35 results about "Weighting filter" patented technology

The invention belongs to the technical field of speech recognition and acoustic modeling, and particularly relates to a high-fidelity end-to-end acoustic modeling method combining speech intelligibility reconstruction and autoregressive feedback optimization, which comprises the following steps of: firstly, establishing an intelligibility loss function based on a human speech intelligibility model; subjective definition features of the voice are reconstructed through a frequency spectrum reconstruction network, intelligibility related features are extracted and embedded in combination with a perceptual weighted filter bank, and the intelligibility related features and Mel frequency spectrum features are input into an acoustic encoder in parallel; based on this, by introducing a voice intelligibility reconstruction and autoregressive feedback mechanism, while the conciseness of an end-to-end voice modeling framework is maintained, systematic improvement of voice sharpness, naturalness and stability is realized, and the method can be widely applied to scenes such as intelligent voice assistants, voice transcription, virtual anchors, voice restoration, cross-language voice generation and the like. And the method has extremely high practical application value and popularization prospect.

Methods and systems are disclosed for a vehicle audio system including, in one example, a method for noise cancellation in a vehicle having a reference sensor configured to acquire a reference signal, a plurality of speakers configured to emit a noise cancellation signal, and a plurality of error microphones configured to acquire a residual signal. The method processes the reference signal with a time domain adaptive weight filter to produce the noise cancellation signal, estimates a frequency domain filtered virtual microphone signal from the noise cancellation signal and the residual signal, and a frequency domain filtered reference signal from the reference signal. The method decomposes the frequency domain filtered virtual microphone signal and the frequency domain filtered reference signal into a plurality of frequency domain subband signals, and updates the time domain adaptive weight filter based on a weight transformation of a plurality of frequency domain subband adaptive filter weights.

The application discloses a low-noise tire pattern pitch optimization method and device based on image processing and a medium, and the method comprises the following steps: S1, processing initial pattern information and calculating pitch length; S2, extracting groove boundaries and tire mark boundaries of a pattern pitch graph; S3, inputting basic parameters such as pitch and mark; S4, introducing an A-weighting filter for acoustic weighting; S5, calculating an initial noise spectrum; S6, optimizing a spectrum peak value in a harmonic range; S7, iteratively adjusting a pitch ratio and quantity until noise energy meets a threshold value, and outputting an optimal pitch sequence. Through the fusion of image processing technology and acoustic analysis, the pattern pitch design is systematically optimized, tire noise is significantly reduced, design efficiency is improved, and a research and development cycle is shortened.

The invention relates to the technical field of intelligent control, in particular to a laser multi-axis flight processing path planning method for adaptive weighted filtering, which comprises the following steps: inputting a three-dimensional curved surface model and a target two-dimensional path, setting constraint thresholds of defocusing amount, a laser incident angle and a galvanometer scanning range, and setting a lower bound of a filtering factor; generating a weight matrix of the midpoint filtering factor and the initialization filtering factor; performing weighted filtering on the target two-dimensional path through an adaptive weighted filtering target function to obtain a smooth two-dimensional path; projecting the smooth two-dimensional path to a three-dimensional curved surface to obtain a platform processing path, calculating constraint parameters and constraint margin, generating segment filtering factor weights and updating a weight matrix; repeating the steps until iteration is finished; judging whether the platform machining path meets the constraint or not, updating the search interval, and repeating the steps to output a final platform machining path; and calculating a galvanometer processing path. According to the method, the path smoothness is improved, so that the average feeding speed and machining efficiency of the motion platform are improved.

The application discloses a bone conduction speech conversion method based on spectral envelope mapping, comprising the following steps: pre-processing and linear prediction analysis of the bone conduction speech signal, and calculating the LP filter coefficient; mapping the LSF coefficient of the air conduction speech signal corresponding to the bone conduction speech signal by using the trained neural network; controlling the minimum error of the original bone conduction speech signal and the synthesized air conduction speech signal according to the timbre weighting characteristic of the bone conduction speech characteristic; estimating the integer pitch of the bone conduction speech signal through the timbre weighting filter, obtaining the reference signal by passing the linear prediction residual signal through the timbre weighting filter, estimating the fractional pitch to obtain the adaptive codebook vector; obtaining the new reference signal by subtracting the adaptive codebook vector from the reference signal, searching for the optimal excitation in the fixed codebook; synthesizing the air conduction speech signal by combining the optimal excitation and the LP filter of the air conduction speech signal, and correcting the air conduction speech signal.

