2002results about "Still image data clustering/classification" patented technology

A method includes receiving training data comprising medical images and associated text reports. The method further includes training an image encoder to associate image embeddings generated by the image encoder when the image encoder is applied to the medical images with text embeddings generated by a text encoder when the text encoder is applied to the associated text reports. The method additionally includes training a vision-language model based on the image embeddings generated by the trained image encoder to represent the medical images and the associated text reports in a language-aligned embeddings space. The method further includes training an adaptor which connects the vision-language model to an LLM to generate LLM-aligned tokens which emphasize portions of medical image information to the LLM via an attention mechanism. The method further includes providing the trained image encoder, the trained vision-language model, and the trained adaptor to perform one or more inference tasks.

A method of training a text model using a plurality of text passage combinations, each text passage combination comprising a respective first text passage and a respective second text passage describing a same matter as the respective first text passage but being differently worded than the respective first text passage. The training comprises minimizing a measure of statistical difference between a respective value of a first text embedding and the corresponding value of a second text embedding over the plurality of text passage combinations. The method then comprises jointly training the text model and an image model based on plurality of image-text combinations, each comprising a respective image and a respective textual report describing the respective image. The joint training comprises minimizing a measure of statistical difference between the value of an image embedding and the corresponding value of a third text embedding over the plurality of image-text combinations.

Systems and methods for classification model training can use feature representation neighbors for mitigating label training overfitting. The systems and methods disclosed herein can utilize neighbor consistency regularization for training a classification model with and without noisy labels. The systems and methods can include a combined loss function with both a supervised learning loss and a neighbor consistency regularization loss.

The invention relates to the technical field of Otwei interactive maps, and discloses a geological data interaction method and system based on an Otwei interactive map. The method comprises the following steps: carrying out feature recognition and structural analysis on original geological data to obtain a standardized data packet, and establishing a geographic space reference conversion index table; performing feature extraction and classification on geological elements in the standardized data packet to obtain structured geological data; importing the structured geological data into an Ovoucher interactive map, and carrying out local registration through a mesh generation technology to obtain a visual geological element map layer; geological element drawing and attribute input are carried out, and an edited geological data set is obtained; and carrying out structure recombination and reverse coordinate conversion to obtain a standard format data file adaptive to the target geological information system. According to the method, accurate conversion of multi-source geological data between different coordinate systems and measuring scales is realized, and the problem of spatial dislocation during integration of different-source geological data in a traditional method is effectively solved.

The invention discloses a knowledge distillation method, electronic equipment and a computer readable storage medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: carrying out the grouping of an attention map matrix of a teacher network based on the number of attention heads in an initial network structure of a student network, a high-dimensional attention space of a teacher network can be divided into subspaces with the same dimension as a student network, and attention map matrixes in groups are spliced, so that each group of spliced matrixes is aligned with an attention head of an initial network structure, a relatively regular corresponding relationship between the attention maps of the teacher network and the student network is realized, and the attention map splicing efficiency is improved. Therefore, the distillation loss can be calculated, and the knowledge of the complex teacher network can be accurately transmitted to the lightweight student network. Therefore, the problem that in a multi-head attention mechanism, the number difference between the teacher network and the student network hinders the student network to fully learn the teacher network knowledge can be solved, and the technical effects of improving the knowledge distillation effect and enhancing the student network performance are achieved.

The invention discloses a safe and rapid storage method and system for image data of an imaging department, and relates to the technical field of medical image data processing. The method comprises the following steps: 1, designing a distributed storage system, and selecting and configuring storage nodes; 2, selecting and updating an encryption algorithm, and implementing file-level and block-level encryption; 3, realizing accurate data deduplication by utilizing a machine learning technology, and selecting a proper compression algorithm; 4, constructing an index system, and establishing a data warehouse and a data lake by using big data analysis and machine learning technologies; 5, establishing an image data backup and recovery mechanism; and 6, constructing a data integrity checking and monitoring system. According to the invention, by adopting the technologies of distributed storage, encryption, data deduplication, compression, big data analysis, machine learning, backup and recovery, data integrity verification, monitoring and early warning and the like, efficient storage and safety protection of image data are realized.

In implementations of systems for searching for images using generated images, a computing device implements a search system to receive a natural language search query for digital images included in a digital image repository. The search system generates a set of digital images using a machine learning model based on the natural language search query. The machine learning model is trained on training data to generate digital images based on natural language inputs. The search system performs an image-based search for digital images included in the digital image repository using the set of digital images. An indication of the search result is generated for display in a user interface based on performing the image-based search.

