19 results about "Local pattern" patented technology

The invention relates to a sketch image retrieval method based on a style adaptive residual network, and belongs to the field of computer vision. The method comprises the following steps: firstly, acquiring a fine-grained sketch data set, and performing feature extraction through a style adaptive residual network; in the network, firstly, features with important semantic information are enhanced by using a channel attention mechanism through a style enhancement door module, so that low-response but key global or local features are effectively transmitted. Then, a style enhancement module is used for adaptively adjusting style enhancement features through a feature relation matrix and learnable transformation operation, and it is ensured that the model does not excessively depend on a local mode of a specific style, but matching can be carried out on the basis of overall structure information; finally, a style adaptive residual network is constructed, so that the model can adaptively adjust feature representation, high robustness is kept under sketch input of different styles, and it is ensured that stable and discriminant depth features can be obtained in the matching stage.

The invention provides a defective mask substrate screening method and system, a medium and a terminal.The defective mask substrate screening method comprises the steps that defect information of a defective mask substrate is obtained; obtaining target layout data; setting different relative rotation angles for the defective mask substrate and the target layout, and carrying out matching analysis: B1, if the defect is a shading type defect, when the local graph density is 0%, judging that the defect does not influence the utilization of the defective mask substrate; b2, if the defect is a light-transmitting defect, judging according to the type of the photoresist adopted by the defective mask substrate: when the positive photoresist is adopted, and if the local pattern density is 100%, judging that the defect does not influence the utilization of the defective mask substrate; and when the negative photoresist is adopted, if the local pattern density is 0%, judging that the defect does not affect the utilization of the defect mask substrate. Whether the defect mask substrate can be used or not is converted into calculation of the local pattern density, and the detection efficiency is greatly improved.

The application provides a box recognition method, system and device based on multi-view vision and a storage medium, and the method comprises the following steps: obtaining a first image and a second image at different angles; inputting the first image and the second image into an image recognition neural network respectively to obtain local pattern information, corner point position, parallax information and 3D matching information representing a box; inputting the local pattern information and the corner point position into a context adjustment layer of a character detection neural network to obtain stereo information corresponding to the local pattern representing the box; and obtaining a spatial box with relative positions based on the corner point position and the stereo information. The application can enhance the algorithm recognition effect, has good real-time effect, consumes less computing resources, is convenient to deploy on a resource-limited vehicle-mounted edge device, is not prone to causing misrecognition and collision, and improves driving safety.

The application discloses a pattern defect detection method and device and a storage medium. The method comprises the following steps: obtaining an overall pattern region and a plurality of local pattern regions corresponding to a target pattern in a to-be-detected image, and obtaining an overall template image and a plurality of local template images from template information corresponding to the target pattern; aligning the overall pattern region with the overall template image, and respectively aligning each local pattern region with a corresponding local template image; determining a first defect position in the overall pattern region based on a first pixel value difference between the aligned overall pattern region and the overall template image, and respectively determining a second defect position in each local pattern region based on a second pixel value difference between the aligned each local pattern region and the corresponding local template image; and comprehensively determining a third defect position of the target pattern in the to-be-detected image based on the first defect position and the second defect positions. Through the above method, the accuracy of pattern defect detection can be improved.

The application provides a ball toy surface spraying device with a limiting structure and relates to the technical field of robot spraying. The application discloses a ball toy surface spraying device with a limiting structure and relates to the technical field of robot spraying. The ball toy surface spraying device with the limiting structure comprises a movable part, which is used for positioning the position, which does not need to be sprayed, downward in advance. Meanwhile, the position of the movable plate on the bottom rod is changed by rotating an adjusting screw, and then the activity range of the swing part in the assembling groove is adjusted. When spraying, the swing part drives the spherical toy to rotate and swing according to a limited angle, so that the spherical toy is only partially exposed to the nozzle, and thus the accurate spraying of the specific area of the spherical body is realized. In this way, the diversified design requirements of local pattern forming and functional area division can be met. The existing spraying process can only realize the full-coverage spraying of the whole outer surface of the toy ball and cannot accurately limit the spraying area according to the production requirements (such as local pattern forming, specific functional area division and the like). Therefore, the diversified product design requirements cannot be met, and the problem of paint waste and increased production cost caused by unnecessary spraying may occur.

This invention relates to a method for predicting non-histone acetylation sites that integrates local convolutional modulation and global self-attention. It addresses the common problems in existing non-histone acetylation site prediction techniques, such as high experimental screening costs, insufficient sensitivity of computational methods to local short motifs, inadequate long-range dependency modeling, and unstable model generalization due to positive-negative sample imbalance. The method includes: S1, data acquisition and processing; S2, candidate site sample construction; S3, embedding and positional encoding; S4, classification model construction; S5, model training and optimization; and S6, model performance evaluation. The advantages of this invention are: maintaining input simplicity and high inference efficiency while achieving collaborative modeling of local patterns and global context, and improving minority class recognition ability and overall stability through an imbalanced training strategy.

