71 results about "Optimal alignment" patented technology

The invention relates to the technical field of video content understanding, in particular to a video content understanding method of a multi-mode advertisement inventory intelligent matching system, and the method comprises the steps: taking a video frame of an original advertisement video as a benchmark, and in a shot where a timestamp of each video frame is located, obtaining a video frame of the original advertisement video; analyzing the similarity of adjacent video frames and the dispersion degree of feature matrix elements corresponding to the voice segments, and determining a first search window; meanwhile, analyzing the occurrence frequency and duration of the text fragment in unit time, and confirming a second search window; based on the search window, analyzing the feature similarity degree of the voice segment and the text segment with the video frame in the corresponding window, and obtaining an optimal alignment path; and aligning the timestamps of the voice segments and the text segments with the timestamps of the video frames, and realizing multi-level semantic analysis through a multi-modal fusion technology and a deep learning model. The invention aims to synchronize the sequential relationship among the audio, the text and the video frame and prompt the accuracy of video content understanding.

The invention relates to the technical field of alignment of photoetching machines, and discloses an alignment method, device and equipment for a photoetching machine and a storage medium. The method comprises the following steps: carrying out image acquisition and space registration on a plurality of small-size alignment marks preset on the surface of a wafer to obtain a virtual large mark image; performing double-light-source imaging on the target area of the virtual large-mark image by adopting an ultraviolet light source and an infrared light source respectively to obtain an ultraviolet light image group and an infrared light image group; performing multi-level fuzzy discrimination and spatial perception double attention processing on the ultraviolet light image group and the infrared light image group to obtain a multi-level optimization feature set; according to the multi-level optimization feature set, guiding filtering and self-adaptive parameter fusion are executed, an optimal alignment coordinate is obtained, and the photoetching machine is driven to execute alignment operation based on the optimal alignment coordinate, so that the optimal configuration of alignment parameters under different process conditions and environment interference is ensured, and the alignment accuracy is improved. And the alignment accuracy and reliability of the photoetching machine are improved.

The invention discloses a memory enhancement-based large language model retrieval fine tuning method. The method comprises the following steps of: performing layer-by-layer dense vector representation on text sequences in a text sequence library; performing layer-by-layer alignment analysis to obtain an alignment loss value of each layer; performing layer-by-layer uniformity analysis to obtain a uniformity loss value of each layer; obtaining an optimal alignment layer and an optimal uniform layer according to the alignment loss value of each layer and the uniformity loss value of each layer; inputting the optimal alignment layer and the optimal uniform layer into a retrieval fine tuning module to obtain a final output vector; and substituting the final output vector into the cross entropy training loss function, and obtaining model parameters of the retrieval fine tuning module by minimizing the value of the cross entropy training loss function. According to the method, effective text retrieval is carried out while internal parameters of the large language model are not finely adjusted and the original generation capability of the large language model is kept, model compatibility of text retrieval and text generation is completed, and a technical basis is provided for constructing a long-term memory module of the large language model.

The invention discloses an intelligent analysis method and system for gas station operation data, and relates to the technical field of operation data analysis, and the method comprises the steps: collecting gas station data for preprocessing, generating symbolized sequence data for the preprocessed data through employing a symbolized approximate aggregation SAX algorithm, and calculating the similarity of the symbolized sequence data by using a dynamic time warping (DTW) algorithm, constructing an accumulative error matrix, and backtracking an optimal alignment path to obtain high-similarity sequence data. According to the method, a continuous time sequence is converted into a discrete symbol sequence through the application of a symbolic approximate aggregation algorithm, the data complexity is remarkably reduced, the trend characteristics of original data are reserved, a symbol quantile interval is generated through a cumulative distribution function of Gaussian distribution and an inverse cumulative distribution function of standard normal distribution, and the probability that the symbol quantile interval is larger than the discrete symbol sequence is reduced. The symbolization process has a strict statistical basis, and the frequent pattern is extracted by using an Apriori algorithm through a high-similarity sequence data set.