An online concierge system trains a fine-tuned generative image model for distinct categories of items based on a generative image model that takes a textual query as input and outputs and an associated image. Training of the fine-tuned generative image model is additionally based on a small set of representative images associated with the various categories, as well as textual tokens associated with the categories. Once trained, the fine-tuned generative image model can be used to generate realistic representative images for items in a database of the online concierge system that are lacking associated images. The fine-tuned model permits the generation of different variants of an item, such as different quantities or amounts, different packaging or packing density, and the like.

The present application discloses an image generation method and apparatus, a device, a storage medium, and a vehicle. The method comprises: obtaining a prompt word input by a user, wherein the prompt word is used for instructing to generate a target image; on the basis of a preset first correspondence, determining an image generation model and a sampler corresponding to the prompt word, wherein the first correspondence is a correspondence among the prompt word, the image generation model and the sampler, and the sampler is used for controlling a noise reduction mode of the image generation model; performing text expansion on the prompt word to obtain a final text; loading the sampler into the image generation model to update noise reduction parameters of the image generation model; and using the image generation model with updated noise reduction parameters to convert the final text into the target image. According to the embodiments of the present application, the quality of generated images can be improved.

The invention discloses a method for detecting industrial hazardous gas leakage based on a multispectral infrared imaging system formed by a micro-lens array and a partition optical filter, which is used for solving the problem that the current infrared imaging equipment cannot identify various leaked gases, and belongs to the technical field of infrared imaging. The multispectral infrared imaging system is constructed on the basis of the microlens array and the partition optical filter, multispectral band design is carried out on common gas in the current petrochemical industry field, and the types of common leaked gas are identified by means of spectral characteristics of the leaked gas in an infrared multispectral channel. Firstly, a snapshot type infrared imaging system is constructed based on a micro-lens array and a partition optical filter, and the system comprises a plurality of spectral intervals which respectively correspond to optical absorption peaks of specific gas; imaging the hazardous gas based on a multispectral infrared imaging system to obtain a multispectral infrared image of a hazardous gas leakage scene; and finally, based on a foreground detection method, extracting gray information of the gas cloud target, constructing a spectral absorption normalization curve of the gas cloud target in each channel, and comparing the spectral absorption normalization curve with the spectral absorption curve in the database to judge the gas type of the current gas cloud target. The invention provides a reliable multispectral imaging scheme for a hazardous gas imaging system based on infrared imaging, gas types can be effectively distinguished during actual detection, interference of safe gases such as carbon dioxide and water vapor is eliminated, and the detection capability and practicability of an infrared imaging hazardous gas detection system are improved.

The invention provides a method and a device for generating a parameterized model, a medium and a program product. The method comprises the following steps: S1, receiving text data which is input by a user and is related to a design intention or a command; s2, preprocessing the text data through a predetermined natural language processing algorithm to obtain structured parameter information of the text data; and S3, constructing a parameterized model through predetermined modeling software according to the structured parameter information. By means of the technical scheme, the modeling efficiency can be improved, the quality and consistency of the model are guaranteed, and the maintainability of the model is further enhanced.

Systems and methods include extracting item parameters from images of items positioned at a POS system. Item parameters associated with each item when mapped into a feature vector for each item are indicative as to an identification of the item. The feature vectors are analyzed to determine whether item parameters when combined and mapped into the feature vectors match a corresponding feature vector stored in a database. The database stores different combinations of item parameters as mapped into different stored feature vectors with different stored feature vectors associated with different items thereby identifying each item based on each different combination of item parameters as mapped into each stored feature vector associated with each item. Each item positioned at the POS system is identified when the feature vectors associated with each item match a corresponding stored feature vector as stored in the database.

The invention relates to an operation and maintenance decision-making system and method based on a large language model and a graph database, the system comprises an intelligent operation and maintenance platform, a large model layer and an operation and maintenance database, the intelligent operation and maintenance platform is connected with the large model layer and the operation and maintenance database, and the intelligent operation and maintenance platform is used for receiving a user query demand and sending the user query demand to the large model layer. After a task instruction is automatically generated, an operation and maintenance tool is called to execute query analysis operation, and an operation and maintenance processing result is correspondingly returned; the large model layer is used for providing various large model bases so as to correspondingly process task instructions output by the intelligent operation and maintenance platform; and the operation and maintenance database is used for storing the operation and maintenance topological relation data, the operation and maintenance historical event data and the operation and maintenance knowledge data so as to provide a data source for the intelligent operation and maintenance platform. Compared with the prior art, the method has the advantages that the query result can be quickly and accurately obtained, the component fault influence can be determined, the historical operation and maintenance event can be queried and matched, and the operation and maintenance decision efficiency and accuracy are improved.

