16results about How to "Improve sparsity" patented technology

The application discloses a machine learning algorithm for screening of metal elements in a diatomic catalyst full cycle based on product yield prediction and application thereof, and the algorithm comprises the following steps: (1) collecting attribute features of metal elements in the diatomic catalyst as a feature space 1; (2) preliminarily screening the attribute features according to comprehensive scores of feature importance to obtain a feature space 2; (3) performing feature operation processing on the feature space 2 to generate a multivariate descriptor space matrix; (4) performing dimension reduction processing on the multivariate descriptor space matrix by combining a gradient boosting regression model with a recursive feature elimination algorithm; (5) further screening to obtain an interpretable descriptor by using a least absolute shrinkage and selection operator; and (6) applying the interpretable descriptor to a random forest regression model to screen a catalyst for a propane dehydrogenation reaction to prepare propylene. The application can solve the problem that a traditional descriptor is difficult to directly predict a catalytic product yield.

The invention discloses a virtual multi-view fusion millimeter wave point cloud imaging method, which belongs to the field of millimeter wave radars, and realizes three-dimensional point cloud imaging of a target by constructing a virtual observation view angle and fusing perception data of different view angles: establishing a multi-view synthetic aperture radar imaging model; a single-bit compressed sensing framework is introduced to construct a sparse imaging optimization problem, an accurate relay angle is solved by minimizing a model error, and a relay plane model error is corrected; and high-precision 3D point cloud imaging of the region of interest is realized based on the accurate angle. According to the method, a three-dimensional point cloud of a target is constructed by fusing virtual visual angle detection shielding characteristics formed by reflection of a plurality of relay surfaces, joint estimation of relay surface angle prior and 3D point cloud is realized by a nested iteration method, and artifacts and offset caused by prior errors are eliminated; the sparsity and the target continuity of the three-dimensional point cloud can be improved through the additional composite norm constraint, and the accuracy of virtual multi-view 3D point cloud imaging is effectively improved.

This invention relates to a power data sparsity compression observation method for non-intrusive load monitoring, comprising the following steps: Step 1, acquiring power monitoring data and determining its electricity consumption behavior pattern; Step 2, initializing wavelet sparse basis for sample data based on simple power behavior patterns; Step 3, based on the determination results of the electricity consumption behavior pattern of the completed training samples and the results after wavelet basis initialization in Steps 1 and 2, performing improved K-SVD training and generating sparse basis; Step 4, training the sample dataset Y based on Step 3 to obtain a sparse dictionary D, and performing power monitoring data compression observation based on the improved sparse basis.

This invention discloses a text summarization method and system based on sparse attention acceleration, comprising: reading long text data, performing word segmentation and embedding encoding processing, extracting sequence feature vectors, and mapping them to a query matrix Q, a key matrix K, and a value matrix V; constructing a summarization generation network, including a sparse attention calculation module, a feedforward calculation module, a prediction head module, and a key-value caching module; inputting the query matrix Q, key matrix K, and value matrix V into the summarization generation network, first passing through a backbone network composed of multiple stacks of the sparse attention calculation module and the feedforward calculation module, and then being processed by the prediction head module, while the key-value caching module caches historical decoding states in real time to obtain an initial text feature vector; based on the initial text feature vector, performing an autoregressive decoding process through the summarization generation network to output the final summary result. This invention can accelerate the decoding process and ensure the semantic integrity and coherence of the generated summary.

This invention belongs to the field of radar signal processing technology and discloses a method for detecting small sea targets based on optimized feature mode decomposition, including: S1: acquiring the signal data to be detected; S2: decomposing the original signal into several modal components using FMD, and selecting envelope spectral entropy as the fitness function; S3: using the SOS algorithm to globally optimize the fitness function in FMD; S4: introducing the PSO algorithm to locally optimize the key parameters of FMD; S5: retaining components with low envelope spectral entropy values ​​and correlation coefficients greater than a threshold; S6: extracting envelope spectral entropy and frequency band energy proportion features from the selected modal components, introducing the Gini coefficient as a weighting factor, and constructing GSEBE joint features; S7: inputting the envelope spectral entropy value into a DELM classifier with controllable false alarm rate, and achieving target detection based on the comparison between the predicted value and the decision threshold. This invention enhances the ability to distinguish between sea clutter and target echoes, achieving more accurate classification and detection.

