23 results about "Stationary wavelet transform" patented technology

The invention provides a low-illumination perception method and system based on space-frequency fusion, and relates to the technical field of image processing, and the method comprises the steps: obtaining a normal light image and a low-illumination image corresponding to the normal light image; performing feature extraction on the normal light image and the low-illumination image through an encoder to obtain multi-scale spatial features; decomposing the multi-scale spatial features through stationary wavelet transform to generate a low-frequency feature component and a high-frequency feature component; performing optimization processing on the low-frequency characteristic component and the high-frequency characteristic component based on a parameter learnable frequency domain adaptive filtering mechanism; fusing the optimized low-frequency feature component and the optimized high-frequency feature component by introducing a cross attention learning mechanism to obtain a frequency domain feature; performing adaptive complementary fusion on the frequency domain features and the multi-scale spatial features to generate fusion features; step-by-step up-sampling is carried out on the fusion features through a decoder; and outputting a segmentation result of the low-illumination image through a sensing head according to the fusion features after up-sampling.

The invention discloses a robust lightweight time sequence prediction method based on bidirectional filling and geometric attention, and belongs to the field of multivariate time sequence data missing filling and prediction. The problem of monitoring data missing caused by frequent failure of the sensor in a complex geological environment is solved, and the accuracy and reliability of geological disaster prediction are improved. The invention provides a co-processing framework which is used for fusing a bidirectional cyclic interpolation time sequence (BRITS) and a SimpleTM light-weight prediction model, and is characterized in that the BRITS and the SimpleTM light-weight prediction model are combined with each other, and the BRITS and the SimpleTM light-weight prediction model are combined with each other. According to the framework, firstly, forward and backward time sequence dependence and multivariate variable relevance of sensor data are captured by utilizing a BRITS model, a failure mode is accurately recognized, and efficient data filling is realized; and then, in combination with a specific stable wavelet transform multi-scale decomposition and geometric product attention mechanism of the SimpleTM model, fine-grained time sequence characteristics of the filled data are deeply mined under low calculation overhead, and reliable support is provided for subsequent disaster prediction. According to the method, the prediction precision and robustness in a data missing scene are remarkably improved, meanwhile, the lightweight deployment requirement in a resource-limited tunnel construction environment is considered, and the railway tunnel construction safety is effectively guaranteed.

The application discloses a laser communication link correction method, system, device, medium and product based on wavelet interpolation, relates to the field of optical communication, and comprises the following steps: performing stationary wavelet transform and discrete wavelet transform on a light spot image respectively to obtain stationary subband coefficients and discrete subband coefficients; correcting the discrete high-frequency subband coefficients after interpolation processing by using the stationary high-frequency subband coefficients to obtain corrected discrete high-frequency subband coefficients; replacing the discrete low-frequency subband coefficients with the coefficients of the expanded light spot image to obtain replaced discrete low-frequency subband coefficients; through inverse discrete wavelet transform, the centroid method is used again to calculate the centroid coordinates of the light spot image after wavelet interpolation processing and the deflection angle of the beacon light, the deflection angle of the receiving end is adjusted according to the deflection angle of the beacon light, and the correction of the laser communication link is completed; and the stability of the laser communication link can be improved by improving the subdivision accuracy of the light spot.

The invention discloses a terahertz near-field radar image enhancement method, which comprises the following steps of: performing data conversion and normalization on a terahertz near-field radar image to be enhanced to obtain a normalized terahertz near-field radar image; performing stationary wavelet transform decomposition on the normalized terahertz near-field radar image to obtain a plurality of low-frequency sub-bands and a plurality of high-frequency sub-bands; performing high-frequency multi-direction filtering on the high-frequency sub-bands by using filters in different directions to obtain a plurality of enhanced high-frequency sub-bands; performing low-frequency adaptive linear enhancement on the low-frequency sub-bands by using the mean value and the standard deviation of the low-frequency sub-bands to obtain a plurality of enhanced low-frequency sub-bands; performing inverse stationary wavelet transform on the plurality of enhanced high-frequency sub-bands and the plurality of enhanced low-frequency sub-bands to obtain a reconstructed terahertz near-field radar image; and carrying out linear transformation on the reconstructed terahertz near-field radar image to obtain an enhanced terahertz near-field radar image. The method can enhance contrast, highlight edge features and enrich detail information.

