23 results about "Soft thresholding" patented technology

This invention discloses an underwater acoustic anti-interference transmission method and system based on sparse time-frequency feature mapping. The method includes: at the transmitting end, adaptively generating a fractional-order linear frequency-modulated waveform with a specific time-frequency shear slope based on the Doppler state of the underwater acoustic channel, achieving physical-layer focusing of transmitted energy; at the receiving end, constructing a hybrid observation model containing wide-block sparse channel components and narrow-block sparse burst noise components, and using a dual-channel variational Bayesian algorithm to jointly iteratively infer the posterior probability distribution of environmental burst noise in the fractional-order domain; finally, recovering the original signal through soft-threshold interference cancellation and fractional-order channel equalization. This invention effectively solves the communication failure problem caused by high-dynamic Doppler diffusion and marine biological impulse noise interference in underwater acoustics by actively mapping the waveform and using heterogeneous sparse joint inference at the receiving end, significantly improving transmission reliability in harsh underwater acoustic environments.

The application discloses a water acoustic signal time-frequency spectrum line spectrum extraction method and system, and belongs to the technical field of signal processing. A noisy time domain signal is generated through simulation, and a mask label of a time-frequency spectrum of the noisy time domain signal belonging to a line spectrum is generated through a soft threshold function; a denoising model is trained according to the time-frequency spectrum and the mask label; a target water acoustic signal is acquired, the time-frequency spectrum of the target water acoustic signal is input into the denoising model, and a mask label corresponding to the time-frequency spectrum input is output through inference; the time-frequency spectrum of the denoised target water acoustic signal is acquired according to the mask label and the time-frequency spectrum; an initial candidate point set of the time-frequency spectrum of the denoised target water acoustic signal is acquired, and an initial candidate point of a current frame time-frequency spectrum in the initial candidate point set is acquired; a correlation cost matrix is constructed, the initial candidate point and a trajectory are correlated and matched with the minimum difference as a target, and a line spectrum of the trajectory and the candidate point dynamic correlation is acquired. The method can balance denoising fidelity, detection accuracy and real-time performance.

This invention discloses an improved method for generating driver fatigue EEG data using a latent diffusion model, belonging to the field of EEG signal processing. The method includes: processing the original multi-channel driver fatigue EEG signal through multi-scale wavelet denoising, adaptive thresholding, and an improved soft thresholding function; then performing inter-channel covariance alignment and whitening to eliminate redundancy to obtain a preprocessed signal; performing short-time Fourier transform and logarithmic energy normalization to obtain normalized time-frequency features; inputting the time-frequency features into an encoder to obtain the mean and variance of latent variables, and obtaining latent variables through reparameterized sampling; reconstructing the time-frequency features using a decoder combined with fatigue state labels; training a conditional variational autoencoder by minimizing reconstruction loss and KL divergence loss; adding noise through forward diffusion in the latent space; training a denoising network to remove noise based on fatigue state labels during reverse denoising; inputting the denoised latent variables into the decoder, combining them with a specified fatigue state to generate new time-frequency features, and reconstructing them into the target EEG signal. This invention can enhance the training dataset and improve the accuracy of driver fatigue monitoring.

This invention discloses a high-resolution range image feature preservation and enhancement method based on a deep unfolded network, comprising: preprocessing radar echo data into an initial feature tensor; constructing a low-rank sparse decomposition model and mapping the iterative solution process of the model to a deep neural network containing cascaded networks, with each layer corresponding to one iterative optimization step; in each layer, a low-rank near-end mapping module captures the global correlation between the range dimension and the Doppler dimension through a dual-path axial attention mechanism to update the low-rank clutter component matrix; a sparse near-end mapping module updates the sparse target component matrix using a complex soft thresholding operator to maintain target phase information while suppressing clutter; and a Lagrange multiplier update module is used to update the Lagrange multipliers. This invention overcomes the limitations of manual parameter tuning by introducing a deep neural network and a dual-path axial attention mechanism, enabling the acquisition of structurally complete and phase-accurate high-resolution range image features while effectively suppressing clutter.