The application provides a vehicle acceleration determination method, device, medium and electronic equipment. The method comprises the following steps: calculating a difference value between a fusion vehicle speed value of a target detection period and a fusion vehicle speed value of a start detection period in a first adjacent detection period group, dividing the difference value by a time length of a plurality of adjacent detection periods in the first adjacent detection period group, and obtaining a to-be-determined acceleration value of the target detection period; performing weighted filtering on the to-be-determined acceleration values of the plurality of adjacent detection periods in a second adjacent detection period group and preset calibration weight values of the corresponding detection periods, and obtaining an estimated acceleration value of the target detection period; and obtaining a target acceleration value of the target detection period based on a measured acceleration value of the target detection period and a preset measured weight value, and the estimated acceleration value of the target detection period and a preset estimated weight value. The target acceleration value of the target detection period is finally generated after comprehensively considering various influencing factors, thereby improving the accuracy of the target acceleration value.

The invention discloses a multilingual off-line simultaneous interpretation method based on a small bidirectional translation model. According to the method, voice front-end enhancement, automatic voice recognition, neural machine translation and text-to-voice conversion are sequentially executed locally on a terminal, and a unified sub-word list and a shared model are adopted to automatically judge translation between at least two languages and generate a target text / voice in a stream mode. In order to adapt to the mobile terminal, the translation model is quantized by an integer and deployed in a lightweight format, and the time delay is reduced in combination with beam width search, early stop and cache mechanisms. The voice front end adopts a causal U-shaped network to jointly suppress reverberation and noise, room impulse response and multi-noise are introduced in the training stage, a signal-to-noise ratio strategy is configured, and time sequence weighted filtering and mean square error gain are combined in the reasoning stage to improve intelligibility. The device is composed of a mobile terminal and a Bluetooth earphone, supports a collaborative chain of earphone side front enhancement and mobile phone side decoding, and realizes stable simultaneous transmission output in a weak network / no network scene.

This invention provides a video signal decoding / encoding method and a bitstream transmission method. The video signal decoding method includes the following steps: determining an intra-frame prediction mode for a current block; obtaining reference samples based on neighboring samples of the current block; obtaining a first prediction sample for the current block based on the reference samples and the intra-frame prediction mode; obtaining a second prediction sample for the current block by applying weighted filtering to the reference samples and the first prediction sample; and obtaining a reconstructed sample for the current block based on the second prediction sample.

The invention discloses a gated jump connection network denoising method fusing multi-scale space-time attention. According to the method, a coding-decoding network fusing a multi-scale parallel convolution structure, a gating jump connection mechanism and a BiLSTM time sequence modeling unit is constructed, so that adaptive denoising and feature reconstruction of a noisy one-dimensional time sequence signal are realized. Wherein the encoder captures local details and long-period characteristics of a signal by using multi-scale parallel convolution; in the decoding stage, a gating jump connection mechanism is embedded, and channel splicing is carried out on shallow layer features of an encoder and deep layer features of a decoder; a driver attention sub-network is embedded to generate a gating mask, the mask is used to carry out adaptive weighted filtering on shallow layer features, and background noise is blocked from spreading to a decoding end before fusion. According to the method, multi-scale representation considering local transient change and long-term evolution trend can be obtained, the input quality of subsequent feature extraction, pattern recognition and life evaluation algorithms is remarkably improved, and the accuracy and reliability of a fault diagnosis process are improved.

This invention relates to the field of image recognition technology, specifically a method for identifying transparent foreign fibers in cotton using polarization detection combined with image processing. The method includes the following steps: acquiring images using a polarization sensor and extracting temporal and angular information; calculating pixel phase delay; performing frequency domain analysis to determine the principal direction and generating a birefringence response; screening skeleton points based on grayscale and constructing fiber paths; combining phase gradients for extreme value analysis and intensity reconstruction to obtain global coordinates; and then weighted filtering based on directional differences to eliminate redundancy and obtain fiber position coordinates. In this invention, by combining image acquisition and feature expression paths, polarization response and texture information are fused to form multi-dimensional discriminative features, reducing noise and uneven illumination interference. Combining fiber continuity and optical response analysis enhances the ability to identify weak responses, making boundaries clearer and more stable. Furthermore, redundant points are compressed under directional classification and spatial constraints, improving positioning consistency and recognition accuracy, and achieving stable identification of transparent foreign fibers in complex scenarios.

