noise reduction model comprising one or more of a deep neural network-based noise reduction module, a Wiener filtering module and a spectral subtraction module; and on the basis of time domain feature extraction and frequency domain feature extraction, voice feature enhancement processing is performed on the noise-reduced voice signal to obtain the target voice signal, so that the transmission quality of the voice signal is improved.

The invention provides a filtering method and device for channel estimation, electronic equipment and a storage medium. The method comprises the following steps: acquiring initial channel estimation of a reference signal, and determining optimal noise power in a current target channel scene based on the initial channel estimation; performing first-stage Wiener filtering on the initial channel estimation to obtain first filtering channel estimation; and performing second-stage Wiener filtering on the first filtering channel estimation based on the optimal noise power to obtain second filtering channel estimation. Therefore, according to the scheme, the optimal noise power in the target channel scene is acquired in advance, so that the second-stage Wiener filtering can be immediately carried out after the first-stage Wiener filtering is completed, the process of carrying out noise estimation on the reference signal after the first-stage filtering can be omitted, the processing waiting time is reduced, and the processing efficiency is improved. And the processing time delay is reduced, so that the two-stage filter circuit can be designed in a tight coupling manner to enhance the stability of the filter performance.

The invention particularly relates to a voice noise reduction method suitable for different noise environments, and relates to the technical field of voice signal processing. Analyzing a noise environment and characteristics; adaptive selection of a noise reduction algorithm; adaptive noise reduction processing is executed; and noise reduction effect evaluation and iterative optimization are carried out. According to the invention, refined noise classification is matched with the algorithm, so that the limitation of a traditional single algorithm in a complex environment is solved; the steady-state noise is dynamically counteracted by adopting an adaptive filter, the frequency domain gain of the unsteady-state noise is adjusted in real time through Wiener filtering, and the pulse noise is subjected to dynamic threshold processing by combining median filtering and wavelet transform; especially, the design of cooperation factors in wavelet transform can accurately distinguish signal details and noise: intensively suppress pulse noise, loosely retain details such as voices and consonants, and realize dynamic balance of noise reduction intensity and signal distortion.

The invention discloses a micro-magnitude embedded system speech enhancement method and device based on a neural network, electronic equipment and a storage medium. The method comprises the following steps: acquiring power spectrum data of voice data to be processed, and calculating a filter bank coefficient for a preset filter bank; the power spectrum data and the filter bank coefficient are input into a pre-training neural network, an initial Wiener filter coefficient is obtained, and the pre-training neural network is obtained through training in a ratio mask mode and a signal approximation mode; performing interpolation processing on the initial Wiener filter coefficient to obtain a target Wiener filter coefficient with the same number as the frequency points of the to-be-processed voice data; and processing the to-be-processed voice data through the target Wiener filter coefficient to obtain voice data after noise reduction processing. According to the technical scheme, the noise reduction effect is improved while the noise reduction calculation amount is reduced.

The invention provides a digital-analog dual-drive combined Wiener filtering unmanned aerial vehicle observation ground target coordinate resolving method, and relates to the field of computer vision and artificial intelligence, and the technical key points are that the method comprises the following steps: constructing a flight parameter-ground coordinate space mapping network, and obtaining a coordinate space prediction value; establishing a three-dimensional geometric physical model, and deducing geometric predicted values of the coordinates according to the aerial coordinates of the unmanned aerial vehicle; establishing a target coordinate track time mapping network, and obtaining a target time prediction value; constructing a Wiener filter, solving a filter coefficient, forming multi-attribute prediction complementation by using linear combination, and obtaining a final coordinate estimation value; according to the method, the measurement error iterative coupling of the sensor is effectively removed by mining the internal law between the observation data and the prediction truth value of the space, geometry and time, combining the parameters of the Wiener adaptive filter and fusing the three prediction data values of physical and data driving, and the ground coordinate calculation accuracy and robustness of the unmanned aerial vehicle observation image target are improved.