This invention belongs to the field of computer vision and artificial intelligence technology, and discloses a sonar image target detection method and system based on soft thresholding and attention mechanisms. It extracts features from sonar images to obtain multi-scale feature maps, then fuses these multi-scale feature maps to obtain a fused feature map. A result prediction module then uses this fused feature map for detection, obtaining predictions of the target category and bounding box location. This invention performs dual optimization at the feature and loss function levels, reducing false alarms caused by noise and improving the detection accuracy and localization precision of small targets in sonar images without changing the image preprocessing workflow. It exhibits strong robustness and practical value.

This application relates to a SAR sparse imaging method and system based on a dual-channel depth unfolding network. The method includes updating image features using a physical observation matrix; dual-channel parallel feature mapping, thereby focusing on the sparse recovery of strong scattering points and compensating for detail loss caused by soft thresholding; fusing the outputs of the two channels to obtain the final output of the layer; and network training and imaging. This application can achieve high-quality reconstruction of three-dimensional targets under sparse observation conditions through dual-channel fusion.

The application discloses a kind of based on optimization's guiding filter image fusion change detection method and device, method includes: by extreme minimum scale difference operator, pixel value difference value is to the picture of same scene different phase, obtains difference map;Using the way of guiding filter, difference map is fused on global scale, to obtain the optimal difference map;Mean filter is used to phase map I1 and I2, and the base layer difference map corresponding to each difference map is obtained, and the optimized brightness saliency map F1 and F2 are obtained by pca fusion as the image of generating saliency map, according to the principle of maximum saliency to determine weight map;Weight map is regularized for double-scale difference map reconstruction, and the final difference map is obtained;In clustering segmentation stage, by introducing soft threshold function, the final difference map is further processed, to suppress existing abnormal noise.The device includes: processor and memory.

This invention discloses a pruning method, apparatus, and electronic device for spiking neural networks. The method includes: initializing an initial vector composed of the weights of each connection in the synaptic connection layer to obtain a weight vector; when pruning the spiking neural network using a backpropagation-based algorithm, calculating the gradient of each loss function value with respect to the hidden parameter vector using a predefined derivative function; updating the gradient of the hidden parameter vector using gradient descent, and calculating a target threshold for subsequent gradient updates using a preset incrementing function; based on the target threshold, mapping the hidden parameter vector back to the weight vector using the soft threshold function; and obtaining a trained spiking neural network model when the number of pruning training rounds reaches a preset number. This invention solves the technical problem of effectively deploying spiking neural networks on neuromorphic computing chips in related technologies.

The application discloses a wide field of view video image change detection method and device, and the method comprises the following steps: using an improved adaptive fast guided filter for video image change detection; after filtering the multi-temporal video image, a logarithmic ratio operator considering neighborhood information is proposed, the mean ratio difference graph generation mode is improved, and a difference graph is generated; the improved MR image and the improved LR image are subjected to image fusion through discrete wavelet transform, and a fused difference graph is obtained; a soft threshold function is used to perform initial classification on the fused difference graph, and an initial change area and an unchanged area are obtained; a cumulative distribution function is used to compress the pixel value of the classification result from [0, 255] to [0, 1]; a super-fast robust constraint fuzzy C-Means clustering algorithm is proposed, and an improved adaptive median filter is used for denoising processing. The device comprises a processor and a memory.

The application discloses a kind of multi-dimensional self-adaptive streaming reconstruction methods of spectrum situation tensor for space-based dynamic incomplete observation, comprising: based on spectrum measurement data, construct dynamic tensor and determine dimension evolution mode by the spatial coverage dimension variation of dynamic tensor;When dimension evolution mode is expansion or invariable, using historical factor matrix as prior constraint to process current observation tensor, obtain the factor matrix of current time slot;When dimension evolution mode is short, using time series prediction method to pre-fill blind area, obtain the factor matrix of current time slot;Based on factor matrix, reconstruct spectrum situation tensor, and separate abnormal interference from reconstruction residual using soft threshold operator;Based on the reconstruction residual after separating abnormal interference, calculate relative change rate, update tensor rank in combination with sliding window regression trend;Based on the tensor rank after updating, adjust factor matrix structure, use the adjusted factor matrix for the processing of next time slot, iterate until the reconstruction result converges.

