54 results about "Scale transformation" patented technology

Disclosed is a technical solution for improving the accuracy of the operation of machine learning by compensating for lower precision with a scaling transform. An example non-transitory computer readable medium includes instructions that, when executed, cause a machine to identify at least a first precision data type and a second precision data type associated with execution of a machine learning model, the first precision data type has a first data precision greater than a second data precision of the second precision data type, determining at least one scaling factor to be applied to a first weight of the machine learning model, the first weight being based on the first precision data type, and converting the first weight to a second weight based on a multiplication of the first weight with the at least one scaling factor, the second weight being based on the second precision data type.

The invention discloses a multi-scale crowd counting method and system based on VSSM and mask reconstruction and a medium, and belongs to the field of computer vision and deep learning, and the method comprises the steps: carrying out the multi-scale transformation and segmentation operation of an input crowd image, so as to obtain an image block under each scale; analyzing each image block to obtain an information entropy graph, performing masking operation on each image block according to the information entropy graph, inputting the masked image block into a VSSM to perform feature extraction so as to obtain coding features, performing decoding reconstruction on the coding features through a Transform decoder, and finally obtaining a reconstructed feature image; fusing the reconstructed feature images of different scales by using a multi-scale fusion module to generate fused features; and generating a crowd density map based on the fusion features, and analyzing the crowd density map to obtain a crowd counting result. Counting errors caused by single-scale analysis are avoided, and the crowd counting precision in a full-scale range is remarkably improved.

The invention discloses a lithology generality joint feature classification model construction method, which relates to the technical field of intelligent geological prospecting, introduces a cross-scale feature alignment mechanism, and utilizes scale transformation factors to dynamically match the resolutions of macroscopic, mesoscopic and microscopic features, thereby avoiding feature distortion caused by direct fusion; for example, in a volcanic rock and sedimentary rock mixed area, rough texture of a remote sensing image originally conflicts with smooth response of a logging curve, but through alignment processing, the model can automatically balance credibility of different data sources, and misjudgment is reduced; inter-scale dependence modeling is further enhanced through introduction of the feature association graph, vertical sequence association of a thin interbed is captured through a graph structure, the boundary of a millimeter-level thin layer is made clear, and the problem of fuzziness caused by neglecting of interlayer interaction in an existing scheme is solved.

The invention relates to the technical field of deburring, in particular to deburring equipment of a tool jig and an automatic burr positioning method. The method comprises the following steps: acquiring a plurality of monitoring images of a to-be-deburred tool jig under each view angle; performing Gaussian pyramid scale transformation on each monitoring image, performing angular point detection, and dividing the angular points in each monitoring image into two categories according to the presentation condition of the angular points in a scale image; obtaining a texture expression value of each angular point in the scale image by combining the relative distance between each angular point and other angular points in the corresponding scale image of each monitoring image and the edge change condition; and obtaining a texture feature value of each angular point in the monitoring image by combining the texture representation value of each angular point in the corresponding scale image in all the monitoring images under the same visual angle, and further dividing and screening a burr region for each category. According to the invention, the positioning accuracy of the burr area and the burr removal effect are improved.

The invention provides a Gaussian neural field dynamic scene reconstruction system based on depth consistency constraint, and relates to the technical field of computer graphics, and the system comprises an estimation module which generates a target frame initial depth map; the calculation module reconstructs the point cloud and obtains a point cloud normal direction and a pixel normal direction; the optimization module is used for iteratively correcting the initial depth based on the two types of normal consistency to obtain an optimized depth map; the alignment module is used for determining a scale parameter through regression by taking the first target frame as a reference, and carrying out scale transformation on the depths of other frames to form a consistent depth sequence; the reconstruction module is used for constructing or training a Gaussian neural field based on the sequence and outputting a three-dimensional representation; in addition, the calculation module can contain multi-dimensional wavelets and sparse reconstruction and is used for multi-scale noise suppression and direction weighted fitting. Reference frame selection is based on frame-level quality, geometry and scale stability indexes; according to the system, the intra-frame geometric credibility and the cross-frame scale consistency are improved, ghosting and tearing are reduced, and the stability and integrity of dynamic scene reconstruction are enhanced.