The invention discloses a handwritten stroke order identification method, which comprises the following steps of: acquiring handwritten electronic information sequence data, constructing a distance matrix and an alignment path of a handwritten and a corresponding printing form in a character library, identifying dislocation point positions and dislocation strokes in the alignment path according to a point position alignment relationship between the handwritten sequence data and the printing form sequence data, and identifying the handwritten stroke order according to the dislocation point positions and the dislocation strokes. And replacing the printed form sequence data according to the handwritten form handwriting point positions and strokes in the distance matrix until an optimal alignment path is obtained, and outputting the current writing stroke order according to the point position pairs and strokes determined according to the optimal alignment path. The handwriting is segmented into standard strokes, DTW alignment can be used, the standard strokes can be used for segmentation, the printed stroke order in the character library is adjusted to be consistent with the handwriting stroke order, the standard strokes are determined, the stroke order is defined, and handwriting recognition of an auxiliary identification system is facilitated.

The invention relates to the technical field of people relationship mining, in particular to a multi-dimensional people relationship discovery method based on multi-modal information fusion, which comprises the following steps: extracting text, visual and audio modal features from a video; according to the Euclidean distance of the modal features, optimal alignment path optimization and nonlinear time adjustment are carried out, and the modal features of time step alignment are obtained; performing self-attention and cross-attention mechanism processing on the modal features to obtain multi-modal representations of texts, vision and audios; multi-modal representation of texts, vision and audios is carried out, and a graph structure network of a social relation, a standing relation and a co-occurrence relation is established; determining the social relationship closeness degree, the standing consistency degree and the co-occurrence strength of each person in the video according to the graph structures of the social relationship, the standing relationship and the co-occurrence relationship; according to the method, the understanding accuracy of the multi-dimensional character relationship can be improved.

The invention provides a cross-language text classification and processing method and system based on deep transfer learning, and relates to the technical field of text processing, and the method comprises the steps: extracting feature representations of a source language text and a target language text at different linguistic levels through a multi-level semantic transfer network; and determining an optimal alignment path, performing nonlinear mapping alignment to obtain fusion features, propagating category semantics by using a semantic bridging function, iteratively updating pseudo-tag confidence distribution of the target language text, and completing classification in combination with a multi-task learning model. According to the method, the problem of text classification in a cross-language scene is effectively solved, and the accuracy and efficiency of low-resource language text processing are improved.

The invention relates to the field of power load monitoring, in particular to a load event detection method based on power time sequence image features. The method comprises the following steps: acquiring load power time sequence data in real time, and preprocessing the collected load power time sequence data; segmenting the load power time sequence data through a window segmentation method based on the preprocessed data; positioning a change point area through a time sequence change point detection algorithm based on the generated two-dimensional power curve image; classifying and positioning load events through a YOLO model based on the obtained change point area; and on the basis of a target detection result, secondary correction is carried out on an event starting point and an event ending point identified in the target detection process based on a local trend analysis algorithm. An optimal alignment path distance between a detection event and a template is calculated by adopting a DTW algorithm, the influence of expansion deformation of a time axis is eliminated, and the problem that a traditional Euclidean distance is sensitive to a time phase is solved.

Correction and / or other adjustments to optical systems of lithographic, metrology, and / or other apparatuses are generally based on information from previous manufacturing and / or metrology operations, user experience of the lithographic and / or metrology systems and / or processes, underlying optical physical principles, and / or other information. Advantageously, the present systems and methods use reinforcement learning to optimize alignment (and associated settings and / or other characteristics) of tens, hundreds, thousands, or millions of optical system components, and make corrections and / or other adjustments to optimize adjustments and / or processing time and / or cost.

The disclosed embodiments concern methods, apparatus, systems, and computer program products for computing the best alignment score between a set of query sequences and a non-acyclic sequence-labelled reference genome graph comprising a processor and non-transitory memory. In an embodiment, said computing is performed employing an alignment algorithm suitable for non-acyclic graph that leverages inter-sequence and intra-sequence parallelism, executed on a heterogeneous computing system, containing at least a CPU (Central Processing Unit) and a GPU (Graphics Processing Units).