The application discloses a kidney tumor segmentation method based on multi-scale feature extraction, adopts a mainstream encoder-decoder architecture, proposes a feature refinement module in the encoder path, divides and fuses the features of each stage on the channel, so that more fine-grained local features can be extracted on the encoder path, which is very helpful for the target with large size change such as kidney tumor; meanwhile, a multi-scale feature extraction module is proposed in the decoder path, the receptive field can be expanded without introducing too much memory consumption by nesting a small U-shaped network in the path, and the global features can be extracted more effectively, in addition, the multi-scale feature extraction module also includes a soft threshold denoising module, the soft threshold denoising module can automatically select a threshold to eliminate the noise-related information, so that more discriminative features can be extracted.

The application relates to the technical field of fault diagnosis, in particular to a method for diagnosing faults of outrigger lines of an aerial work platform, which comprises the following steps: acquiring various types of electrical signals of each outrigger line of the aerial work platform in real time; decomposing the various types of electrical signals into a plurality of IMF components, and decomposing each IMF component into a plurality of wavelet signals; constructing a noise interference confidence degree based on the degree of noise interference on each type of electrical signal of each outrigger, the random fluctuation degree and the proportion of prominent sharp peaks of each layer of wavelet signals, and the correlation degree between each IMF component and the original electrical signal, so as to optimize the soft threshold function in the wavelet threshold denoising algorithm, and then obtain filtered various types of electrical signals, and then judge whether faults occur in each outrigger line in the current monitoring period. The soft threshold function in the wavelet threshold denoising algorithm is optimized, the denoising effect of the electrical signal is improved, and the accuracy of the self-diagnosis of the outrigger line fault is improved.

This invention relates to a method for constructing a terrain navigability cost map based on cosine soft thresholding and probabilistic fusion, belonging to the field of robot environmental perception and map building technology. It includes elevation grid construction, step height feature extraction, local slope feature extraction, roughness feature extraction, cosine soft thresholding, probabilistic OR fusion, visual semantic-assisted correction, and cost map generation. Advantages include: cosine soft thresholding allows the cost map to reflect terrain difficulty in a gradient manner; probabilistic OR fusion enables multi-dimensional safety-priority fusion; semantic-geometric joint correction reduces the terrain misclassification rate from approximately 18% to below 5%; the parameterized design of six physical parameters adapts to different robot platforms; the output is compatible with the ROS costmap_2d interface; and zero-derivative boundary conditions eliminate cost jumps in critical regions.

This invention discloses a method and system for identifying the corrosion state of wind turbine bolt materials. The method includes: acquiring raw signals using a ring ultrasonic transducer array in full-matrix acquisition mode; obtaining focused echo signals through full-focusing post-processing; performing wavelet packet decomposition and soft-threshold denoising on the focused echo signals to extract the energy entropy of each sub-band; concatenating the raw time-domain signal, sub-band reconstructed signal, and energy entropy into a multi-channel feature sequence; inputting the sequence into a deep convolutional neural network combined with a channel attention mechanism to extract corrosion feature semantic vectors; acquiring load and environmental data and mapping them to external feature vectors, fusing them with the corrosion feature semantic vectors through tensor product fusion to obtain a dynamic embedding vector; constructing the dynamic embedding vectors of the same bolt at different times as an input sequence, and outputting the corrosion level, residual strength ratio, and remaining life in parallel through a bidirectional long short-term memory network. This invention can improve the accuracy and multi-dimensional quantitative assessment capability of identifying the corrosion state of wind turbine bolts under complex and harsh working conditions.