The invention belongs to the technical field of time prediction, and particularly relates to a time prediction method and device based on different channel attention attention mechanisms. Inputting the scaled household electricity consumption time sequence data vector into an MLP (Multimedia Library Protocol) to obtain a processed output vector, and carrying out stationary wavelet transform on the processed output vector; based on the high-frequency detail components, carrying out Granger causal analysis, and carrying out inverse Granger transformation on the high-frequency detail components to obtain reconstructed household power utilization time sequence data; on the basis of the low-frequency component and the intermediate-frequency component, calculating correlation between attention of different channels, and screening a plurality of channels with the strongest dependency; self-attention coupling is used for channels with high dependency, geometric self-attention coupling is used for channels with low dependency, and output of self-attention and output of geometric self-attention are fused to obtain comprehensive features; and inverse transformation is carried out on the reconstructed frequency domain sequence data and the comprehensive features, and a final prediction result is obtained through FFN.

The invention discloses an anti-interference radar signal extraction method and device and electronic equipment, and the method comprises the steps: carrying out the whitening processing of a to-be-processed radar signal through a zero-phase component analysis whitening algorithm, and obtaining a to-be-separated radar signal after the whitening processing. An initial separation matrix is obtained by performing iterative optimization processing on a preset target separation function for multiple times. And performing fine adjustment on the initial separation matrix according to a second-order blind source separation algorithm to obtain a target separation matrix. Performing signal separation processing on the radar signal and the interference signal according to the target separation matrix to obtain an initial radar signal, and performing denoising processing on the initial radar signal according to a target denoising algorithm to obtain a radar signal, the target denoising algorithm being any one of a wavelet transform denoising algorithm and an improved stationary wavelet transform denoising algorithm. The problems of high calculation complexity and low efficiency in the prior art are avoided, and the stability and accuracy of radar signal extraction are improved.

The invention belongs to the field of computer vision, designs a WBM-MSF model for monocular scene flow estimation, and designs a frequency domain context feature extraction network and an iterative update network WBM based on wavelet transform. The frequency domain context feature extraction network decomposes an input image into frequency sub-bands with different scales and directions by using discrete stationary wavelet transform, and extracts a low-frequency global structure and high-frequency detail features under the condition of not losing spatial resolution. An iterative update network WBM iteratively updates the hidden state through a state space model Mamba, introduces stationary wavelet transform to enhance the frequency domain feature representation capability, and ensures that the hidden state does not lose local high-frequency information in the iteration process through the global modeling capability of the Mamba. According to the method, the problems that details are lost, the edge is fuzzy and wrong matching is generated in a non-texture area due to the fact that an existing model is difficult to capture high-frequency information of the edge and low-frequency information of a smooth area at the same time are effectively solved, frequency domain features are fully utilized, and finally the precision of monocular scene flow estimation is improved.

The invention provides a feedback type denoising heart beat detection method suitable for a high noise condition. The method comprises a signal quality evaluation classification step, a threshold value updating step, a preliminary noise reduction step, a layered denoising step, a heart beat recognition step and a feedback type R peak detection correction step. According to the method, based on feedback denoising cardiac beat detection combining stationary wavelet transform with neural network classification, denoising is performed through a staged wavelet threshold algorithm, signal time sequence characteristics are reserved by using stationary wavelet transform, cardiac beats are classified and screened in combination with a neural network, and a detection result is corrected through repeated iteration, so that the accuracy of cardiac beat detection is improved. Therefore, high-precision R peak detection is realized and signal noise reduction is achieved. According to the method, the detected R peak position is used for providing auxiliary information for denoising, so that the denoising effect is enhanced, meanwhile, the denoised signal can verify the accuracy of R peak detection in turn, and the accuracy of R peak detection is improved. The heart beat detection anti-noise performance is good, and the method is suitable for real-time heart beat monitoring and extraction of wearable equipment.