The invention discloses a concrete structure damage variable-scale finite element simulation method based on two layers of grids, which comprises the following steps: step 1, respectively carrying out structure scale and material scale finite element grid subdivision on a concrete structure to obtain a structure-material two-layer grid covering the whole structure simulation area; 2, endowing material parameters to a structure scale unit and a material scale unit, and respectively establishing a structure scale finite element model and a material scale finite element model of the concrete structure; and 3, based on a simulation scale transformation criterion and a structure scale-material scale connection technology, dynamically updating the structure scale-material scale collaborative finite element model, and simulating a concrete structure damage evolution process. According to the method, the simulation scale of the damage area is converted based on the two layers of grids, so that the simulation precision of the concrete damage evolution process can be ensured, the calculation efficiency is relatively high, and the method has a wide application prospect in the refined analysis of the concrete structure damage evolution process.

The invention provides a lightweight semantic segmentation method and system for a remote sensing image of an urban scene, and relates to the technical field of semantic segmentation, and the method comprises the steps: carrying out the multi-scale transformation of the remote sensing image of the urban scene, carrying out the semantic feature extraction, obtaining a first semantic feature map under each reduced scale, adjusting the first semantic feature map, and obtaining a second semantic feature map under each reduced scale; using a second semantic feature map; performing forward, snake-direction and spiral scanning fusion operation on the second semantic feature map to obtain a third semantic feature map, and fusing the third semantic feature map with the third semantic feature map of the previous scale to obtain a fused semantic feature map; splicing the third semantic feature map and the up-sampling result of the fused semantic feature map of the previous reduced scale according to the channel dimension, and then performing two convolution operations and fusing; obtaining a fused feature image corresponding to a reduced scale; and taking an up-sampling result of the last scaled-down fusion feature image as a semantic segmentation result image.

This invention relates to a decoupled evolutionary modeling and service recommendation method for user interest state distribution. The method first collects user interaction data within a service system to construct a sequence of user-item-time triples. A graph convolution algorithm is used to perform initial static embedding learning of users and items, extracting collaborative signals from the user-item interaction graph. A decoupled state space model is constructed, representing user interest states as distributed variables composed of an orthogonal rotation matrix and a diagonal scaling matrix, respectively capturing the evolution of interest direction and the changing diversity of interest. A user interest precision matrix is ​​generated based on the rotation-scaling transformation, and the anisotropic Mahalanobis distance is used to measure the degree of match between users and items. A recommendation list is constructed based on the distance scores, with the top K best items selected as the recommended results. This method overcomes the static embedding limitations of user interest modeling, achieving invariant preservation and potential preference extrapolation in a dynamically evolving structure, thereby improving the expressiveness and generalization performance of the recommendation system.

The application discloses a scale transformation-based secure numerical multiplication calculation method and device, relates to the technical field of secure numerical multiplication calculation, and comprises the following steps: each participant determines a private input value; each participant performs scale transformation on the private input value to generate a private input matrix; each participant takes the private input matrix as input, and calculates a private output matrix by using a secure two-party matrix multiplication protocol based on a secure data confusion technology; each participant calculates the sum of each element of the private output matrix to obtain a private output value; and each participant sends the private output value to a calculation requester to obtain a secure numerical multiplication calculation result. By introducing scale transformation and the secure two-party matrix multiplication protocol based on the secure data confusion technology, the application can improve the calculation efficiency, the communication efficiency and the calculation accuracy, and does not depend on a third-party cloud platform, thereby further guaranteeing security.

The present invention provides an object detection method, device and medium based on multi-scale morphological operators, belonging to the field of object detection in image processing. It includes: processing a single-frame image to obtain a preprocessed image; performing multi-scale morphological dilation operation pixel by pixel to obtain the dilation result and erosion result of each pixel under each scale structuring element; constructing a gray-scale transformation index using the dilation result and erosion result, and marking the scale of the structuring element with the smallest gray-scale transformation index value as the optimal scale; calculating the mean value of the pixel points corresponding to the optimal scale structuring element; for each pixel point, constructing a morphological feature value using the mean value, dilation result, and the gray value of the pixel point, and determining whether the pixel point is a target. The multi-scale morphological operation of the present invention has the characteristics of high detection probability and low false alarm compared with traditional methods, and can effectively alleviate false alarm events caused by local gray-scale changes.