The invention relates to the field of voice signal processing and voice recognition, and discloses a voice recognition enhancement method and system based on harmonic model fundamental frequency optimization and RNN noise suppression, and the method comprises the steps: collecting a voice signal in a noise environment in real time, carrying out the framing and windowing of the voice signal, and generating multi-dimensional time-frequency data; estimating the fundamental frequency of each frame of voice through a cepstrum analysis method, and screening effective fundamental frequency frames according to a confidence coefficient threshold to form a fundamental frequency characteristic matrix; dynamically adjusting the noise power spectrum of the Wiener filter under the drive of the fundamental frequency characteristic matrix, and carrying out dislocation fusion on the filtering output and the original spectrum to form an enhanced spectrum first draft; inputting the enhanced spectrum first draft into a recurrent neural network in a framing manner, predicting the gain of each frequency band, calculating an inhibition factor, and generating a multi-frame continuous enhanced spectrum sequence; multiple frames of continuous enhanced spectrum sequences are synthesized into voice signals through inverse short-time Fourier transform, an end-to-end enhanced recognition process is formed, and the method has the advantage of improving accuracy.

The invention discloses a highway bridge multi-mode safety monitoring method and system. The method comprises the steps that ultrasonic information, voiceprint information and image information of a highway bridge are collected; performing graying and normalization processing on the image information by adopting an improved Laplace transformation algorithm, and automatically adjusting the threshold value of the ultrasonic information according to the energy distribution of the ultrasonic information by utilizing a dynamic threshold value filtering method; carrying out noise reduction processing on the voiceprint information by adopting a Wiener filtering algorithm; a Modern-TCN model is adopted to train the voiceprint information obtained after noise reduction is carried out on the voiceprint information; the image information and the ultrasonic information are processed by adopting a Modern-TCN model improved by a Lite-EMA attention mechanism; a progressive multi-modal fusion method optimized by a PID search algorithm is adopted, the Modern-TCN model improved by the Lite-EMA attention mechanism and the Modern-TCN model are fused, and the fused model processing result is combined with an experience knowledge base to obtain a highway bridge safety monitoring result; the overall performance and reliability of highway bridge safety monitoring can be improved.

The invention discloses a micro-lens array focal length detection method and detection equipment based on an image moving fitting method, and the method comprises the steps: setting a camera, adjusting the camera until the camera is focused on a micro-lens array, adjusting the optical axis distance between the camera and the array in a floating manner, and collecting star point light spot images and Perot plate line-to-plate images at a plurality of adjusted positions; according to the light spot sub-images corresponding to the same sub-lens in the plurality of light spot images, focal plane coordinates are determined, and corresponding PSF reconstruction is carried out; selecting a pair of light plate images and line-to-plate images, and performing recovery processing on sub-images corresponding to the sub-lenses in the images by using a Wiener filtering algorithm; performing convolution operation on the restored image and the reconstructed PSF function to obtain a reconstructed line-to-board image corresponding to the focal plane coordinates; and the focal length of the sub-lens is calculated according to the focal length of the collimator, the actual line distance of the Perot plate and the line distance determined by the reconstruction line to the plate image. Array images near the focus are collected in an image moving mode, and the focal lengths of the multiple sub-lenses are detected in parallel.

The present application relates to the technical field of image processing, in particular to an insulator image enhancement method based on infrared image denoising of a UAV. The original infrared image collected by the UAV is preprocessed, and the preprocessed image is decomposed in multiple scales by wavelet transform to obtain high-frequency detail coefficients and low-frequency approximation coefficients; the high-frequency detail coefficients are processed by adaptive thresholding, while the low-frequency approximation coefficients are subjected to Wiener filtering, and after wavelet reconstruction and guided filtering, a denoised image is obtained; based on the denoised image, an edge detection operator is used for preliminary edge extraction, edge information is obtained by combining non-maximum suppression and double thresholding, and the edge gap is connected by using morphological edge reconstruction technology; through adaptive contrast enhancement based on local temperature distribution and image fusion processing, the final image with enhanced features is generated. The present application can realize efficient denoising and accurate feature enhancement of the insulator infrared image collected by the UAV.