The application discloses a hybrid DC line protection method and system based on wavelet transform and energy ratio, and relates to the technical field of high-voltage direct current transmission system relay protection. The method comprises the following steps: synchronously collecting polar voltage and polar current signals at the left end and the right end protection installation of a hybrid DC line; performing stationary wavelet transform on the fault transient components of the double-end line mode voltage and the line mode current respectively, and locating the first arrival time of the fault wave head of each double end; taking the first arrival time of the fault wave head as the starting point, respectively intercepting a characteristic window with a preset time length, calculating the left end transient energy and the right end transient energy, and performing amplification processing to obtain the amplified double-end energy ratio; and judging the physical section where the fault occurs according to the preset ratio interval where the amplified double-end energy ratio is located. The application can fully adapt to the harsh operating environment containing the impedance mutation of a submarine cable-overhead line and long-distance deep attenuation, and realizes fault determination and protection of the double-end DC line.

The application belongs to the technical field of time prediction, and particularly relates to a time prediction method and device based on different channel attention focusing mechanisms. A scaled household electricity time sequence data vector is input into an MLP to obtain a processed output vector, and a stationary wavelet transform is performed on the output vector; based on high-frequency detail components, Granger causality analysis is performed, inverse Granger transformation is performed on the high-frequency detail components, and reconstructed household electricity time sequence data is obtained; based on low-frequency components and medium-frequency components, the correlation between different channel attentions is calculated, and multiple channels with the strongest dependency are screened; self-attention coupling is used for the channels with strong dependency, and geometric self-attention coupling is used for the channels with weak dependency; the outputs of the self-attention and the geometric self-attention are fused to obtain comprehensive features; the reconstructed frequency domain sequence data and the comprehensive features are inversely transformed again, and a final prediction result is obtained through an FFN.

The invention relates to the technical field of artificial intelligence, in particular to an X-Ray image enhancement method and related equipment based on unsupervised learning, and the method comprises the steps: obtaining a to-be-enhanced X-Ray image and a trained unsupervised image enhancement network; inputting an X-Ray image to be enhanced into the unsupervised image enhancement network, and performing noise suppression processing on the input X-Ray image through the first denoising sub-network to obtain a denoised feature map; after deep features in the de-noised feature map are extracted through a CBR module and a ResCBR module in an illumination estimation network, non-uniform illumination is processed through a multi-scale spatial pyramid structure, and a smooth and boundary-sensing illumination feature map is generated; processing the illumination feature map through a second denoising sub-network, and performing structured noise reduction through alternate operation of convolution and stationary wavelet transform to obtain an enhanced image; according to the invention, the definition and contrast of the defect area in the X-Ray image can be improved.

The present application relates to the field of artificial intelligence, in particular to an X-Ray image enhancement method based on unsupervised learning and related equipment, the method comprising: obtaining an X-Ray image to be enhanced and a trained unsupervised image enhancement network; inputting the X-Ray image to be enhanced into the unsupervised image enhancement network, performing noise suppression processing on the input X-Ray image through a first denoising subnetwork to obtain a denoised feature map; after extracting deep features in the denoised feature map through a CBR module and a ResCBR module in an illumination estimation network, processing non-uniform illumination through a multi-scale spatial pyramid structure to generate a smooth and boundary-aware illumination feature map; processing the illumination feature map through a second denoising subnetwork, and performing structured denoising through the alternative operation of convolution and stationary wavelet transform to obtain an enhanced image; the present application can improve the clarity and contrast of the defect area in the X-Ray image.

The application provides a low-illumination perception method and system based on space-frequency fusion, and relates to the technical field of image processing. The method comprises the following steps: acquiring a normal light image and a low-illumination image corresponding to the normal light image; performing feature extraction on the normal light image and the low-illumination image through an encoder to obtain multi-scale spatial features; performing decomposition on the multi-scale spatial features through a stationary wavelet transform to generate low-frequency feature components and high-frequency feature components; performing optimization processing on the low-frequency feature components and the high-frequency feature components based on a parameter-learnable frequency domain adaptive filtering mechanism; introducing a cross-attention learning mechanism to fuse the low-frequency feature components and the high-frequency feature components after optimization processing to obtain frequency domain features; performing adaptive complementary fusion on the frequency domain features and the multi-scale spatial features to generate fusion features; performing step-by-step up-sampling on the fusion features through a decoder; and outputting a segmentation result of the low-illumination image through a perception head according to the up-sampled fusion features.