According to the erosion gully extraction method based on the noise label, firstly, an erosion gully remote sensing image is obtained, multi-scale transformation and extraction are carried out on the erosion gully remote sensing image to obtain a result image, and a multi-scale similarity loss item is calculated according to the obtained result image to improve the multi-scale aggregation capability and the anti-noise capability of a model; secondly, according to the erosion gully remote sensing image and the result image, calculating to obtain a refined label image; and finally, according to a cross entropy loss function of the original scale extraction result graph and the noise label graph, a cross entropy loss function of the original scale extraction result graph and the refined label graph, and a multi-scale similarity loss item, summation is carried out to obtain an erosion gully extraction model loss function. Optimizing the erosion gully extraction model according to the obtained erosion gully extraction model loss function; the method has excellent effects in erosion gully segmentation tasks and erosion gully image information extraction.

The invention relates to a user interest state distribution-oriented decoupling evolution modeling and service recommendation method, which comprises the following steps of: firstly, collecting interaction data of a user in a service system, and constructing a user-article-time triple sequence; performing initial static embedding learning on the user and the article based on a graph convolution algorithm, and extracting a cooperative signal in the user-article interaction graph; constructing a decoupling state space model, representing a user interest state as a distribution variable composed of an orthogonal rotation matrix and a diagonal scaling matrix, and respectively capturing interest direction evolution and diversity change characteristics; generating a user interest precision matrix according to rotation-stretching transformation, and measuring the matching degree between a user and an article by adopting an anisotropic mahalanobis distance; and constructing a recommendation list according to the distance scores, and taking the first K optimal articles as recommendation results. According to the method, static embedding limitation of user interest modeling is broken through, invariant keeping and potential preference extrapolation in a dynamic evolution structure are achieved, and the expression ability and generalization performance of a recommendation system are improved.

The present invention relates to the field of natural language processing, and in particular to an information reasoning method, apparatus, computer equipment, and storage medium, which perform multi-scale transformation on a knowledge graph, construct a multi-scale zoom graph of the knowledge graph, and perform multi-scale interaction in combination with a subgraph to be inferred and a multi-scale transformation weight matrix constructed based on the subgraph to be inferred, aggregate neighboring node information from different scales, make full use of information at different scales, and construct a more accurate retrieval subgraph for performing graph information reasoning, thereby improving the accuracy of accurate information reasoning on the graph.

The invention relates to a quantum phase change critical index measuring method and system, a medium and a program product, and relates to the field of quantum calculation and quantum simulation. The method comprises the following steps: for each candidate value of a critical index, determining a plurality of value combinations of a change speed and a system size of a system parameter of a quantum system, so that the candidate value, the change speed and the system size meet a first condition; for each value combination of the change speed and the system size, linearly changing system parameters by the change speed to enable the quantum system to generate quantum phase change; measuring a first correlation function of the quantum system; performing scale transformation meeting a second condition on the first correlation function to obtain a second correlation function; for each candidate value, comparing differences between the second correlation functions; and selecting a candidate value that minimizes the difference as a measurement of the critical index. The method can overcome the problems of short coherence time and limited system size of a quantum system, and improves the measurement accuracy.

The present invention relates to a method and device for counting repetitive sports actions based on a multi-scale transformation network. The method includes: detecting skeletal key points in an original video to obtain an original skeletal sequence, sampling the original skeletal sequence at three scales to obtain three down-sampled skeletal sequences; encoding the temporal periodicity features of each down-sampled skeletal sequence for each skeletal joint point to obtain a temporal autocorrelation matrix for each skeletal joint; performing a coarse-to-fine periodic selection on the temporal autocorrelation matrix of each skeletal joint to obtain three weighted fusion period cubes; performing multi-scale periodic fusion on the three weighted fusion period cubes to obtain a multi-scale period cube; performing impulse graph regression on the multi-scale period cube based on a multi-scale period transformation network to obtain an impulse graph; adding all elements in the impulse graph to obtain a repetitive action count result. The embodiments of the present invention are accurately tested, and the count result is interpretable.