The application discloses a TDI-CCD image deartifact method and device, equipment, medium and product, and relates to the technical field of image processing. The method comprises the following steps: constructing a deep deblurring network; the deep deblurring network comprises a Wiener filter module and a generative adversarial network, and the Wiener filter module is arranged at the input end of a generator of the generative adversarial network; the Wiener filter module is used for filtering a frequency domain image; the deep deblurring network is trained, and the trained deep deblurring network is used as a deartifact model; during the training of the deep deblurring network, a point spread function and a regularization parameter of the Wiener filter module are used as learnable network parameters; and an image with row smear obtained from a TDI-CCD camera is processed by using the deartifact model to obtain a deartifact image. The application can effectively remove image artifacts and improve the definition of a TDI-CCD image.

The invention relates to the field of trachinotus ovatus swarm sound production identification, and discloses a method for identifying sound production of a trachinotus ovatus swarm in a net cage based on dual-channel time-varying Wiener filtering. According to the method, modeling and target signal counteracting are respectively carried out according to the conditions of the left sound channel and the right sound channel of the input signal so as to analyze the noise signal-to-noise ratio characteristics, enhancement of target biological sound and suppression of noise signals are realized in combination with a Wiener filtering method, trachinotus ovatus sound production can be identified in a complex noise environment, and the method has the target positioning capability. According to the method, by introducing a dual-channel signal equalization and interference compensation mechanism and fusing posterior and prior signal-to-noise ratio estimation, the sound production of the trachinotus ovatus swarm can be effectively identified under the complex noise condition without training data, the problems that the performance is poor and a large amount of training data is needed when non-stationary noise is separated by a traditional noise separation method are solved, and the method is suitable for popularization and application. Technical prerequisites are provided for net cage acoustic monitoring work, and possibility is created for rapid, efficient and accurate sound production analysis of subsequent net cage fish schools.

The invention relates to a power transmission network security management system, and the system comprises a final-stage conversion mechanism which is used for successively executing Wiener filtering processing and median filtering processing on a secondary conversion picture; and the model application device is used for intelligently identifying whether the field spacing distance between the grounding piece and the operation electrician is smaller than or equal to the safe spacing distance or not by adopting an AI identification model. The power transmission network security management system is intelligent in design and simple to operate and control. The image content capturing action can be executed on the replacement scene for replacing the strain insulator string so as to obtain and output the corresponding replacement scene picture, and the AI identification model is adopted to intelligently identify whether the field spacing distance between the grounding piece and the operation electrician is smaller than or equal to the safe spacing distance based on various visual information in the replacement scene picture; therefore, when a strain insulator string is replaced in a power transmission network, a reliable and safe distance between an operation electrician and a grounding body is maintained.

The invention provides a multi-output channel Wiener filtering implementation method with low operand. The method comprises the following steps: (1) respectively calculating a reference signal cross correlation coefficient and a secondary path cross correlation coefficient; (2) the central part of the reference signal cross correlation coefficient is subjected to convolution of the secondary path cross correlation coefficient, and an obtained result is the central part of the filtered reference signal cross correlation coefficient; (3) performing secondary path filtering on the error signal, performing summation, and performing convolution with the reference signal to calculate a cross correlation coefficient between the filtered reference signal and the error signal; and (4) solving the coefficient of the filter by using a Levinson-Durbin algorithm. According to the method, solving of the standard Wiener filter is achieved with low calculation amount, the noise reduction amount is high, stability is good, the structure is simple, and the noise reduction amount and the convergence speed of an active noise reduction system can be improved.

The present invention proposes an image restoration method for optical synthetic aperture imaging (OSAI), and the implementation steps are as follows: (1) Filter the optical synthetic aperture imaging (OSAI) image based on the improved Wiener filtering algorithm; (2) Obtain the optimal cut-off frequency D best ; (3) Divide the spectrum diagram of the preliminarily restored image into high and low frequencies; (4) Perform high-frequency emphasis and bilateral filtering on the high and low frequency spectrum diagrams of the preliminarily restored image; (5) Obtain the restoration result of the OSAI image. By calculating the comprehensive evaluation index Q of the preliminarily restored image after processing the degraded image and selecting the optimal solution of the parameter K to replace the parameter K in the Wiener filtering algorithm, the present invention can more accurately select the appropriate K value of the Wiener filtering algorithm, so that the preliminary restoration effect reaches the optimal effect of the Wiener filtering algorithm. Then, high-frequency emphasis and bilateral filtering are performed on the high and low frequency spectrum diagrams of the preliminarily restored image, effectively improving the resolution of the restored image and meeting the needs of space observation.