This invention discloses a method and system for scene text super-resolution based on text prior and stationary wavelet domain transform. The method includes: constructing an initial super-resolution image generation neural network model and inputting historical low-resolution text images into the model to obtain historical high-resolution text images; applying stationary wavelet transform to the historical high-resolution text images to obtain low-frequency sub-bands and several high-frequency sub-bands, and performing loss calculation to construct a frequency domain loss function; constructing a discriminator network model and calculating an adversarial loss function; constructing a generator total loss function based on the frequency domain loss and adversarial loss, and using the total loss to update and iterate the initial super-resolution image generation neural network model and the discriminator network model to obtain a super-resolution image generation neural network model; acquiring the low-resolution text image to be processed, and using the super-resolution image generation neural network model to process the low-resolution text image to be processed to obtain the super-resolution result.

The application discloses a robust light-weight time series prediction method based on bidirectional imputation and geometric attention, and belongs to the field of missing data filling and prediction of multivariate time series. The problem of missing monitoring data caused by frequent sensor failure in complex geological environment is solved, and the accuracy and reliability of geological disaster prediction are improved. The application proposes a collaborative processing framework that fuses bidirectional recurrent imputation for time series (BRITS) and SimpleTM light-weight prediction model. Firstly, the BRITS model is used to capture the forward and backward time series dependence and multivariate correlation of sensor data, accurately identify the failure mode and achieve efficient data filling; then, combined with the specific stationary wavelet transform multi-scale decomposition and geometric product attention mechanism of the SimpleTM model, the fine-grained time series characteristics of the filled data are deeply mined under low computational overhead, providing reliable support for subsequent disaster prediction. The application significantly improves the prediction accuracy and robustness in the missing data scenario, while taking into account the light-weight deployment requirements in the resource-constrained tunnel construction environment, effectively ensuring the safety of railway tunnel construction.

This application discloses a dual-modal anti-drone detection method based on a geometrically perceptive space-frequency interaction network, belonging to the field of computer vision and pattern recognition technology. The method includes: preprocessing visible light and infrared images by homography transformation matrix registration; constructing a pseudo-twin dual-stream backbone network to independently extract dual-modal multi-scale feature pyramids; enhancing feature interaction through a cross-gated large-kernel attention fusion module using cross-modal gating and decomposed large-kernel attention; reconstructing spatial geometric structure and frequency domain texture using a space-frequency perception module employing strip convolution and stationary wavelet transform; and training the network end-to-end based on a multi-task joint loss function to output the target category and bounding box. This application enhances the representation of small target features while preserving the unique semantics of each modality, significantly improving the accuracy and robustness of anti-drone detection in complex environments.

The invention relates to a digital archive OCR processing method based on adaptive enhancement, and belongs to the technical field of computers. The method comprises the following steps of: decomposing an input image into four sub-bands by adopting lifting wavelet transform (LWT), and performing up-sampling and reconstruction on the high-frequency sub-bands by adopting a bicubic interpolation method; decomposing an input image into four sub-bands by adopting stationary wavelet transform (SWT); adding the two groups of high-frequency sub-bands direction by direction to realize complementary fusion of multi-scale features; finally, gamma-times interpolation reconstruction is carried out on the high-frequency sub-band fusion result, inverse lifting wavelet transform ILWT is applied to reconstruct an image with the original resolution, and a preliminary fusion image is obtained; and inputting the input image and the preliminary fusion image into a generative adversarial network (GAN) model for further fusion to form a high-quality text image. According to the method, the background noise is effectively suppressed while the character stroke edge definition is ensured, so that the image is optimally balanced in the aspects of brightness, contrast ratio, texture fidelity and the like.

The invention relates to the field of ocean monitoring and prediction, and discloses a buoy trajectory prediction method based on stationary wavelets and wind-flow-position interaction, and the method comprises the steps: obtaining a trajectory sequence of a buoy in a historical time period and corresponding ocean environment field data, and carrying out the standardization processing; decomposing the normalized trajectory sequence of the buoy through stationary wavelet transform to obtain a plurality of subsequences of different scales, and then obtaining a plurality of segmented token sequences through dynamic segmentation; performing feature extraction on each segmented token sequence after feature mapping through a wind-flow-position interaction feature extraction mechanism; and reconstructing interaction features output by the wind-flow-position interaction feature extraction mechanism by using stationary wavelet inverse transformation, and finally outputting a trajectory coordinate predicted value of the buoy in a future time period after linear layer mapping. According to the method, by introducing wavelet multi-scale decomposition and dynamic segmentation, the model can explicitly capture trends and details in the trajectory, and the non-stationarity can be effectively dealt with.

