13 results about "Variational regularization" patented technology

The invention discloses a hyperspectral anomaly detection method and system based on tensor multi-subspace learning. The method comprises the steps of obtaining hyperspectral image tensor data of a to-be-detected region; carrying out band-by-band normalization processing; decomposing into an abnormal tensor, a noise tensor and a structure background component; non-convex tensor correlation total variation regularization is designed; constructing an iterative sparse weight tensor; structuring-norm is adopted; constructing a robust dictionary tensor; establishing an anomaly detection model; optimizing by adopting an effective iterative updating algorithm based on an alternating direction multiplier method to obtain an optimal abnormal tensor; and detecting the obtained optimal abnormal tensor to generate an abnormal detection result graph. According to the method, the background component, the abnormal component and the noise component are subjected to regularization constraint optimization, and the robust background dictionary is constructed, so that effective separation among the background, the abnormity and the noise is realized.

The application relates to the field of intelligent monitoring, and discloses a device defect detection method based on infrared identification, which comprises the steps of infrared thermal imaging data acquisition, noise reduction processing, nonlinear heat conduction modeling, high-order partial differential optimization and variational regularization inversion solving. An infrared thermal imager collects device surface temperature distribution data, after multi-scale wavelet noise reduction and boundary heat flow balance optimization, a nonlinear heat conduction model is established by using temperature-related thermal conductivity coefficients, the heat diffusion boundary is optimized in combination with a high-order Laplace operator, and the defect heat source distribution is inverted through a variational regularization method, so that the geometric characteristics, thermal parameters and position coordinates of the defects are accurately extracted. The application can adapt to complex working conditions, realize high-precision detection and characteristic analysis of device defects, overcome the problems of noise interference, insufficient boundary processing and weak defect inversion capability in the prior art, and is widely applicable to fault diagnosis and health monitoring of industrial equipment.

The invention discloses a social recommendation method and system based on denoising and preference balance, and belongs to the technical field of graph neural networks and social recommendation. The method comprises the following steps: constructing a social graph; denoising the social graph to obtain a denoised social graph; taking the normalized user item interaction matrix and the de-noised social graph as input, performing double-branch variational coding learning probabilistic user embedding, and performing variational regularization, gating de-noising and attention fusion to obtain user fusion representation; based on user fusion representation, joint reconstruction of an interaction graph and a social graph is carried out, and a training target is obtained through a preference balance optimization target. According to the method, through adaptive graph structure optimization and a long-tail preference constraint mechanism, accurate modeling of a real social relation and balanced expression of diversified preferences are realized, and finally, the reliability, fairness and recommendation precision of the model in a noise sparse social environment are improved.

The application provides a phase retrieval method, device, system and medium based on enhanced total variation regularization, and relates to the technical field of image processing. The method comprises the following steps: constructing a non-convex optimization model for phase retrieval based on an enhanced total variation regularization model and a phase retrieval problem model; converting the non-convex optimization model into a convex optimization model by using a convex function difference algorithm; obtaining an image to be recovered, inputting the image to be recovered into the convex optimization model, iteratively solving the convex optimization model by using an alternating direction multiplier method, and obtaining a recovered image. The application can improve the definition of the recovered image.

This invention relates to the field of detection and imaging technology, specifically to a muon quasi-scattering imaging device and method based on sparse representation and dictionary learning. The device includes a modular muon detector array deployed on one side, an FPGA high-speed data acquisition system, and a computational control unit, and can also integrate a momentum spectrometer module. The method includes offline dictionary preparation and online sparse reconstruction steps. First, a complete dictionary is trained using GEANT4 simulation and the K-SVD algorithm. Then, data is collected to construct an enhanced objective function containing block sparseness and anisotropic total variational regularization. The reconstructed image is obtained by optimizing the solution using the ADMM algorithm. This invention overcomes the limitations of two-sided detection, adapts to single-sided detection scenarios in geotechnical engineering, significantly improves imaging efficiency and quality under low-throughput data, has strong algorithm robustness and a scalable framework, and can accurately identify anomalies such as underground cavities and isolated boulders.

The application discloses a kind of turbulent flame geometry three-dimensional tomography method and system based on multi-probe CGI, comprising: N single-pixel detector is arranged around flame to form multi-probe detection array, dynamic space illumination modulator is used to generate structured light field to illuminate flame, each detector synchronously acquires barrel detection signal;Through compressed sensing reconstruction algorithm combined with convex optimization solving method, flame two-dimensional projection image under each view angle is recovered;Using adaptive histogram equalization, combined with filtering method, the calculation ghost imaging noise is inhibited, and two-dimensional projection image preprocessing is realized;Based on multi-view geometric projection relationship, construct tomographic projection matrix, use algebraic reconstruction technique combined with full variation regularization iterative algorithm to solve three-dimensional voxel distribution, and the flame front geometry three-dimensional structure is reconstructed.The application uses single-pixel detector to replace traditional area array camera, with the advantages of low cost, strong ability to resist harsh environment, high time resolution, suitable for high space-time resolution three-dimensional measurement.