The invention relates to the technical field of satellite navigation, and particularly discloses an anti-interference and anti-cheating method and equipment for civil aviation satellite navigation, and the method comprises the steps: constructing a multi-beam pointing and multi-stage Wiener filtering anti-interference method, solving an optimal anti-interference weight vector under the significance of a maximum output signal to interference plus noise ratio, and obtaining an optimal anti-interference weight vector under the significance of a maximum output signal to interference plus noise ratio; the maximum interference suppression degree and the minimum satellite signal receiving signal-to-noise ratio loss are considered; detecting whether deception exists in the captured signal after interference resistance through a correlation peak ternary detection and attitude position auxiliary detection method; if deception exists, deception is inhibited through a deception regeneration cancellation method, and satellite signal purification is achieved. Aiming at the anti-interference and anti-cheating requirements of civil aviation, a multi-beam pointing and multi-stage Wiener filtering anti-interference method, a correlation peak ternary detection method, an attitude position auxiliary detection method and a cheating regeneration fire-fighting cheating method are constructed; the method has the advantages of being universal for various kinds of interference, high in anti-interference capability, adaptive to forwarding / generation type deception, high in deception prevention capability, high in engineering practicability and the like.

The invention provides an image processing method based on the combination of Wiener filtering and a Richardson dewson algorithm. The image processing method comprises the following steps: acquiring a to-be-processed blurred image through an image acquisition module; fourier transform is carried out on the to-be-processed blurred images through the blurring kernel estimation module, the spectral characteristics of the to-be-processed blurred images are analyzed, and the to-be-processed blurred images are classified into first-class images or second-class images according to the spectral characteristics; pre-processing the first type of images or the second type of images through a Wiener filtering algorithm and a Richardson-Mexi algorithm to obtain pre-processed images; an edge gradient threshold of a preprocessed image is calculated through a Sobel operator, a fusion weight is given to an intermediate image generated in the process of preprocessing the blurred image to be processed according to the gradient threshold, a final fusion image is output through fusion, and the blurred image to be processed is obtained through blurring type adaptive judgment, staged algorithm adaptation and regional weight fusion. The method achieves the efficient processing of mixed blurring and multi-type noise, and gives consideration to the deblurring effect, noise suppression and edge detail reservation.

The present application discloses a method and an electronic device for reducing echo residue. The method for reducing echo residue can be applied to the electronic device, and comprises: performing echo cancellation on a voice input signal according to an echo reference signal to obtain an echo cancellation signal; converting the echo reference signal into a reference spectrum signal of each frame by fast Fourier transform; converting the echo cancellation signal into a voice spectrum signal of each frame by fast Fourier transform; obtaining an a priori signal-to-noise ratio of a current frame by using the reference spectrum signal of the current frame and the voice spectrum signal of the current frame according to an additive noise principle; filtering the voice spectrum signal of the current frame by a Wiener filter coefficient of the current frame determined by the a priori signal-to-noise ratio of the current frame to obtain a target spectrum signal of each frame; and converting the target spectrum signal of each frame by inverse fast Fourier transform to obtain a target voice signal. Therefore, the method for reducing echo residue can accurately filter out residual echo.

The application relates to a cloud-edge collaborative computing framework and processing method for multi-modal data stream fusion processing, which comprises the following steps: S1: based on the original data stream collected by the edge computing node, an improved Wiener filtering algorithm is used for noise suppression, signal denoising is realized through adaptive threshold wavelet transform, and a time stamp alignment technology is used to solve the time delay difference problem of multi-modal data, and a space-time aligned and purified data stream is generated. The application has the effects that through the combination of the improved Wiener filtering algorithm and the adaptive threshold wavelet transform, the noise suppression efficiency of the original data stream is significantly improved, the time stamp alignment technology effectively solves the time delay difference of multi-modal data, the generation of the space-time aligned and purified data stream ensures that the subsequent processing has a unified time sequence reference, and the space-time attention fusion network adopts the collaborative architecture of a bidirectional gate recurrent unit and a lightweight 3D convolution network.