The invention discloses a deep learning time sequence alignment method and system for a multi-view asynchronous sequence image, and the method comprises the following steps: carrying out the image screening of a multi-view asynchronous collected angiography sequence, and constructing a short sequence composed of three frames of continuous images; the method comprises the following steps of: extracting a backbone network by adopting an OfficientNet-B0 as a feature extraction backbone network, removing a classification head of the feature extraction backbone network, replacing the classification head with identical mapping, outputting convolutional features, and performing network optimization in combination with a soft dynamic regularization alignment loss function (soft DTW); extracting a feature vector describing the heart cycle progress through a network; calculating the cosine similarity of the feature vectors among different view angles, constructing a similarity matrix, and searching an optimal alignment path in the similarity matrix by using a soft dynamic time warping algorithm; and outputting a time sequence alignment frame and a corresponding similarity score according to the optimal alignment path. According to the method, the multi-view asynchronous sequence image can be effectively processed, time sequence alignment is realized, and a reliable alignment basis is provided for cardiovascular image analysis.

The invention discloses a synthetic seismic record alignment and calibration method based on forced dynamic time warping, and belongs to the technical field of exploration, and the method comprises the following steps: S1, data preprocessing and seismic wavelet extraction; s2, geological constraints are set, and alignment and non-alignment points are determined; s3, generating a synthetic seismic record by using the reflection coefficient and the seismic wavelet; and S4, calculating an optimal alignment path by applying a forced dynamic time warping (FDTW) algorithm. According to the synthetic seismic record alignment and calibration method based on forced dynamic time warping, priori geological information can be fully fused, matching deviation caused by seismic record abnormity is effectively avoided, and the alignment precision between a synthetic record and an actual record is greatly improved.

This invention discloses an automatic construction method for catenary support equipment archives based on multi-frame visual sequence alignment, belonging to the field of railway operation and maintenance technology. The method includes: acquiring trackside inspection videos; segmenting the frame data sequence based on the area change trajectory of the pole boundary frame to obtain several pole trajectories compressed into candidate pole nodes; extracting representative frames, kilometer marker area characters, license plate area characters, and pole type labels to obtain a candidate pole node sequence arranged by time; acquiring pre-stored real structured trackside data; constructing a key point dictionary based on the uniqueness of kilometer markers; generating a real pole number sequence arranged in line order; scoring all video frames covered by the candidate pole nodes frame by frame; using the Viterbi algorithm to solve the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence; and generating catenary support equipment archives. This invention solves the problems of low accuracy in catenary support equipment archive construction, inability to effectively handle missed detections, false detections, and disordered sequences.

The application discloses a kind of energy storage system multi-frequency sampling data processing method, system, equipment and medium, it is related to energy storage system data processing technical field, by the data slice division of the multi-modal multi-frequency sampling data obtained and adding frequency position coding, the input embedding sequence corresponding to each mode is generated;Based on the acquisition time stamp of each mode data, construct positive sample pair and negative sample pair, input embedding sequence is input into transformer encoder to extract features, and the encoder parameters are adjusted to complete self-supervised training with the goal of maximizing positive sample pair similarity and minimizing negative sample pair similarity;After the same processing, the multi-modal data to be aligned is input into the trained encoder to obtain the aligned perception feature vector, the optimal alignment path is determined according to the similarity of each vector, and the aligned multi-modal data is output by resampling, the application can automatically learn cross-modal alignment without manual annotation and hardware synchronization, and the alignment accuracy and computational efficiency are considered.

This invention discloses an embodied data generation method for category generalization operations, belonging to the field of embodied operation technology. The method includes: constructing a shape library containing warp mappings between templates and target shape spaces; acquiring the 3D model, pose trajectory, and function point annotations of source demonstration data; generating a new shape model through the shape library mapping; calculating the geometric deviation of key points; solving for the optimal alignment transformation sequence and correcting the robotic arm's motion trajectory; integrating the new model into the scene point cloud; and driving the virtual robotic arm to synthesize a new sequence of observation action pairs. This invention achieves function-aware nonlinear warp mapping, solves the function alignment problem under complex geometric deformations, and improves the generalization ability of embodied operations; it adopts a reality-to-reality paradigm, completing efficient data augmentation with a small number of annotations, significantly reducing data collection costs.