The invention relates to a health state recognition method based on image statistical characteristics. Comprising the following steps of: firstly, acquiring an original vibration signal of a bearing by using a sensor, converting the original vibration signal into a time-frequency domain color image by adopting short-time Fourier transform, and then converting the time-frequency domain color image into a grayscale image by using grayscale transform; secondly, extracting six types of image statistical characteristics from the grayscale image, wherein the six types of image statistical characteristics comprise a mean value, a variance, a contrast ratio, correlation, energy and homogeneity; then, six types of image statistical features are used as input, and a fusion feature vector is output through a designed stability-enhanced sparse PCA (Principal Component Analysis) module; and finally, performing health state recognition on the fused feature vector by using a state change detection function. The method is good in recognition effect and high in industrial scene adaptability, and provides high-precision and high-reliability technical support for predictive maintenance of equipment; the method is high in calculation efficiency, good in real-time performance and high in robustness, unification of precision, efficiency and robustness is achieved, and the method is very suitable for online real-time monitoring application in engineering practice.

This application provides a training method, a spiking neural network, and an apparatus for a spiking neural network. The method includes: iteratively training the spiking neural network using training data of the object to be processed, wherein the spiking neural network contains at least one neuron model; calculating a task loss function based on the training results of the spiking neural network and updating the weight parameters; updating the current discharge threshold parameter of the neuron model based on the historical discharge frequency of the neuron model, wherein the value of the current discharge threshold parameter increases with the increase of the historical discharge frequency; and obtaining the trained spiking neural network model using the updated weight parameters and the updated discharge threshold parameter after multiple iterations of training. This method facilitates the training of a sparsely discharging spiking neural network, reduces the energy consumption during spiking neural network applications, and improves model energy efficiency by using the trained spiking neural network model to process the data of the object to be processed.

The application discloses an image rain removing method and device for unmanned aerial vehicle power grid inspection, and belongs to the technical field of unmanned aerial vehicle aerial image rain removing.The image rain removing method provided by the application comprises the following steps: acquiring an aerial image collected by an unmanned aerial vehicle, performing step-by-step information mining operation on the aerial image to obtain a denoising sub-image; in at least one information mining operation process, a spatial dimension reduction operation and a spatial dimension expansion operation are performed in sequence; and a target image with rain streaks removed is constructed based on the denoising sub-image, so that the image rain removing operation is completed.In the feature information mining process of the aerial image, the spatial dimension reduction operation is performed first, the content of rain streak noise is greatly compressed, and the influence on other objects in the image is relatively small, then the spatial dimension expansion operation is performed, the sparseness and the discrete degree of residual rain streak information are improved, the rain streaks are efficiently removed, and other image features are well reserved.

This application provides a source code generation method and apparatus. The method includes: determining target features based on a first fragment, wherein the first fragment includes one or more codes input by a user before a first moment, and the target features are used to indicate code attributes of the codes in the fragment; generating multiple candidate fragments; and selecting a second fragment from the multiple candidate fragments based on the target features, wherein the features of the second fragment have the highest similarity to the target features among the features of the multiple candidate fragments. The target features are obtained based on the source code input by the user and serve as the user's sequence preference. The source code suggestions provided to the user are generated based on the user's sequence preference. In other words, the generated source code is related to the source code input by the user. Therefore, the accuracy of the generated source code can be improved.

This invention relates to the field of remote sensing image compression technology, specifically to a remote sensing image compression method based on dual-tree complex wavelet convolution and a frequency dictionary entropy model. The method includes: compressing remote sensing images using a remote sensing image compression model based on dual-tree complex wavelet convolution and a frequency dictionary entropy model; the remote sensing image compression model includes a dual-tree complex wavelet convolution module, a frequency dictionary entropy model module, and a residual feature extraction module; the dual-tree complex wavelet convolution module is used to perform downsampling and upsampling feature processing on the remote sensing image, effectively removing frequency domain redundancy in the latent representation; the frequency dictionary entropy model module is used to perform frequency division on the obtained latent representation of the remote sensing image to establish a probability model; the residual feature extraction module is used to extract features from the remote sensing image, effectively capturing long-range contextual information. This invention enhances the model's image compression effect by introducing dual-tree complex wavelet convolution and a frequency dictionary entropy model, significantly improving the fidelity of remote sensing images at high compression ratios.

A low-complexity OTFS channel estimation method based on windowing processing, a storage medium and a computer are provided, which solve the problems of high complexity and large pilot overhead of the cross-correlation channel estimation method. The method comprises the following steps: constructing a Doppler axis function in a time-delay-Doppler domain equivalent channel processed by a window function, and performing interception to obtain a matching function; setting a threshold coefficient, a Doppler axis resolution, an initial path cycle number and a pilot area; obtaining an estimated time-delay-Doppler domain impulse response matrix according to the pilot area position relationship; performing a cross-correlation operation on the matching function and the impulse response matrix according to the Doppler axis resolution to obtain a maximum cross-correlation value, a Doppler grid point value, a time-delay grid point value and a cross-correlation value thereof; estimating corresponding channel parameters; updating all values on the lth time-delay grid point of the impulse response matrix; setting an iteration stopping condition, and outputting the estimated value when the stopping condition is met. The method is applied to the field of high-speed mobile communication.