The invention discloses a card mixing detection method and system based on visual features, and belongs to the technical field of machine visual detection, and the method comprises the steps: carrying out the multi-station image cooperative collection to obtain front and side images of cards; performing multi-level semantic feature extraction to obtain global layout, local patterns, character contents and edge texture features; according to hierarchical goods mixing judgment, series-level coarse screening, variety-level fine screening and version-level fine screening are executed in sequence; according to the material isomerism detection, non-visual visible mixed goods are identified through near-infrared and polarization imaging; according to the invention, the problem of accurate detection of all types of mixed goods in a card packaging production line is solved.

1. The name of the design product: the decorative layer of mobile phone protective case. 2. The use of the design product: the overall use of the design product is to protect the mobile phone; the local use is to decorate the mobile phone case, and the mobile phone and the support are fixed through the magnetic attraction ring on the back. 3. The design points of the design product: the combination of the local pattern and color to be protected. 4. The picture or photo that best indicates the design points: design 1 perspective view 1. 5. The design to be protected contains color. 6. Design 1 is designated as the basic design. 7. Other circumstances that need to be explained: the unprotectable part is represented by the semi-transparent blue coating area in the figure.

The application discloses an industrial process zero-sample soft measurement method based on retrieval enhanced large language model. The application uses known working condition data as a retrieval knowledge base; by identifying historical segments with invariant local patterns from the knowledge base and the target query, robust features are extracted using a multi-granularity representation module to overcome noise interference, and a parameter efficient fine-tuning strategy is used to bridge the modal gap between industrial signals and language models. In the training stage, the application does not need to contact the target working condition data, and only relies on the experience prior provided by retrieval to realize instant generalization and high-precision prediction of unknown states, effectively solving the data distribution deviation problem caused by the multi-mode characteristics of industrial processes, and significantly improving the adaptability and reliability of the soft measurement model in unknown environments.

The invention provides a clothing version intelligent adaptation method, a server, a medium and a product, and relates to the technical field of computer assistance. The method comprises the following steps: firstly, disassembling newly obtained costume design CAD data to obtain component-level data; attribute characteristic data such as component categories and style labels in the data are called, and indexes are established and associated to a clothing component database. Monitoring the current canvas of the designer in real time, and determining target clothing areas such as a collar area and cuff areas; and based on the database, in combination with the attribute characteristics of the target area and other areas of the canvas, retrieving and generating an inspiration window recommendation candidate matching list and displaying the list. And when an operation of selecting the target candidate collocation by a designer is received, calling the corresponding component-level data, and replacing the original version data of the canvas target area. According to the whole process, efficient reuse of local version resources is achieved, the design period of new style development is shortened, meanwhile, the process attribute consistency of reuse parts is guaranteed, and the efficiency and quality stability of garment version design are improved.

Aspects of the present disclosure include methods for generating a sampled profile including a plurality of sampling points having a plurality of characteristic values associated with the detected non-visible light, identifying one or more macroblocks each includes a subset of the plurality of sampling points, calculating a number of occurrences of the local pattern value within each subset of the plurality of the sampling points for each of the one or more macroblocks, generating a first array including a plurality of weighted values by calculating the plurality of weighted values based on the numbers of occurrences of the local pattern value and corresponding sizes of the one or more macroblocks, assigning a unique index to each of the plurality of weighted values, generating a second array of the unique index by ranking the plurality of weighted values, and generating a third array including a plurality of ranking distances.

The invention discloses a heterogeneous log analysis method and system based on multi-feature fusion, and the analysis method employs a data optimization mechanism based on log features to carry out the preprocessing and enhancement of input log data, so as to reduce the data noise and extract more representative features. Secondly, the analysis method combines a plurality of feature measurement methods, including sequence features, n-gram features and Jaccard similarity features, and comprehensively utilizes the advantages of the features through a feature combination mode. Specifically, the sequence features are used for capturing the sequence patterns of the logs, the n-gram features are excellent in local pattern matching, and the Jaccard features can measure the global similarity between the logs. By fusing the features, the analysis method can more accurately analyze and cluster similar log events. Moreover, a dynamic template generation mechanism is introduced into the analysis method, and a template generation strategy is automatically adjusted according to characteristics of different log sets so as to cope with log data of different formats and complexity.