The invention provides a noise robust text-to-image pedestrian retrieval method and device based on a soft label. The method comprises the following steps: calculating cosine similarity based on image global features and text global features, and normalizing the cosine similarity to generate a soft label representing image-text pairing confidence; distributing a sample weight for each training sample according to the soft label, and obtaining a joint weight for current iteration in combination with a dynamic weight factor progressively increased along with the training process; respectively constructing cross-modal contrast learning loss and similarity distribution matching loss by using the joint weight, and carrying out weighted summation on the cross-modal contrast learning loss and the similarity distribution matching loss to obtain a total loss function; and updating parameters of the image encoder and the text encoder by using the total loss function until training convergence, and obtaining a cross-modal alignment model. According to the method, robust cross-modal alignment can be realized, and the pedestrian retrieval accuracy in a noise scene is improved.

The invention relates to the field of data visualization, in particular to a visual scheduling method and system for a scientific and technological achievement evaluation system. The method comprises the following steps: identifying system access behavior data of a user; performing attention hotspot field prediction on system access behavior data of the user, performing deep personalized demand evolution, and constructing a personalized scientific and technological achievement demand graph; real-time data flow matching retrieval is carried out according to the personalized scientific and technological achievement demand graph, semantic feature analysis is carried out, and deep semantic features of each text item and visual semantic features of each image and table item are extracted; and performing dynamic clustering processing according to the deep semantic features of each text item and the visual semantic features of each image and table item, and performing multi-dimensional scientific and technological achievement item aggregation to construct a scientific and technological achievement display library. According to the invention, comprehensive, dynamic, accurate and personalized scientific and technological achievement evaluation visualization is realized.

The invention discloses a customs intelligent specimen cabinet monitoring system, relates to the technical field of specimen cabinet monitoring, and aims to solve the problems that the monitoring strength of an existing specimen cabinet is reduced, and abnormality checking cannot be carried out in time. According to the invention, the most suitable transmission channel is flexibly selected through different transmission requirements and environmental conditions, the adaptability and flexibility of the system are enhanced, the feature extraction of the environmental monitoring data covers the monitoring parameters of each sensor, and the comprehensive monitoring of environmental factors is ensured. The monitoring of real-time video data provides intuitive visual information, so that workers can intuitively know the real-time condition in the specimen cabinet, manual intervention is reduced through automatic detection, recognition and alarming, the automation and intelligence level of the monitoring system is improved, and the monitoring efficiency is improved through automatic data classification, visual display and anomaly analysis. And the intelligent level of the customs intelligent specimen cabinet monitoring system is improved.

A method comprising generating image descriptions of images in an original training set of images; determining, using at least one LLM, at least one domain and / or class which is under-represented in the original training set; generating, using a second LLM and based on the determination of the at least one domain and / or class, at least one instruction for a third LLM to generate at least one text prompt; generating, using the third LLM and based on the at least one instruction, the at least one text prompt for a text-to-image model; generating, using the text-to-image model and based on the at least one text prompt, at least one synthetic image; and generating an enhanced training set of images for use in training an image processing machine learning, ML, model, the enhanced training set of images comprising the original training set of images and the at least one synthetic image.

The invention discloses a method for generating a three-dimensional building GLTF file based on index compression. The method comprises the following steps: S1, collecting three-dimensional coordinate data of a building; s2, grouping and classifying the data according to floors to form a classification index structure based on a hash table; s3, selecting a reference model and recording Mesh data and vertex coordinates of the reference model; s4, calculating a translation matrix T of the polyhedral structure, and performing coordinate transformation; s5, generating Mesh data of other polyhedrons based on index compression, and reducing redundant surface data by adopting a topology preserving surface optimization algorithm; s6, independently processing the geometric data of each floor, and generating corresponding Mesh data and Node nodes; and S7, combining all floor data to generate a complete GLTF file, and performing compression by using a dynamic Huffman coding algorithm. Through the improved index compression and dynamic Huffman coding technology, the file size is remarkably reduced, the transmission efficiency is improved, and the geometric and topological integrity of the model is ensured.