The invention provides a forest fire early recognition method, medium and system based on multi-modal monitoring data, and belongs to the technical field of forest fire early recognition. Establishing a macroscopic abnormal feature recognition system, calculating a reference threshold value, generating an abnormal feature distribution matrix through spatial clustering analysis, and performing microscopic feature deep mining on an abnormal region boundary and an intensity gradient change region to establish a candidate feature set; a multi-scale feature fusion model is adopted to realize deep fusion of different modal and scale features through a hierarchical convolution structure and a multi-scale pyramid attention mechanism, and the fused features are input into a fire behavior identification and discrimination model to carry out risk grade evaluation and establish a response priority order. The technical problem that the identification precision of multi-mode forest fire monitoring data is insufficient under the condition of weak signals is solved.

The application discloses a kind of deep learning end-to-end lens design methods based on fuzzy kernel correction wiener filtering, in training stage, the nominal fuzzy kernel of each patch area of clear image can be generated by differentiable ray tracing, add tolerance to generate tolerance fuzzy kernel and blur clear image;The simulation fuzzy image and nominal fuzzy kernel are input into correction network to obtain correction fuzzy kernel, wiener filtering is carried out based on the kernel to simulation fuzzy image, and the restoration image is obtained by filtering result input image restoration network, parameter updating is carried out by loss function, gradient back propagation, and iteration is carried out until it meets the exit requirement;In experimental stage, according to the training parameter, manufacture lens and shoot real fuzzy image, use the nominal fuzzy kernel and network parameter after training to complete similar restoration steps with training stage, and generate final restoration image.The application does not need additional calibration and fine adjustment, can effectively resist the influence of various tolerances, along with lens design completion, can significantly reduce lens production cycle and cost.

The application discloses a low-complexity cascaded generalized sidelobe canceling beam forming method and device, and through constructing a beam former of a cascaded generalized sidelobe canceling structure, a high-dimension generalized sidelobe canceler is decomposed into two low-dimension generalized sidelobe canceling units in front and back cascades, so that the operation complexity can be remarkably reduced and the robustness is improved; through using a front and back level feedback iteration processing mode based on frames, the real-time performance of the algorithm is improved, and the application value is higher. The application further utilizes the idea of multi-channel Wiener filtering to construct an interference noise canceling module, directly updates the filtering coefficient between frames, does not need to additionally perform noise estimation, and has simple realization and high robustness, and can balance the interference noise suppression and target speech distortion. Through a multi-branch fusion and interleaving strategy, the application can improve the degree of freedom of beam weight coefficient searching, is more favorable to the algorithm to obtain an optimal solution, and improves the performance of target enhancement and interference noise suppression.

The invention relates to the technical field of intelligent household appliances, in particular to a dishwasher cleanliness detection method adopting T-Faster R-CNN, comprising the following steps: constructing a tableware picture library; de-noising the images of the tableware picture library by using a nonlinear bilateral Wiener filtering method; dividing the denoised tableware picture library into a tableware training set and a tableware testing set; marking the images in the tableware training set; constructing a T-Faster R-CNN (Convolutional Neural Network) model, and performing feature extraction by adopting a T-SR network; the tableware training set is sent to a T-Faster R-CNN model for training, and a trained T-Faster R-CNN model is obtained; a G-Focal cleaning loss function is calculated, and a final T-Faster R-CNN (Convolutional Neural Network) model is obtained; and testing the final T-Faster R-CNN by using the tableware test set, and outputting a detection result to obtain the cleanliness of tableware cleaned by the dish washing machine. By fully utilizing the feature representation capability of the neural network, the dirt in the dish-washing machine can be accurately detected, the dependence on human intervention is reduced, and the robustness and adaptability of the algorithm are improved.