Provided is a secure real-number multiplication method and apparatus based on scale transformation, which relates to the technical field of secure real-number multiplication. The secure real-number multiplication method includes: determining, by each participant, a private input value; performing, by each participant, scale transformation on the private input value to generate a private input matrix; with the private input matrix as input, computing, by each participant, a private output matrix by utilizing a Secure Two-Party Matrix Multiplication (S2PM) protocol based on a secure data disguising technology; calculating, by each participant, a sum of all elements in the private output matrix to obtain a private output value; and sending, by each participant, the private output value to a computation requester to obtain a secure real-number multiplication result.

Embodiments of the present application disclose a tongue feature recognition method and device, electronic equipment and a storage medium. The method comprises: acquiring a tongue image; performing multi-scale operation on the tongue image to determine at least two scale transformation images; inputting the at least two scale transformation images into a recognition model to output target feature information corresponding to the tongue image through the recognition model; and the recognition model is a cross-network model composed of a segmented convolutional neural network (Seg-CNN) structure and a point convolutional neural network (Point-CNN) structure. In this way, when analyzing the tongue image, the multi-scale operation is performed on the tongue image, and the network model composed of the Seg-CNN structure and the Point-CNN structure is used, so that the global feature information with points, lines and surfaces can be recognized, thereby the crack feature in the tongue image can be clearly recognized, and the recognition accuracy of the crack feature is improved.

The present invention relates to the field of epistaxis recognition, and specifically relates to an epistaxis recognition method and a computer device based on multi-scale feature fusion. The solution includes: using a fully convolutional network to extract local features of an epistaxis image, using a Transformer to extract global features and capture global context information, performing different scale transformations on the local feature map and the global feature map to generate feature maps of different scales, performing multi-scale feature fusion, performing context fusion, calculating attention weights, and finally outputting the final features, then calculating the cross-entropy loss and training the model, and finally recognizing epistaxis through the trained model. The present invention addresses the problems of dataset imbalance and small bleeding areas being easily overlooked, dynamically adjusts the attention weights during the training process, focuses on difficult negative samples and small sample areas, thereby improving the accuracy. Ultimately, the accuracy of epistaxis image recognition is improved. The present invention is applicable to epistaxis recognition.

The present invention provides an image recognition method for a pepper automatic picking machine, which relates to the field of image recognition. It introduces a non-uniform scale transformation factor to replace the traditional fixed-factor scale scaling strategy, realizing the dynamic adjustment of the scale level features according to the target distribution; improves the traditional scale fusion method, calculates the dynamic scale weighting coefficient through scale response measurement, and realizes the weighted fusion of the scale features after channel adjustment; combines the adaptive transformation of the color feature channel, the dynamic interaction modeling of the color channels and the global color adjustment, effectively perceives the color features of peppers at different maturity stages, and improves the color expression ability under complex lighting conditions; optimizes the information fragmentation problem caused by the complete separation of the classification and regression tasks in the existing decoupling head through a cross-task feature sharing mechanism, introduces a cross-task sharing coefficient to enhance the effective feature interaction between the classification and regression tasks in the deep feature layer, and improves the accuracy and robustness of recognition.

The present invention provides an image stitching method and a stitching system based on adaptive multi-scale transformation, belonging to the technical field of image processing. By performing multi-scale feature extraction on the images to be stitched, adaptively assigning fusion weights to the multi-scale features extracted from adjacent images, using an edge detection algorithm to divide adjacent images into multiple regions, and adjusting the fusion parameters between regions of adjacent images according to the attributes of the regions, iteratively calculating the fusion weights and fusion parameters through a genetic algorithm, determining the optimal parameter combination based on evaluation metrics, and finally generating a high-quality and high-definition stitched image. The present invention can make full use of the feature information of images at different scales, achieve more accurate and efficient image stitching, and the method of adjusting fusion parameters based on regional attributes can reduce seams and distortions in the stitching region, improve the visual effect of the stitched image, and automatically find the stitching parameters most suitable for the current image, improving the automation degree and robustness of the stitching process.