Methods and systems for a vehicle audio system are disclosed, in one example, including a method for noise cancellation in a vehicle having: a reference sensor configured to acquire a reference signal; a plurality of speakers configured to emit a noise cancellation signal; and a plurality of error microphones configured to acquire a residual signal. The method: processes the reference signal with a time domain adaptive weight filter to generate the noise cancellation signal; a frequency domain filtered virtual microphone signal is estimated from the noise cancellation signal and the residual signal, and a frequency domain filtered reference signal is estimated from the reference signal. The method comprises: decomposing the frequency domain filtered virtual microphone signal and the frequency domain filtered reference signal into a plurality of frequency domain sub-band signals; and updating the time domain adaptive weight filter based on a weight transform of a plurality of frequency domain sub-band adaptive filter weights.

The invention provides a pressure assignment method and system based on an open air spring system, and belongs to the technical field of vehicle intelligent chassis control. The method comprises the following steps: acquiring a vehicle state signal in a whole vehicle CAN network and a sensor signal of the open air spring system in real time; performing multi-dimensional feature extraction on the signal, and identifying a vehicle motion state and a state feature parameter through an improved random forest algorithm; dynamically distributing weights of a plurality of filtering modules according to an identification result, and performing weighted filtering on the air spring pressure original signal; a closed-loop feedback mechanism is introduced, the deviation between a filtering pressure value and a theoretical pressure value is calculated in real time, the filtering weight and parameters are dynamically corrected, and finally accurate assignment of the air spring pressure is achieved. According to the method, the problem of pressure assignment deviation caused by inaccurate working condition recognition, single filtering mode and lack of calibration in the prior art is effectively solved, and the control precision of the electric control shock absorber and the operation stability and comfort of the whole vehicle are remarkably improved.

The invention relates to a method, a corresponding device and a medical technology system in the form of an MRI system and / or a diagnostic system for reconstructing images in magnetic resonance imaging from provided raw data in the form of asymmetrically acquired k-space data, wherein the raw data have a symmetric and an asymmetric part, the method comprising the steps: - Reconstructing a phase image using an image space transformation of k-space data from the symmetric part of the raw data, - Applying an iterative k-space reconstruction method, starting with a working image reconstructed from the raw data to generate a new working image using the phase image, and including forming a working space through a k-space transformation of the working image, - Applying an image reconstruction method to the workspace, providing a weighting filter which, if the workspace corresponds to the raw data, has a weighting of zero wherever no raw data has been recorded, and has exclusively non-zero weightings for all other workspaces, and which has a lower weighting with respect to the raw data recorded symmetrically in k-space than with respect to the raw data recorded asymmetrically in k-space, and generating a complex intermediate image by means of an image space transformation from the k-space data weighted according to the weighting filter, which is phase-corrected with the phase image, and generating a result image as the real part of the phase-corrected intermediate image.

This application provides an image enhancement method, electronic device, storage medium, and program product. The method includes: acquiring pixel difference data of a first chromaticity plane, a second chromaticity plane, and a luminance plane of an image to be processed, wherein the pixel difference data characterizes the difference between the pixel value of each pixel within a sliding window and the pixel value of the center pixel; the first chromaticity plane and the second chromaticity plane are obtained by performing resolution transformation processing on the first initial chromaticity plane and the second initial chromaticity plane respectively; determining filtering weights based on the pixel difference data, the filtering weights characterizing the contribution of the pixel value of each pixel within the current sliding window to the pixel value of the center pixel; performing weighted filtering processing on the luminance plane according to the filtering weights to obtain a target luminance plane; and determining a target image based on the target luminance plane, the first initial chromaticity plane, and the second initial chromaticity plane. This method can improve image quality during image conversion.