This invention discloses a filtering method, device, storage medium, and underwater vehicle based on forward-looking sonar. The filtering method employs the following steps: sonar physical detection area modeling and mask mapping; first-level global adaptive benchmark extraction and coarse filtering; acoustic attenuation sensing and radial block partitioning; second-level non-uniform sensitivity local fine filtering; spatial smoothing fusion and image reconstruction, outputting a denoised forward-looking sonar image. By cleverly combining an acoustic radial distance model with the local statistical features of a single-frame image, and through pure mathematical cascade matrix operations and adaptive soft thresholding, a perfect balance between reverberation suppression and weak target protection is achieved with extremely low computational overhead. It preserves the features of weak targets in the far field, solves the denoising distortion problem caused by drastic changes in the dynamic range of sonar images, reduces the temporal and spatial complexity of the algorithm, ensures real-time deployment at the edge of the underwater unmanned platform, and ensures smooth edges and no geometric distortion in the denoised image, providing high-fidelity input for subsequent target recognition and path planning.

This invention provides a noise reduction method for machine tool spindle rotation signals based on VMD-PE-MWSTD, belonging to the field of precision measurement technology. The method includes: first, acquiring and preprocessing the rotation signal of the machine tool spindle; then, using Variational Mode Decomposition (VMD) to decompose the signal into several Intrinsic Mode Function (IMF) components; next, calculating the permutation entropy (PE) of each IMF component, and automatically identifying and filtering noise-dominated IMF components and useful signal-dominated IMF components based on a preset PE threshold; processing the filtered noise-dominated IMF components using an improved wavelet soft thresholding (MWSTD) method to remove noise while retaining the effective components in the signal; finally, reconstructing the denoised IMF components and the useful signal-dominated IMF components to obtain the final denoised rotation signal. This invention effectively separates signal and noise, significantly improves the signal-to-noise ratio, and retains the true characteristics of the signal, providing a high-quality data foundation for machine tool spindle accuracy monitoring and health diagnosis.

This invention discloses an intelligent analysis method for the operating status of ring main units based on edge computing, belonging to the field of power system automation. It involves deploying high-frequency electromagnetic wave sensors and environmental sensors to synchronously collect partial discharge signals and temperature and humidity parameters. At the edge nodes, dynamic soft thresholding denoising based on wavelet transform and sparse decomposition of overcomplete atomic libraries are sequentially performed to achieve interference filtering and data compression. Subsequently, a lightweight network constructed using depthwise separable convolution and channel attention is used to extract state feature vectors. Based on the gas discharge mechanism, a diagonal environmental influence correction matrix is ​​constructed with temperature and humidity deviations and their cross-coupling terms as independent variables. This matrix performs nonlinear correction on the feature vectors to remove feature drift caused by environmental fluctuations at its root. Finally, a classifier outputs the operating status category, thereby achieving highly reliable closed-loop real-time diagnosis on the edge side. This significantly improves the accuracy of fault identification under complex climatic conditions and greatly reduces the false alarm rate.

This invention discloses an adaptive Doppler signal noise reduction filter, comprising: a sensor interface impedance matching and electrical isolation unit, an analog-to-digital conversion and Doppler signal preprocessing unit, a Doppler signal multi-scale wavelet decomposition hardware unit, a Doppler background noise statistical characteristic estimation unit, a hierarchical adaptive threshold hardware calculation and nonlinear correction unit, a wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit, a noise reduction effect quantitative evaluation and parameter closed-loop optimization unit, and a power management module. This invention solves the reconstruction oscillation problem of the hard threshold function and the constant amplitude deviation defect of the soft threshold function by dynamically adjusting the hierarchical adaptive threshold based on the hyperbolic tangent square operator and using a fully hardware-based design. It filters out noise while completely preserving the amplitude and phase information of the Doppler signal, achieving high-fidelity noise reduction with low hardware resource consumption and meeting low-latency processing requirements in various scenarios.