The invention discloses a motor bearing fault diagnosis method based on multi-scale data enhancement and integration, and the method comprises the steps: collecting vibration signal samples of a bearing in a normal operation state and a fault state, and carrying out the multi-scale transformation of the vibration signal samples; enabling the processed fault vibration signal sample to pass through a multi-scale generative adversarial network, generating a multi-scale generated fault sample, and achieving data enhancement; training an integrated convolutional neural network model by using the enhanced bearing vibration signal sample; and performing fault diagnosis by using the trained integrated convolutional neural network model. According to the motor bearing fault diagnosis method based on MSGAN data enhancement and ECNN, the problems of motor bearing data imbalance and multi-scale are solved, the MSGAN is used for generating multi-scale samples to achieve data enhancement, training of a fault diagnosis model is facilitated, and meanwhile the ECNN is used for improving the fault diagnosis precision.

The invention provides a steel bridge deck pavement performance multi-scale test evaluation method which comprises the following steps: acquiring test data of steel bridge deck pavement, and extracting a multi-scale performance parameter set from the test data; based on the multi-scale performance parameter set, establishing a scale conversion function for mapping local performance parameters to an overall structure state; constructing an initial digital model based on the scale transfer function and the multi-scale performance parameter set, and performing parameter reverse calibration on the initial digital model in combination with actually measured response data in the full-scale segment scale data to obtain a multi-scale performance digital model; and based on the multi-scale performance digital model, simulating the obtained service environment data of the target pavement structure to obtain performance evaluation data of the target pavement structure, the performance evaluation data including a predicted performance degradation value and a predicted remaining service life. By the adoption of the method, more reliable technical support can be provided for performance management and control of steel bridge deck pavement.

The invention relates to the technical field of image processing, and discloses a scale change-based real-time video jitter removal method, which comprises the following steps of: 1, acquiring a video frame at a previous moment as a pre-frame through single video shooting equipment, and acquiring a video frame at a current moment as a processing frame; step 2, pre-processing the obtained pre-frame and the processed frame; and step 3, extracting a preset number of feature points in the pre-processed pre-frame and the processed frame, matching the obtained feature points, removing mismatching points by using an RANSAC algorithm, and performing scale transformation and mapping on the matched feature points to the original size to obtain matching points of the pre-frame and the processed frame under the original size. The method comprises the following steps: preprocessing an acquired pre-frame and a processed frame, generating a low-resolution image through scale transformation downsampling, extracting feature points from the low-resolution image, removing mismatching points, and mapping the feature points to matching points with original sizes.

Provided are a secure two-party data comparison method and apparatus based on scale transformation. The method includes: transmitting, by a computation requesting party, a two-party data comparison request to two participant nodes; performing, by each participant node, scale transformation and linear scaling on private data locally after receiving the two-party data comparison request, to obtain an encrypted vector; determining, by each participant node, a real number locally using a secure two-party dot product protocol based on the local encrypted vector, and sharing the real number with the other participant node; determining, by each participant node, a comparison sign based on the obtained real number and transmitting the comparison sign to the computation requesting party; and determining, by the computation requesting party, a comparison result of the private data of the two participant nodes based on the comparison signs transmitted by the two participant nodes.

The present invention belongs to the field of 3D detection and measurement, and specifically relates to a 3D point cloud registration method, system and storage medium. The present invention overcomes the problem that the existing point cloud registration algorithm has low registration accuracy when the two sets of point clouds to be registered have scale transformation. The present invention mainly includes the following steps: first, determine the point cloud {p i}The scale transformation factor, rotation axis vector, and cosine and sine values ​​of the rotation angle of the points A, B, and C in the registration unit are transformed to points A', B', and C'; then the point cloud {p i} to point cloud {p i '}; Finally, using the point cloud {p i} to point cloud {p i '} scale transformation factor and angle transformation matrix to calculate the point cloud {p i} to point cloud {p i '} displacement transformation matrix. In the registration process, the possible scale transformation in different point clouds is considered, and the displacement change matrix is ​​calculated based on the scale transformation factor, which improves the registration accuracy and reduces the registration error.