The invention discloses an improved BM3D image denoising method based on multi-element fusion block similarity measurement. According to the method, wavelet decomposition is carried out on a noise image, a noise standard deviation is stably estimated on a highest-frequency sub-band, BM3D parameters are initialized, in a basic estimation stage, pixel intensity, local variance and gradient difference are fused to construct block similarity measurement, similar block matching grouping and three-dimensional block group construction are realized, and the method has the advantages of being high in robustness and high in robustness. And in a final estimation stage, a basic estimation image is obtained through three-dimensional collaborative hard threshold filtering and weighted aggregation, in the final estimation stage, secondary matching is guided by the basic estimation image, a three-dimensional block group is constructed again, a structure and texture adjustment factor is introduced to form an adaptive Wiener coefficient for Wiener collaborative filtering, and finally, a target de-noised image is output through weighted aggregation. In the whole process, the phenomena of mismatching and over-smoothing of a traditional BM3D algorithm under the condition of high noise can be reduced, the edge and texture detail keeping capability is enhanced, and the denoising quality of a noise image is improved.

This invention relates to the field of image processing and remote sensing analysis technology, and provides a robust deep non-negative matrix factorization method for hyperspectral image unmixing. The method includes: decomposing hyperspectral data into a product of multiple non-negative low-rank matrices using a multi-level non-negative matrix factorization method; representing the similarity and differences of data in each non-negative low-rank matrix based on a dual-graph adversarial learning mechanism; constraining the Gram matrix during the decomposition process using inner product-based structural sparsity regularization; and constructing a comprehensive optimization model to obtain the unmixing results of the hyperspectral data. This invention represents hyperspectral data as multiple non-negative low-rank matrices, and introduces adversarial graph regularization terms, hierarchical sparsity constraints, and truncated activation mechanisms to improve the robustness and expressive power of the model. Finally, it utilizes the alternating direction multiplier method (ADMM) for efficient optimization, enabling the proposed method to accurately extract endmembers and abundance in complex noisy environments, achieving robust hyperspectral image unmixing.

The application provides a fast phase unwrapping method based on tail non-convex regularization, and aims at solving the following two technical problems of the existing phase unwrapping algorithm: one is that details are lost due to the discontinuity of the image edge; and the other is that it is difficult to accurately realize phase unwrapping when facing high complexity or large phase jumps. The steps of the application are as follows: based on the wrapped image, a non-convex tail optimization model with box constraint is constructed; the non-convex tail optimization model is solved by using a completely split primal-dual algorithm under the condition that the support set is empty, so that an initial phase is obtained; errors in the initial phase estimation are corrected through tail minimization, and the support set is dynamically updated, so that an unwrapped image is obtained. Compared with the traditional phase unwrapping algorithm, the application combines the non-convex regularization with the tail minimization strategy, so that the accuracy and robustness of the phase unwrapping are improved, and the application has significant superiority in key evaluation indexes such as signal-to-noise ratio and structural similarity.

Embodiments of the present application provide a hybrid pruning method and device for a neural network model, a storage medium and a program product. The method can be applied in the field of artificial intelligence. The method comprises: obtaining first weights of a neural network model, the first weights being weights obtained by pre-pruning; and performing online pruning on the first weights according to statistical characteristics of current online activation data of the neural network model to obtain second weights of the neural network model. The method improves the sparsity of the neural network model, thereby reducing the computational complexity of the neural network model.

This invention relates to the field of wireless communication and signal processing technology, specifically to a shortwave spectrum high-efficiency sampling method and system based on compressed sensing. The method includes: acquiring power spectral density data across the entire frequency band using a low-speed scanning receiver; determining a set of strong interference frequency points; generating a binary null observation sequence; loading the binary null observation sequence onto an analog front-end mixer circuit to perform analog domain orthogonal mixing and integration processing on the input RF signal, outputting voltage sample values; calculating signal sparsity coefficients using a compressed sensing reconstruction algorithm; calculating the reconstruction residual based on the voltage sample values ​​and the signal sparsity coefficients; triggering a re-acquisition process of power spectral density data in response to the reconstruction residual exceeding a preset error threshold; and outputting the signal sparsity coefficients in response to the reconstruction residual being less than or equal to the preset error threshold. This invention solves the problems of limited dynamic range and weak signal overload, significantly improving the receiving sensitivity of weak signals under strong interference environments.