This invention discloses a hyperspectral / multispectral chlorophyll-a inversion method based on feature-level residual learning. First, a regularized regression model is trained using multispectral broadband features and measured chlorophyll-a ground truth values. The difference between the ground truth and predicted values ​​is used to construct a residual error vector. Then, using this residual as the target, a multivariate dimensionality-reduction regression model is trained using hyperspectral high-frequency derivative features. The multispectral feature matrix of each pixel in the target water body is input into the trained regularized regression model to obtain the baseline predicted value of chlorophyll-a concentration for each pixel. The hyperspectral derivative feature matrix of each pixel is input into the trained multivariate dimensionality-reduction regression model to obtain the high-frequency residual correction value for each pixel. Finally, the two values ​​are summed and post-processed using a hyperbolic tangent soft thresholding function to obtain the final predicted value of chlorophyll-a concentration, and the inversion map is output. This invention achieves a fundamental breakthrough in the multi-source fusion paradigm, possesses strong physical noise resistance, and improves the accuracy of high-concentration chlorophyll inversion.

The application discloses a double-end traveling wave fault location method based on wavelet transform and Teager-Kaiser energy operator (TKEO), which is applied to a 800V direct current hydrogen production line. In order to improve the accuracy of fault identification and the precision of fault location, firstly, wavelet transform is performed on a fault signal, a soft threshold is used for denoising and the signal is reconstructed; characteristic information of each wavelet decomposition layer is extracted and analyzed, and the fault type is determined according to the energy ratio of high-frequency and low-frequency decomposition layers; secondly, TKEO is used to extract the instantaneous energy spectrum after wavelet decomposition, and the sampling points of the first wave head reaching the two ends of the direct current line are accurately calibrated; finally, the double-end ranging method is used to accurately solve the fault distance.

The application relates to the technical field of nondestructive testing, and discloses a steel structure weld defect intelligent detection method, which comprises the following steps: collecting a weld area A-scan echo sequence signal by using an ultrasonic probe; performing adaptive noise reduction processing on the signal based on wavelet packet transformation, obtaining pure echo by using a minimum Shannon entropy criterion and soft threshold denoising; converting the pure echo into a two-dimensional time-frequency spectrum by using continuous wavelet transformation; inputting the time-frequency spectrum into a preset convolutional neural network model, extracting multi-layer convolutional features, and outputting a defect type through classification; and calculating the depth and horizontal position of the defect in the weld according to the sound path time of the pure echo signal and the probe parameters. The application combines signal adaptive enhancement with time-frequency image depth learning, effectively overcomes structural noise interference, and realizes automatic high-precision identification of the weld defect type and accurate positioning of physical coordinates.

This invention relates to a channel estimation method based on a Time-Frequency Joint Sparse Attention Network (TF-JSANet) in OFDM systems. The method first obtains a preliminary estimate of the pilot position at the receiver using the least squares (LS) criterion, forming a low-resolution noisy input. Then, a TF-JSANet network structure is constructed, extracting channel features in the time and frequency dimensions through parallel time-domain and frequency-domain branches, respectively. An Adaptive Sparse Attention Module (ASAM) is introduced to achieve dynamic soft-threshold denoising and feature enhancement. Finally, an upsampling module reconstructs the complete channel state information. Simulation results show that this method outperforms traditional algorithms and existing typical deep learning channel estimation methods, demonstrating good generalization ability and robustness.

This invention discloses a multi-timescale fault detection and isolation method for integrated navigation systems. Targeting integrated navigation systems employing a federated filtering structure, the method acquires state estimation information for each local filtering branch, constructs a residual quantity reflecting changes in consistency between branches, and normalizes this residual by combining it with corresponding uncertainty information to obtain a standardized deviation. Normal fluctuations are suppressed through soft thresholding, and non-negative effective deviations are extracted. Based on the effective deviations, fast-timescale and slow-timescale statistics are constructed respectively, and the impact of operational condition changes on the detection results is mitigated through baseline updates and platform removal processing. Finally, fault detection and isolation conclusions are output based on the joint decision results of the fast and slow-timescale statistics. This invention can balance rapid response to abrupt faults with cumulative perception of gradual faults, and is applicable to local branch fault detection and isolation in multi-source integrated navigation systems.