A method of processing video data, performed by a decoder, is provided. The method includes decoding a bitstream to obtain video data and coding information; obtaining a picture based on the video data; determining a weighting map using a weighting map function, the weighting map comprising a plurality of weights mapped to respective spatial locations of the picture, wherein the picture and / or coding information are used as inputs to the weighing map function; determining a filter; and applying the weighting map and the filter to the picture to obtain a filtered picture.

The invention discloses an improved power distribution method for stabilizing wind power fluctuation through hydrogen composite energy storage, and the method comprises the steps: obtaining wind power output power data, decomposing power through employing an algorithm combining adaptive moving average and amplitude limiting weighted filtering, and obtaining grid-connected reference power and composite energy storage execution reference power; dividing a composite energy storage execution reference power frequency band through wavelet packet decomposition based on the response characteristic difference between the hydrogen energy storage and the super capacitor, realizing primary power distribution and outputting initial output reference power; then introducing a reference line variable, and calculating the minimum difference value of the accumulated charge and discharge quantities to compensate the power loss and balance the charge and discharge quantities; and constructing a power collaboration mechanism and a collaboration rule by combining the operation characteristics of the electrolytic cell and the state of the energy storage equipment, secondarily correcting the initial output reference power by using a double-end Logistic function correction coefficient, constraining the output in a safety interval, and outputting the corrected output power, thereby realizing the wind power fluctuation stabilization and the efficient collaboration operation of the energy storage system.

The invention belongs to the field of satellite navigation, and particularly relates to a random channel fault-oriented anti-interference adaptive degradation method, which comprises the following steps of: realizing channel power estimation and fault detection through intermediate frequency data; judging whether the reference main array element has a fault or not through the fault list; an array with the best satellite receiving condition is set as a temporary reference main array, and meanwhile, a constraint vector is adjusted to carry out weight calculation and weighted filtering; dynamic optimization and cyclic detection are set to realize adaptive tracking of channel faults and self-recovery of an anti-interference algorithm after the faults are cleared. According to the method, a fixed main array is not set in an anti-interference algorithm, and the temporary reference main array is adaptively adjusted by performing periodic detection on the channel fault, so that the anti-interference adaptive degradation capability after the channel fault is realized, and the robustness of an anti-interference receiver facing the random channel fault is effectively improved.

The invention discloses an intelligent lightning arrester with optimized energy absorption. The intelligent lightning arrester is characterized in that a reference signal is processed by adopting an adaptive energy weight filter so as to generate an energy dispersion control signal; decomposing the residual heat signal into a plurality of sub-band heat error signals and into a plurality of sub-band energy reference signals; applying the sub-band gradient transfer function to generate a plurality of sub-band energy reference signals; determining a sub-band fusion gradient through the filtered sub-band energy reference signal and the filtered sub-band heat error signal; determining an adaptive step size; updating the weight based on the adaptive step size and the fusion gradient; and integrating each weight to generate an updated weight of the adaptive energy weight filter, wherein the updated weight is applied to the adaptive energy weight filter to adjust the energy dispersion control signal. According to the lightning arrester, the calculation limitation of a traditional lightning energy absorption optimization system is solved, and energy dispersion and heat dissipation management in a wide frequency spectrum are improved.

According to another aspect of the present disclosure, a system is provided. The system may include a processor and a memory storing instructions. The instructions stored in the memory, when executed by the processor, cause the processor to identify whether a filtered intra block copy (FIBC) mode is enabled for a current block. The instructions, when executed by the processor, cause the processor to generate a filtered predictive block by applying a shift-invariant weighting filter to a reference block in response to the FIBC mode being enabled for the current block. The instructions, when executed by the processor, cause the processor to encode the current block based on the filtered predictive block.

A method of processing video data performed by an encoder is provided. The method comprises the following steps: acquiring a plurality of original images from original video data; a plurality of reference images are obtained, each reference image in the plurality of reference images corresponds to an original image in original video data, and the resolution of the plurality of reference images is lower than the resolution of the plurality of original images in the original video data; performing up-sampling on the plurality of reference images to obtain a plurality of up-sampled reference images; a weighted filter is obtained to reduce overall error between the plurality of up-sampled reference images and corresponding original images by determining a weighted mapping for each reference image using a weighted mapping function, the weighted mapping comprising a plurality of weights mapped to respective spatial locations of the plurality of up-sampled reference images, and determining a filter that is applied to the plurality of up-sampled reference images by respective weighted mappings to obtain a filtered up-sampled reference image such that the filter is applied to different spatial positions of the plurality of up-sampled reference images with different weights; performing inter prediction on a plurality of blocks of the first image based on a plurality of reference blocks in one or more of the plurality of up-sampled reference images; and encoding the video data encoding information into a code stream, the encoding information comprising information on the weight mapping function and / or the filter to be used at a decoder, where the method further comprises applying the weight filter to the plurality of upsampled reference pictures prior to performing the inter prediction, and applying the weight filter to the plurality of upsampled reference pictures prior to performing the inter prediction. Or during the inter prediction, applying the weighting filter to an up-sampled reference block of the plurality of reference pictures.