The invention provides a night lighting scheduling method for a power plant. According to the method, an existing security camera is combined with an improved YOLOv4-tiny algorithm to realize rapid and accurate detection of personnel, and lighting scheduling of a specific area is triggered. An infrared induction device array with a Fresnel lens is deployed in key areas such as a power distribution cabinet, signals are processed through wavelet transformation, and after cross validation is conducted on the signals and camera detection results, a full-power brightening mode is started through ZigBee transmission signals. A fuzzy Petri network control model is introduced, and an optimal illumination strategy is generated according to the position, the track and the illumination intensity of the personnel. An improved MFCC algorithm is combined with an HMM to recognize a voice instruction, when a person enters a forbidden zone, warning voice is played, the definition is guaranteed through multi-channel Wiener filtering, meanwhile, a differential privacy protection LBP algorithm is used for analyzing facial expressions, and a second-level alarm is triggered when the facial expressions are abnormal. And a self-adaptive model based on deep reinforcement learning is constructed, scheduling is optimized through a PPO algorithm and an epsilon-greedy exploration mechanism, and power consumption minimization is realized.

The application discloses a method for improving voice conference call quality, relates to the technical field of computer voice processing, and comprises the following steps: using an Escape-TDNN model, adopting diversity samples and double noise reduction technology; using a multi-index fusion algorithm, providing personalized audio services, using adaptive beam forming technology and Wiener filtering algorithm, adopting a human voice separation model, combining with a voiceprint recognition result for post-processing optimization, using natural language processing technology, combining with migration learning technology, using voice synthesis and post-processing technology, fusing microphone array space information and video visual features, combining with environmental prior information, introducing an attention mechanism, and supporting online update of the human voice separation model. The application separates background sound by using an algorithm, isolates noise, makes voice clear, introduces text-to-speech, copes with situations where voice cannot be emitted or a microphone fails, accurately identifies and separates human voice by means of voiceprint technology, improves recognition and purity, and enhances audio effect.

This invention discloses a method and system for restoring defocused images based on estimating the point spread function from a spot image. The method includes: 1) reading the original spot image and the defocused blurred image; 2) cropping the area where the spot is located; 3) projecting the cropped spot image into grayscale along the vertical axis to obtain the original line spread function curve; 4) performing Gaussian fitting on the original line spread function curve to obtain a smooth line spread function curve; and then calculating the standard deviation of the Gaussian function; 5) calculating the blur coefficient of the Gaussian defocus model using the standard deviation; 6) substituting the blur coefficient into the Gaussian defocus model to obtain the point spread function; 7) performing Wiener filtering on the defocused blurred image in the frequency domain to obtain the initial restored image in the frequency domain; 8) performing a second Wiener filtering on the initial restored image to obtain the final restored image in the frequency domain; and 9) performing an inverse Fourier transform to obtain the restored image. This invention effectively reduces the blurriness of defocused images and solves the problem of computational complexity.

This invention discloses a modulation domain speech enhancement method based on group sparse representation, comprising the following stages: a training phase: calculating sub-band spectra of a mixed training signal, speech signal, and noise signal at different modulation frequencies; obtaining signals of different clustered frames of the sub-band spectra through frame clustering analysis; training a joint sub-dictionary using signals from different clustered frames of the sub-band spectra, and then concatenating the sub-dictionaries into a mixed, speech, and noise structured dictionary. A testing phase: calculating the sub-band spectrum of the mixed test signal, preserving its acoustic and modulation phase; calculating the projection coefficients of the mixed test signal's sub-band spectrum onto the mixed structured dictionary through group sparse coding; multiplying the structured dictionary and projection coefficients to recover the speech and noise modulation amplitude spectra, and combining this with a noisy modulation phase inverse transform to obtain the acoustic amplitude spectrum; calculating the weights of the two estimated sets based on the Gini coefficient, and optimizing the estimation using a Wiener filter. This invention utilizes the structured characteristics of modulation domain features as prior information, improving the performance of single-channel speech enhancement.