The invention relates to the technical field of building construction quality management, and discloses an intelligent management system for a premixed concrete pouring construction site. The system comprises a data acquisition and verification module, a multi-source data fusion module, a feature learning module, a dynamic calibration module, an adaptive partition module, an association mining module and an intelligent decision module. The system collects and verifies sensor data of a construction site in real time, carries out fusion and feature extraction, and calibrates a quality state by using a dynamic standard feature library. And when the calibration result is uncertain, the system can adaptively divide the construction area into a plurality of analysis units according to the feature distribution so as to perform local pattern mining and correlation analysis, and finally trigger early warning based on the risk score. The problems that in the prior art, due to the fact that static standards are used, adaptability is poor, and accurate positioning cannot be carried out when global evaluation is uncertain are solved, and dynamic, accurate and intelligent management on the concrete pouring quality is achieved.

The application discloses a teaching data dynamic classification processing method and device and a storage medium, constructs a composite modeling architecture based on combination of a multi-head attention, a long short-term memory network and a hidden Markov model, extracts global time sequence dependent features through the multi-head attention mechanism, and performs feature-level fusion with local state features, so that the limitation of a single model in structured data modeling is solved. Secondly, a Tra-BiLSTM double-encoding feature enhancement module is proposed, on the basis of a pre-training model, the global interaction advantage of a transformer and the local bidirectional modeling capability of a BiLSTM are combined, multi-level semantic feature extraction of Chinese time sequence data is realized, and the joint capturing capability of long-distance dependence and local patterns is significantly improved. Thirdly, a dynamic weight multi-task optimization mechanism is designed, the gradient conflict and the unbalanced optimization target existing in multi-task learning are solved, and the overall training efficiency and the generalization performance of the model are improved.

This invention relates to the field of character recognition technology, specifically to a method and system for recognizing Chinese characters in multiple fonts based on a pre-trained graph model. The method includes acquiring original character images and performing graph construction processing to generate structured character graphs; pre-training a graph neural network based on an unlabeled multi-font dataset by introducing a computational tree as a transferable local pattern, and constructing a discrete tree vocabulary using a vector quantization mechanism; wherein the graph neural network uses GraphSage as the backbone network; in the fine-tuning stage, the pre-trained graph encoder and tree vocabulary are loaded, and a residual mechanism is introduced to fuse the original graph embeddings and quantized graph embeddings, and a dedicated learning model for multi-font recognition tasks is obtained through fine-tuning on a target dataset; the dedicated learning model is used for multi-font generalization recognition; this invention effectively alleviates the intra-class discrepancy problem caused by font style differences and improves the model's recognition generalization ability in multi-font scenarios.

The embodiment of the invention provides a master control display device, a controlled display device and a reverse screen transmission method, the master control display device responds to a touch screen operation, displays a first local pattern on a first display interface and generates operation information corresponding to the touch screen operation, and the first local pattern comprises a sub-local pattern generated based on the touch screen operation; controlling the communicator to send the operation information to the controlled display equipment; controlling the communicator to receive image information which is sent by the controlled display equipment and is obtained based on the operation information; deleting at least a part of sub-local patterns contained in the first local pattern based on the image information to obtain remaining sub-local patterns; and displaying a second local pattern on the first display interface, wherein the second local pattern comprises a sub-pattern contained in the image information and the remaining sub-local patterns of the first local pattern. According to the embodiment of the invention, a user feels that the contents displayed on the master control display equipment and the controlled display equipment are synchronous and consistent, and the user experience is improved.

An identity recognition method includes: acquiring biometric pattern information describing a biometric pattern of a first object; performing feature extraction on the biometric pattern information to obtain a global pattern feature and a local pattern feature; fusing the global pattern feature and the local pattern feature to obtain a fused pattern feature of the first object; and performing identity recognition on the first object based on the fused pattern feature of the first object.

The invention relates to the technical field of printed matter quality control, and discloses a microstructure master mask repairing method which comprises the following steps: detecting the flatness of a microstructure master mask, and screening a to-be-repaired master mask; placing the mother set in a solvent to remove impurities on the surface; a mold pressing sample sheet is obtained, and a microstructure defect spot area on the mold pressing sample sheet is marked; carrying out hollowing-out treatment on the microstructure defect spot area on the mould pressing sample sheet, and manufacturing a mask and a repair file according to the position and the size of the microstructure defect spot area; the mask is attached to the mother set to be repaired, the micro-structure defect spot area is exposed out of the hollowed-out area, and the mother set attached with the mask is placed in material filling equipment for material filling; according to the method, the mask on the mother set to be repaired is removed, and local pattern photoetching and microstructure reconstruction are carried out on the repaired area of the mother set, so that the damaged microstructure on the mother set can be accurately recovered, the problem that a traditional microstructure mother set cannot be repaired is solved, the consistency and quality stability of microstructures among batches are improved, and the production cost is reduced. And the quality consistency between batches is ensured.