The invention discloses a dynamic load change rate calculation method, system and device and a readable storage medium, and the method comprises the steps: collecting the input power and the output power of a load system, calculating the difference value between the input power and the output power, and obtaining the energy deviation caused by the transient change of a load; calculating an instantaneous load change rate based on the energy deviation caused by the transient change of the load and the dynamic time window; and performing time weighted filtering on the instantaneous load change rate to obtain a dynamic load change rate. According to the method, the problem of noise amplification caused by direct differential of the power signal is effectively avoided, and the estimation accuracy of the load change trend is improved on the premise that a complex control algorithm is not introduced.

The application discloses a kind of dynamic residual threshold adaptive selection space-time filtering method, device, equipment and storage medium, the method includes obtaining the remote sensing time series image data of target area;By model training and predicting the possible change trend of remote sensing time series data obtains the prediction result of remote sensing time series data.Furthermore, the absolute residual between image predicted value and image true value is calculated, and the residual threshold determined based on the absolute residual median, quartile range, skewness and kurtosis is selected as the threshold of filtering method selection, specifically, for high residual area, spatiotemporal weighted filtering strategy combined with Euclidean distance is used;For low residual area, the filtering reconstruction method in time domain is used, so as to ensure the continuity of image while suppressing non-systematic noise.Compared with prior art, it has higher degree of automation and adaptability, and is especially suitable for remote sensing time series data processing in high-altitude, frequent cloud cover or complex terrain area.

The invention provides a sound velocity imaging method based on a concave ultrasonic transducer and a related device. The method comprises the following steps: acquiring full-matrix ultrasonic radio frequency data of a tested tissue through a semi-ring array ultrasonic transducer in a synthetic aperture acquisition mode; complex radio frequency image reconstruction is carried out on the full-matrix data, mask and Gaussian weighted synthesis is carried out based on semi-ring array sound field distribution characteristics, and a focused complex radio frequency image with a higher signal-to-noise ratio and richer angle information is obtained; obtaining a stable and accurate cross-event propagation time delay difference through cross-event phase difference extraction and two-stage Gaussian weighted filtering processing according to the focused image; the time delay difference serves as an observation vector, a ray tracing inversion model matched with semi-ring array geometry and regularization solution are combined, a high-resolution sound velocity distribution diagram of the measured tissue is reconstructed, and the tiny time difference in observation data can be decoded into a high-quality space sound velocity distribution diagram; it is ensured that high-precision and high-robustness quantitative sound velocity imaging can still be achieved under the scene constraint of the geometrical characteristics of the semi-ring array.

A method for reconstructing MR tomography images from asymmetrically acquired k-space raw data, with symmetrical and asymmetrical parts, may include reconstructing a phase image from the symmetrical k-space data, applying an iterative k-space reconstruction starting with a base image, and forming a working space via k-space transform. A weighting filter may be applied, assigning zero weight where no raw data exists, lower weight to symmetrical data, and non-zero weight to other data. A complex intermediate image is generated by image space transform of weighted k-space data, phase-corrected with the phase image, and the final result image is obtained as the real part of the phase-corrected intermediate image.

A diffusion skewness imaging method with positive semi-definite constraint based on Q-space trajectory imaging comprises the following steps: 1) expanding a diffusion signal and a diffusion tensor distribution model, and introducing a high-order skewness tensor; 2) solving the diffusion tensor distribution model expanded to the high-order skewness direction by utilizing positive semi-definite programming to obtain a high-order skewness tensor; diffusion tensor lt is carried out on related solving variables; dgt; performing positive semidefinite constraint on the covariance tensor and the skewness tensor; and 3) designing a linear trajectory weighting filter and a secondary trajectory weighting filter according to a result obtained by calculation. A large number of experimental verification is carried out on a public data set, a noisy data set and a synthetic data set, and experimental results show that the method can obtain an estimation result closer to a true value on the synthetic data set and shows high robustness on the noisy data set.