The invention discloses a multi-modal power grid fault diagnosis method and system based on a causal event atlas, and belongs to the technical field of intelligent operation and maintenance of power systems. The method comprises the following steps: preprocessing historical fault case data of power grid equipment, and constructing a causal event atlas database; when a fault diagnosis request is received, analyzing the fault diagnosis request by the planning agent, generating an initial fault hypothesis set in combination with power field knowledge, endowing a corresponding credibility score to the fault hypothesis, retrieving the cause subgraph as evidence in an iterative loop, and updating the credibility score of the fault hypothesis by the reasoning agent; and generating a multi-modal diagnosis report after the termination condition is met. According to the method, the problems of insufficient causal modeling and poor interpretability of a traditional method are solved, and the diagnosis accuracy, efficiency and user credibility are remarkably improved.

The invention provides a city physical examination method and device and electronic equipment. The method comprises the following steps: acquiring two-dimensional map data and overground and underground full-space three-dimensional model data of a target city; grouping components in the two-dimensional map data and the above-ground and underground full-space three-dimensional model data according to a preset business dimension to obtain a plurality of component groups corresponding to the target city; based on the multi-dimensional component information of the plurality of component groups, constructing a knowledge graph corresponding to the target city; performing city physical examination on the target city based on the knowledge graph to obtain a city physical examination report; and generating a three-dimensional space-time thematic map corresponding to the target city based on the city physical examination report, the two-dimensional map data and the overground and underground full-space three-dimensional model data. According to the invention, the accuracy and diversity of urban physical examination question judgment can be improved, and the three-dimensional space-time thematic map can be automatically generated.

The invention discloses a cross-domain expression motion unit detection method based on text bridging, and the method comprises the steps: extracting AU visual features of different scales through employing a visual encoder, introducing a group of learnable prompt vectors, and combining with AU text description to obtain AU text features; the source domain features and the text features are aligned through comparative learning, and the discrimination of the source domain AU visual features is improved; a cross-modal attention mechanism is adopted to realize preliminary interaction of vision and texts, and AU interaction feature representation is obtained; further promoting deep interaction and fusion of the text and the visual information through a graph neural network to obtain a fused AU feature; and on the basis of a negative sample AU description-based comparative learning method, target domain features are aligned with text features, and feature distribution differences between a source domain and a target domain are reduced. According to the method, the field invariance of the text is utilized, the visual features of the source domain and the target domain are unified, the text features are aligned, and the generalization performance of an AU detection model and the accuracy of the AU detection model on the target domain are remarkably improved.

A data processing method, relating to the field of artificial intelligence, and comprising: acquiring an image and an object detection request, the object detection request containing a natural language description of an object to be detected in the image; by means of an image encoder, processing the image to obtain a feature representation of each image region among a plurality of image regions, each image region corresponding to a candidate detection box; and, by means of a language model, processing the object detection request and the multiple feature representations to determine from among the plurality of image regions a region where said object is located and the category of said object. The present application uses the image encoder to obtain fine-grained features, i.e. the features of each image region, and performs object detection on the basis of the language model, thereby improving the capability of processing the fine-grained features and the precision of object detection under the guidance of human languages.

Embodiments of the present invention provide computer-implemented methods, computer program product, and computer systems. One or more processors assign an identifier that specifies a number of resources and a category associated with a respective image layer of a plurality of image layers. One or more processors, in response to receiving a user request, identify image layers of the plurality of image layers that match the identifier based on dependencies between the plurality of image layers. One or more processors can retrieve matched layers based on the functionality of respective image layers and the dependencies of those respective image layers.

The invention discloses an unmanned aerial vehicle-satellite scene matching method based on two stages. Precise scene matching and positioning are completed from coarse granularity to fine granularity. An existing satellite base map retrieval task is used as a first-stage task, and a local feature point matching method is introduced in a second stage. Compared with an existing method, the method has the advantages that the matching robustness in a complex scene is remarkably improved, meter-scale accurate positioning of the unmanned aerial vehicle can be realized on an existing satellite map, and a new universal solution is provided for the field of scene matching.