The invention provides alignment marks with different duty ratios and an alignment method, and the alignment marks comprise a plurality of alignment marks which are located on a cutting channel and are distributed along the extending direction of the cutting channel, each alignment mark comprises a plurality of gratings which are arranged in parallel, the plurality of gratings are arranged at intervals along the extending direction of the cutting channel, and a gap is formed between every two adjacent gratings; the extension direction of the gratings is perpendicular to the extension direction of the cutting channel, and the plurality of gratings are rectangular; wherein in each alignment mark, the grating widths are the same, the duty ratios formed by the grating widths and the sum of the distances between the grating widths and the adjacent gratings are the same, and the duty ratios of the multiple alignment marks are gradually increased or decreased in the extending direction of the cutting channel. After the film layer is formed, a plurality of alignment marks are scanned in sequence, and the optimal alignment mark is selected from the alignment marks and fed back to serve as the alignment mark of the corresponding film layer during next photoetching, so that an ideal alignment signal is obtained.

The invention discloses a hydraulic ring geological disaster intelligent identification method based on deep learning, and the method comprises the following steps: 1, collecting original data of a plurality of sensors, and carrying out the preprocessing of the original data, and generating the original data of a unified structure; 2, projecting the original data of the unified structure to a low-dimensional space through local linear embedding, and generating a spatial feature vector; 3, calculating a Manhattan distance between the spatial feature vectors to construct a cumulative distance matrix, obtaining an optimal alignment path, and generating a time sequence feature vector; 4, analyzing the persistent features of the time sequence and spatial features, and constructing a persistent feature map; 5, generating a multi-dimensional feature vector set through an improved Markov random field; and 6, performing anomaly detection according to the Gaussian mixture model, generating disaster early warning information, and performing visual output. According to the method, multi-sensor data and deep learning are fused, and the accuracy and real-time performance of hydraulic ring geological disaster early warning are improved.

This invention relates to a method and system for constructing geographic entities and generating relationships, specifically including: identifying target vector layers from multiple sources, and extracting layer names, attribute field names, and their corresponding attribute values ​​from the target vector layers; performing multi-level filtering and comprehensive scoring based on layer names to determine the standard layer category for optimal alignment of layer names; mapping the attribute field names of the target vector layers to target standard fields using the standard layer category as a reference, and converting the attribute values ​​corresponding to the attribute field names into standard codes to achieve attribute standardization; acquiring the geometric data of the target vector layers, generating target vector entities based on the geometric data, and binding the corresponding standard attributes to the target vector entities; establishing topological relationships, hierarchical relationships, and cross-source entity referencing relationships between entities based on spatial topology analysis and semantic association analysis, thereby forming a structured geographic relationship network.

The present invention relates to a non-aligned SAR-optical image coupled contrast fusion building segmentation method. The present invention comprises the following steps: constructing a multimodal dataset for building remote sensing image segmentation; constructing a non-aligned SAR-optical image coupled contrast fusion building segmentation model; training a non-aligned SAR-optical image coupled contrast fusion building segmentation model; and obtaining the result of the non-aligned SAR-optical image coupled contrast fusion building segmentation model. Compared with the prior art, the present invention solves the problem that the prior method focuses on strict modal alignment and ignores the gap between modalities when dealing with the weak alignment problem of multimodal image pairs. The present invention captures the optimal alignment position by guiding the adaptive feature alignment module through offset, thereby realizing adaptive alignment between modalities. Further, the complementary features between modalities are extracted through the coupled contrast learning framework of multimodal image fusion, thereby reducing the redundant information of the fusion result and ensuring the accuracy and robustness of the segmentation result.

This invention relates to the technical field of geomagnetic positioning and discloses a geomagnetic sequence matching method, comprising the following steps: First, a sequence library is constructed using a geomagnetic reference map, and real-time geomagnetic data is received for matching. Second, sequence similarity is evaluated using a particle swarm optimization algorithm combined with a multidimensional dynamic time programming algorithm, incorporating the Maltrov criterion and roulette wheel optimization search. Then, the velocity and position of the particles are continuously adjusted using adaptive inertia weights. In each iteration, the fitness of each particle is calculated, and the individual optimal position and global optimal position of the particle are updated. When the maximum number of iterations or the particle's fitness threshold is reached, the iteration terminates, the optimal matching position is output, and finally, the user's final position is calculated using the geomagnetic sequence. The technical solution of this invention can find the optimal alignment path between two sequences in the time dimension using particle swarm optimization and simulated annealing algorithms without increasing computational load, improving matching accuracy and shortening computation time.