The invention belongs to the technical field of electric field monitoring signal noise reduction processing, and provides an improved algorithm-based electric field intensity time domain monitoring signal noise reduction method, which comprises the following steps of: improving a salmons swarm optimization algorithm by adopting improved Sine chaotic mapping, a spiral search strategy and a salmons alert mechanism integrating sparrow alert and krill random diffusion; the method comprises the following steps of: applying an improved salat swarm optimization algorithm to accurate self-adaptive selection of key parameters in variational mode decomposition, substituting a selected optimal parameter combination into a variational mode decomposition algorithm, and performing self-adaptive decomposition on an original electric field intensity signal to obtain intrinsic mode components with different center frequencies; then, each modal component is constructed into a Hankel matrix, and signal principal component enhancement and noise separation are realized through a singular value decomposition technology; and finally, carrying out multi-scale fine noise reduction on the processed component by adopting stationary wavelet transform, and obtaining a denoised electric field intensity time domain monitoring signal through signal reconstruction. According to the method, the signal smoothness and the signal-to-noise ratio are remarkably improved, key electric field characteristics are completely reserved, more reliable signal support is provided for industrial process monitoring, and the method has important engineering application value.

The invention discloses a comprehensive evaluation method for water quality of a penaeus vannamei pond based on deep learning. The method comprises the following steps: firstly, acquiring multi-dimensional water quality monitoring data of a penaeus vannamei aquaculture pond; carrying out preprocessing operation on the obtained original data; then, stationary wavelet transform is executed on the processed water quality time sequence data, multi-scale characteristic decomposition and denoising are achieved, and low-frequency trend characteristics and high-frequency disturbance characteristics of water quality signals are extracted; and finally, inputting the features subjected to wavelet decomposition into a pre-trained deep learning model, extracting spatial features through a convolutional neural network, modeling a time dependency relationship through a long-term and short-term memory network, and obtaining five types of water quality grade evaluation results through a Softmax output layer. Based on the method, automatic, refined and real-time comprehensive evaluation of the water quality of the aquaculture pond can be achieved, the accuracy and stability of water quality evaluation are improved, and effective technical support is provided for intelligent monitoring and health management of the aquaculture environment of the penaeus vannamei boone.

The application discloses a kind of enhancement methods of terahertz near-field radar image, comprising the following steps: the data conversion of the terahertz near-field radar image to be enhanced is normalized, and normalized terahertz near-field radar image is obtained;The normalized terahertz near-field radar image is decomposed using stationary wavelet transform, and a plurality of low-frequency subbands and a plurality of high-frequency subbands are obtained;The high-frequency multi-direction filter of high-frequency subband is carried out using different direction filters, and a plurality of enhanced high-frequency subbands are obtained;The low-frequency adaptive linear enhancement of low-frequency subband is carried out using the mean and standard deviation of low-frequency subband, and a plurality of enhanced low-frequency subbands are obtained;The inverse stationary wavelet transform is carried out to a plurality of enhanced high-frequency subbands and a plurality of enhanced low-frequency subbands, and reconstructed terahertz near-field radar image is obtained;Linear transformation is carried out to reconstructed terahertz near-field radar image, and enhanced terahertz near-field radar image is obtained.The application can enhance contrast, highlight edge features and rich detail information.

This invention discloses a lightweight long-term time series prediction method and system for industrial time series data. The method first maps the input multivariate time series to a high-dimensional feature space and decouples the input features into a coarse-grained trend term and a residual term containing fine-grained features. For the residual term, a stacked stationary wavelet transform is used to convert the time-domain signal to a multi-scale frequency domain, and a geometrically sparse attention mechanism is designed. This mechanism calculates the geometric correlation between variables by combining dot products and wedge products, and then uses a sparsity strategy to highlight key dependencies and reduce computational complexity. For the trend term, it is converted from the time domain to the frequency domain, and complex weight parameters are introduced to obtain time-domain features. After convolution processing of the time-domain features, spatial domain features are obtained. The spatial domain features are fused with the time-domain features to obtain the trend term output. Finally, the extracted features of the residual term and the trend term are fused and mapped to obtain the final prediction result.