The invention discloses a buried non-metal pipeline positioning method based on deep learning, and relates to the field of buried pipeline detection, and the method comprises the following steps: firstly, collecting a multi-channel time wave field signal generated by the excitation of a seismic source on the earth surface; carrying out singular spectrum analysis and weighted anisotropic total variation regularization denoising processing on the multi-channel time wave field signal to obtain a denoised multi-channel time wave field signal; a deep neural network TransUNet is constructed, and training is carried out by using the simulation time wave field signal after data enhancement and a pipeline position binary image data set corresponding to the simulation time wave field signal; then, the multi-channel time wave field signals after denoising are input into the trained TransUNet model; and finally, the TransUNet model directly outputs and displays an image of the approximate two-dimensional position of the buried nonmetal pipeline. The method does not need to accurately measure the wave velocity or extract the specific waveform, achieves the automatic positioning from the original acoustic signal to the visual image, and provides an efficient and reliable technical means for the two-dimensional position detection of the buried nonmetal pipeline.

The invention discloses an industrial operation data filling and repairing method and device based on dynamic graph learning, equipment and a medium, and relates to the technical field of data processing, and the method comprises the steps: constructing a non-convex low-rank regularization term through introducing a truncated SCAD penalty function, so as to protect main signal energy; a graph attention neural network is utilized to construct dynamic space correlation evolved along with time among dynamic graph total variation regularization item capture variables, an adaptive graph wavelet contraction mechanism is combined to distinguish mutation signals and random noise in a multi-scale frequency domain, and finally an alternating direction multiplier method is adopted for iterative solution. According to the method, the non-convex low-rank regularization item of the data is accurately reserved, dynamic association of the industrial data is adapted, effective abrupt change and noise are distinguished, and the precision and adaptability of data filling and repairing are improved.

The application discloses a hyperspectral anomaly detection method and system based on tensor multi-subspace learning, and the method comprises the following steps: acquiring hyperspectral image tensor data of a detection area; performing waveband-by-waveband normalization processing; decomposing into an anomaly tensor, a noise tensor and a structural background component; designing a non-convex tensor correlation total variation regularization; constructing an iterative sparse weight tensor; adopting a structured-norm; constructing a robust dictionary tensor; establishing an anomaly detection model; adopting an effective iterative updating algorithm based on an alternating direction multiplier method to perform optimization, acquiring an optimal anomaly tensor; and performing detection on the obtained optimal anomaly tensor to generate an anomaly detection result image. The application realizes effective separation among the background, the anomaly and the noise by respectively performing regularization constraint optimization on the background component, the anomaly component and the noise component and constructing a robust background dictionary.

The invention discloses a hyperspectral anomaly detection method and system fusing low rank and smoothness. The method comprises the following steps: acquiring hyperspectral image tensor data of a to-be-detected area; carrying out band-by-band normalization processing on the data; decomposing the data into a background tensor and an abnormal tensor; weighted tensor correlation total variation regularization is designed; constructing a significance weight tensor; establishing an anomaly detection model; optimizing by adopting an effective iterative updating algorithm based on an alternating direction multiplier method to obtain an optimal abnormal tensor; and detecting the obtained optimal abnormal tensor to generate an abnormal detection result graph. According to the method, weighted tensor related total variation regularization is designed to simultaneously represent global low-rank and local smoothness priori of the background tensor, additional regularization parameters are avoided to balance the two priori regularization, and priori distribution information of gradient tensor singular values is fully considered. Therefore, the capability and the flexibility of processing actual problems are improved.

The invention discloses a method for enhancing an edge detection effect of a low-contrast image. The method comprises the following steps of: 1, filtering noise and ripples in the low-contrast image by using a bilateral filtering algorithm, reducing interference and simplifying calculation; step 2, Gaussian filtering is carried out on the image I (x, y) after noise and ripple filtering in the step 1, an estimated illumination layer L (x, y) is obtained, and a reflection layer R (x, y) is inversely solved; 3, associating the total variation regularization mathematical model with a Proximal gradient method for the estimated illumination layer u to be recovered in the step 2, and calculating gradient delta g (u) of a smooth item g (u); for the non-smooth item h (u), iterating the intermediate variable v for k times by applying a Proximal operator to obtain an illumination layer u with the edge of the reserved image; carrying out self-adaptive high-lift sharpening on the reflecting layer to obtain a sharpened and strengthened image Is (x, y); 4, combining and outputting the illumination layer and the reflection layer in the step 3; step 5, carrying out graying processing on the merged image; and step 6, carrying out edge detection on the image after graying processing to obtain an edge result.

The application relates to the technical field of power systems and discloses a graph generation adversarial learning estimation method for an active distribution network topology under incomplete measurement, which comprises the following steps: distribution network topology transformation; transforming an original distribution network topology to obtain a line topology graph; graph convolution feature extraction; based on the line topology graph, a graph convolutional neural network transmits the dependency relationship between node features through a graph structure, extracts the spatial features of graph data, and constructs a generation adversarial topology estimation model; a generation adversarial network is constructed, the extracted spatial features are input into the generation adversarial network for training, wherein a generator function is trained by minimizing a logarithmic probability, and a discriminator function is trained by maximizing a logarithmic probability; a observability layout optimization design model is constructed; variational dropout regularization is added to an input layer of the GCN-GAN, and the observability layout optimization design model learns to reduce the dropout rate of important features of a GCN-GAN model for correctly predicting a topology.

A method is described for performing full waveform inversion with partial variation regularization on seismic data to generate the multi-dimensional map of physical properties of the earth's subsurface. The method must be executed by a computer system.