A feature dimension reduction method for a small sample classification task belongs to the technical field of data processing, and comprises the following steps: dividing image data into a training set and a test set; constructing a small sample image classification model based on class related feature orthogonal alignment; performing sample feature extraction, sample feature dimension reduction and sample feature orthogonal alignment on an input image by using the small sample image classification model, calculating sample similarity and classification loss, optimizing the model, and outputting an optimal category related feature reconstruction model; and classifying the test image by using the optimal model. According to the method, efficient dimensionality reduction can be completed on the premise that discrimination information is reserved to the maximum extent, the compactness and discrimination of feature representation are improved, and the generalization performance of a classification model is improved. According to the method, the optimal alignment of the query sample and the support sample in the local feature space is realized, the geometric structure integrity of the features is kept, the precision and stability of cross-task feature alignment are improved, and the accuracy and robustness of cross-domain classification are improved.

The invention relates to the technical field of hollow-core optical fiber alignment, in particular to a hollow-core optical fiber automatic alignment method and device, and the method comprises the steps: obtaining a first end face image of a first hollow-core optical fiber and a second end face image of a second hollow-core optical fiber; filtering the first end face image to obtain a first filtered image, and filtering the second end face image to obtain a second filtered image; the first filtering image is rotated angle by angle, and the rotated first filtering image and the second filtering image are subjected to cross-correlation calculation until the optimal alignment angle corresponding to the maximum value of the cross correlation coefficient is obtained; rotating the first hollow-core optical fiber to an optimal alignment angle so as to align the first hollow-core optical fiber with the second hollow-core optical fiber; according to the scheme, based on the end face image of the hollow-core optical fiber, the end face information of the hollow-core optical fiber can be efficiently extracted, the image processing process is simplified, angle matching is optimized through a self-adaptive rotation adjustment algorithm, and fiber cores and anti-resonance structures of the two hollow-core optical fibers can be perfectly aligned.

The invention provides a welding method after optical fiber coupling, a computer device, a storage medium and a program product. The method comprises the following steps: acquiring a first change relation between optical fiber output power and offset from an optimal position; welding at the target position is controlled, and a second change relation of the optical fiber output power along with time change is recorded; determining the highest power of the solidification process and the power after solidification in the second change relation; when it is determined that the difference between the power after solidification and the optimal alignment power in the first change relation is not within a first preset range, fitting operation is carried out according to the highest power in the solidification process, the power after solidification and the first change relation, and the calculation offset moving to the optimal position corresponding to the optimal alignment power is obtained; adjusting the target position by calculating the offset; and the adjusted target position is controlled to be welded. The welding efficiency after optical fiber coupling is improved, and material loss in the welding process is reduced.

The invention discloses a converter installation position deviation detection method applying point cloud processing, and the method employs an HWD module integrated with a deep residual network and feature image resolution to be integrated into a network model of point cloud extraction features, and the HWD module is connected with a point cloud extraction layer in parallel. Point cloud semantic segmentation is enhanced in a complex construction environment, it is guaranteed that the overall accuracy of local and global information extraction of a converter point cloud model is obviously improved, and the needed converter model extraction time is saved; and secondly, intercepting a circular slice from the extracted optimal model data, performing fitting circle solution based on the circumferential contour of the slice, recording a least square circle center parameter, determining an optimal alignment mode between two axes, and calculating coaxial deviation of mounting positions between components, so that errors are effectively reduced. The method still has good monitoring performance under the complex environment of site construction, can well solve the problem that the error of the assembly position of the converter exceeds the standard in construction, and has good detection capability.

The application provides a speech text alignment model training method and device, electronic equipment and a storage medium, and relates to the field of large models. The method comprises the following steps: obtaining a pre-training model and text data, and performing time information prediction on each word in the text data based on the pre-training model to obtain a time prediction result corresponding to each word; obtaining speech data, and performing text reconstruction on the speech data based on the time prediction result to obtain reconstructed text data; performing optimal alignment path prediction on the reconstructed text data and the speech data based on a dynamic programming algorithm to obtain an optimal alignment path; and adjusting parameters of the pre-training model based on the optimal alignment path to obtain a speech text alignment model. The application solves the problem of low applicability of a speech text alignment model in related technologies.