The application relates to the field of 3D vision, in particular to a point cloud shape analysis method based on a space geometry perception convolutional neural network. According to a K nearest neighbor algorithm, a point cloud is constructed as graph structure data, and a domain adaptive convolution kernel with different geometric shapes is adaptively generated according to features in each neighborhood. Subsequently, unit direction vectors, elevation angles and azimuth angles of neighborhood nodes are taken as prior geometric information, and convolution operation is performed on the generated convolution kernel to realize space geometry perception convolution operation. Subsequently, graph attention pooling is designed to coarsen the point cloud, realize multi-scale analysis and reduce the calculation cost. Finally, space geometry perception convolution and graph attention pooling are taken as basic units to construct two networks to realize point cloud classification and component segmentation tasks. The application can effectively classify and component segment target point clouds, and ensures the invariance of point cloud translation and scaling transformation, so that the model has stronger robustness.

The invention discloses an RGB-D salient target detection method based on texture enhancement guidance. The RGB-D salient target detection method is specifically implemented according to the following steps: step 1, constructing a data set and an encoder; step 2, constructing a texture enhancement module; step 3, constructing a dual-path adaptive interaction module; and 4, constructing a dynamic decoding module. In order to solve the key problems of heterogeneous modal feature degradation, depth noise cross-layer propagation, multi-scale semantic mismatch and the like in the prior art, noise suppression of depth features is realized by using high-frequency texture prior constraints, and cross-modal semantic association is established through a channel-space collaborative dynamic interaction mechanism. A deformable cross-scale transformation technology is adopted to realize progressive calibration of multi-level features, and the method shows remarkable boundary integrity and noise suppression advantages in input scenes of multiple targets, low-quality depth and the like.

The application discloses a kind of tax-related legal text-oriented named entity recognition dependent enhancement method, comprising: tax named entity recognition is regarded as span classification task, a large number of spans are enumerated from input text by sliding window, and the deep representation of each span is generated by feature splicing method;Introduce a contrastive learning loss, and the contrast relationship is mined from highly overlapping span;Scale transformation mechanism is used to realize span interaction, and the geometric information of each candidate span is embedded in native span representation, to encode the interaction dependency between spans.The application converts tax named entity recognition into span classification task, and fully mines the interaction dependency between entities, realizes strong inference relationship, and introduces contrastive learning to improve the discrimination degree between different types of highly overlapping entities, so that the named entity in tax legal text can be more accurately and reasonably recognized, laying a foundation for downstream tasks such as tax incentives.

The invention relates to the field of information extraction, and provides an information extraction method, device and equipment and a computer storage medium, and the method comprises the steps: obtaining a to-be-processed image, and carrying out the scale transformation of the to-be-processed image, and obtaining a to-be-enhanced image of at least one scale; inputting each to-be-enhanced image into a corresponding feature extraction network to obtain a feature image corresponding to the to-be-enhanced image of each scale; performing feature fusion on each feature image to obtain an enhanced image of an original scale; performing reconstruction processing on the enhanced image based on the to-be-processed image to obtain a target image; and based on a preset information extraction vector, obtaining target information matched with the information extraction vector in the target image.

The present invention discloses an image fusion method based on multi-scale transformation and convolution sparse representation, including: decomposing medical images of different modalities by rolling guided filtering to obtain basic sub-images and detail sub-images, decomposing the basic sub-images into high-frequency sub-bands and low-frequency sub-bands in multi-scale directions by NSST, encoding them into high-frequency sparse coefficients and low-frequency sparse coefficients by convolution sparse representation, fusing the sparse coefficients by corresponding different fusion rules for the high-frequency and low-frequency coefficients, reconstructing the fusion coefficients by convolution sparse representation, reconstructing the fusion sub-bands by inverse NSST to obtain the fused basic sub-image, and simultaneously using the weighted improved Laplace energy and fused detail sub-images, adding the fused basic sub-images and detail sub-images to obtain the final fused image. The present invention solves the problems existing in the existing medical image fusion methods, provides a better solution for the field of medical image fusion, and achieves the acquisition of high-quality fused images.