The invention discloses a point cloud filtering method based on curvature entropy adaptive parameters, and belongs to the technical field of computer vision and three-dimensional data processing. The method comprises the following steps: firstly, for each target point in a point cloud, calculating a curvature representing a local geometrical shape of the target point; secondly, statistical binning is carried out on curvature values of the target point and all points in the neighborhood of the target point, and a discrete local curvature probability distribution histogram is constructed; then, the curvature entropy of the target point is calculated according to the probability distribution; thirdly, establishing a dynamic response model between the filtering parameters and the curvature entropy, and adaptively adjusting each parameter in a filtering weight function by using the calculated curvature entropy; and finally, performing weighted filtering on the target point by adopting the adaptively adjusted filtering parameters to complete filtering processing. According to the method, the curvature entropy based on statistical distribution is introduced to serve as a control variable of geometric features, point-by-point self-adaptive optimization of filtering parameters is achieved, and the protection capacity for fine features in a complex scene and the noise suppression effect are remarkably improved.

The application provides a dual-mode end-to-end target detection method for unmanned aerial vehicles and complex scenes, and relates to the technical field of unmanned aerial vehicle target detection. In the input and feature extraction stage, a channel splitting operation is used to physically isolate the visible light and infrared data, and the data is transmitted to a parallel dual-flow convolutional network for independent multi-scale feature extraction. In the deep feature fusion stage, a cross-modal difference attention module is added. The spatial cosine similarity is calculated to generate a difference saliency mask, which helps the model to give high weights to the infrared features in areas with inconsistent feature structures. A semantic-guided spatial weighted filtering module is embedded in the neck network. The deep high-level semantic features are used to perform pixel-by-pixel weighted filtering on the shallow features, which can retain the edge details of small targets while greatly suppressing background clutter. At the same time, a scale adaptive weight mechanism is introduced during model optimization to increase the regression penalty for small targets.

The application relates to the field of data processing, in particular to a data acquisition method and system for an electric energy storage box, the method comprising: binning energy storage box data of each dimension to obtain a data range, and constructing an item set according to the data range; calculating a frequent coefficient according to the item set; calculating a weight coefficient of the energy storage data according to the frequent coefficient, and obtaining a filtered value after filtering the energy storage data according to the weight coefficient. The application improves the limitation of the weighted filtering on the processing of burst noise, and has the effect of improving the accuracy of the energy storage box data filtering.

The invention provides a coherent fading suppression method and system based on phi-OTDR (Optical Time Domain Reflectometer), and the method can effectively improve the phase demodulation precision through frequency domain weighted filtering instead of moving averaging in the conventional time domain MRVA (Multi-Redundant Frequency Amplitude), and can remarkably shorten the data processing time during large data volume processing or large-window moving averaging; coherent fading suppression can be realized in a digital domain under the condition that the system complexity is not increased and a typical heterodyne coherent detection optical path is adopted. According to the technical scheme, the technical problem that high-precision and real-time phase demodulation cannot be met at the same time in the prior art is solved.

The invention discloses a multi-channel sound mixing processing method and system, and belongs to the technical field of multi-channel sound mixing processing, and the method comprises the step of carrying out millisecond-level delay calibration between channels through dynamic phase alignment according to multi-channel original audios. A multi-dimensional frequency deconstruction technology realizes high-precision separation of a basic frequency body, a formant matrix and a transient profile, and extraction of the basic frequency body can independently enhance fundamental frequency energy and improve voice definition or plumpness of musical instrument timbre; the accurate positioning of the formant matrix can simulate the spatial reflection characteristics of different musical instruments; the impact force of the sound is reserved by the capture of the wavelet packet decomposition on the transient event, the self-adaptive window length design is combined with the energy weighted filtering, so that the calculation complexity is effectively reduced, the multi-scene requirement is compatible, the gain or delay of the three types of separated spectrum components can be independently adjusted, and the controllability of audio processing is remarkably improved.