An extreme ultraviolet light source includes a laser, a target and a controller. The target may be a fuel droplet. The controller may determine optimal laser-to-droplet alignment positions in two orthogonal directions and update the position of the laser to maintain optimal alignment between the laser and fuel droplet during radiation generation. A method of determining an optimal laser-to-droplet alignment position includes fitting a polynomial to a radiation generation metric that varies based on the laser-to- droplet alignment position, and determining an optimal alignment position for the laser relative to the fuel droplet based on an apex of the polynomial fit; or evaluating commanded laser energy versus laser- to-droplet alignment position across an available alignment space to determine where commanded laser energy is maximum, and determining an optimal alignment position for the laser relative to the fuel droplet so that the alignment space is symmetric between the maximum points.

The invention discloses an equipment fault diagnosis method and device, electronic equipment and a storage medium. The method comprises the following steps: according to different wavelet scales, respectively executing continuous wavelet transform on an acquired vibration signal, a current signal and a voltage signal of equipment to be diagnosed, and extracting an effective wavelet coefficient under each wavelet scale; calculating energy probability distribution of effective wavelet coefficients to obtain wavelet scale entropies of the vibration signal, the current signal and the voltage signal under each wavelet scale; determining dominant scale entropy corresponding to each time point in the three signals; based on the dominant scale entropy difference of the vibration signal and the voltage signal and the dominant scale entropy difference of the current signal and the voltage signal calculated for each time index tuple in the constructed alignment path set, solving a pre-constructed optimal alignment path function in combination with path smoothing constraints; performing time-frequency domain alignment on the multi-modal signal according to the solved optimal alignment path set; and carrying out equipment fault diagnosis based on the aligned multi-mode signals.

This invention discloses a method for measuring cavitation in plant xylem, comprising: calculating image parallax in an image sample set and recovering scene depth information; extracting the center line of the main stem based on the segmentation results of the main stem; determining the two-dimensional pixel coordinates of the target vibration measurement point above the plant roots along the direction of the center line of the main stem; and calculating the three-dimensional spatial position of the target vibration measurement point by combining the scene depth information corresponding to the target vibration measurement point; automatically coarsely aligning the target vibration measurement point above the plant roots; obtaining the optimal alignment posture of the laser Doppler vibrometer; collecting vibration signals from the target vibration measurement point above the plant roots; and measuring the cavitation in the plant xylem using the collected vibration signals. This invention addresses the problems of manual reliance, low efficiency, poor repeatability, and unstable signal quality caused by on-site interference in existing technologies for measurement point identification and LDV alignment, and achieves automatic positioning, automatic alignment, and cavitation event measurement of the position above the plant roots.

The invention relates to the technical field of geomagnetic positioning, and discloses a geomagnetic sequence matching method, which comprises the following steps of: firstly, constructing a sequence library by using a geomagnetic reference map, and receiving real-time geomagnetic data for matching; secondly, evaluating the sequence similarity through a particle swarm algorithm in combination with a multi-dimensional dynamic time planning algorithm, and introducing a Meltgov criterion and a roulette optimization search; then, the speed and the position of the particles are continuously adjusted through the self-adaptive inertia weight; and in each iteration, the fitness of each particle is calculated, and the individual optimal position and the global optimal position of the particle are updated. When the maximum number of iterations or the fitness threshold value of the particles is reached, iteration is terminated, the optimal matching position is output, and finally the final position of the user is calculated through the geomagnetic sequence. According to the technical scheme, the optimal alignment path of the two sequences in the time dimension can be found by using the particle swarm and the simulated annealing algorithm without increasing the calculation amount, the matching precision is improved, and the calculation time is shortened.

This invention discloses a multi-parameter monitoring method and system for seismic isolation bearings. The method includes: dividing multiple seismic isolation bearings into associated node groups and specifying a reference node and reference time; calculating in real time the time difference between the slave nodes and the reference node in the physical data within the physical associated node group, determining whether there is time drift in the slave nodes (the slave nodes are nodes other than the reference nodes); marking the slave nodes with time drift as nodes to be repaired, finding the optimal alignment path between the physical data sequence of the node to be repaired and the template sequence; establishing a nonlinear time mapping function; and being able to perform full waveform correction for complex time distortions such as delay, stretching, and compression, thereby suppressing the formation of data tearing at its source.