The invention discloses a lightweight gateway electric energy meter thermal defect target detection method based on improved YOLOv8s. The method comprises the following steps: firstly, extracting features by using MobileNetV4 as a backbone network; secondly, the Neck part is replaced with a CCFF feature fusion module in the RT-DETR model; in addition, a light-weight C2fStar module is provided in combination with the Star Block thought to replace an original C2f module; meanwhile, a lightweight shared convolution detection head LSCD is introduced; then, an original CIoU loss function is replaced by a Focaler-SIOU (System Identity of Units); and finally, pruning the network model by means of an LAMP mode. The method has the characteristics of light weight, real-time performance, high accuracy and the like.

The invention discloses a three-area three-line database automatic quality inspection evaluation method based on deep learning. The method comprises the following steps of S1, obtaining three-area three-line database data and performing preprocessing; s2, checking a directory structure and a file name of the original data set; s3, extracting spatial layer data from the hierarchical data set; s4, performing segmentation and attribute information extraction on various pattern spots in the layer data after spatial verification; s5, inputting the pattern spot relation characteristic data into an artificial fish swarm algorithm; s6, automatically comparing the pattern spot area and distribution data with a quality inspection standard preset in a three-area three-line database; and S7, performing parallel processing on an automatic comparison result through a multi-thread thread pool. According to the method, an artificial fish swarm algorithm, a relational attention network and a task allocation and thread pool management technology are combined, and three-area three-line database automatic quality inspection evaluation based on deep learning is realized.

The embodiment of the invention provides a recommendation method and device based on a large language model, and the method comprises the steps: obtaining at least one main body inquired by a user, and enabling the at least one main body to be obtained through recognition from a target image; obtaining first information of the user, wherein the first information is used for indicating the interest tendency of the user; determining a target subject in the at least one subject based on the first information; and through the large language model, based on the target subject, generating at least one item of recommended content for the user.

The invention discloses an image retrieval system and method based on color moments and deep learning features. The system comprises an image feature extraction module, an index construction module, a retrieval trap door generation module and an image retrieval module. The image feature extraction module extracts color moment features and deep learning features of images in the image data set to obtain image multi-level features; the index construction module is used for constructing an index tree from bottom to top for color moment clustering; the trap door generation module extracts hierarchical features of the query image to generate a retrieval image trap door; the image retrieval module traverses the index tree from top to bottom according to the retrieval trap door to obtain a similar image candidate set, calculates the similarity between the depth feature of the image in the candidate similar image set and the depth feature of the query image, and outputs top-k images similar to the query image; according to the method, the color moments of the images are extracted and clustered, and the index tree is constructed, so that the similarity between the comparison images and the query images does not need to be calculated one by one during retrieval, and the image retrieval efficiency is improved.

The invention discloses a dance costume prop retrieval method based on causal reasoning and cross-modal matching, which comprises the following steps: in a matching task, calculating the similarity between costume prop images and text matching features by a model, and judging whether the costume prop images and the text matching features are matched or not according to a threshold value; and when the task is retrieved, calculating the similarity between the query text and the image feature set, and sorting to obtain a result. The process of calculating the features and the similarity comprises the following steps: firstly, extracting multi-scale visual features of an image by using Faster R-CNN, and extracting multi-granularity semantic features of a text by using BERT; then, the two types of features are subjected to self-attention and gating fusion respectively and then are input into a Transform layer, and final visual and text representation is obtained; then, multi-head attention calculation is carried out on the two representations, self-attention calculation is carried out after sequences are combined, and matching features are obtained; and finally, dividing the dot product of the two feature vectors by the modular length product to calculate the similarity. According to the invention, the matching accuracy of dance costume props can be improved.

The invention relates to the technical field of wastewater treatment, in particular to a regulation and control method and system for an activated sludge process and electronic equipment. The method comprises the following steps: acquiring activated sludge property evaluation information; matching the activated sludge property evaluation information with data in an activity standard database of biochemical sludge; and implementing negative feedback optimization of the working condition of the activated sludge system on the activated sludge property evaluation information which is unqualified in matching with the activity standard database of the biochemical sludge. According to the regulation and control method of the activated sludge process, stable and reliable operation of the activated sludge process is achieved according to the constructed activity standard database of biochemical sludge, parameters such as system regulation and control logic and signal intensity of the method are subjected to mass data adjustment and calibration, self-adaptive regulation and control are achieved based on real-time monitoring results, timeliness is higher, and the method is suitable for popularization and application. The actual operation condition can continuously tend to an ideal condition point, the system unbalance risk is remarkably reduced, and the labor cost is greatly reduced.