172 results about "Parallel encoding" patented technology

The invention provides an anomaly detection multi-classification method based on multi-source operation and maintenance data fusion. Comprising a data input layer, a parallel coding layer realized through dissimilatory multi-modal coding and a hierarchical multi-modal fusion architecture, a space-time feature fusion layer realized through a space-time perception dynamic gating attention enhancement mechanism, and a dynamic decision optimization layer realized through a gradient perception dynamic smooth loss function. The spatio-temporal feature fusion generates a feature representation and weight matrix with a dynamic attention weight through a spatio-temporal perception dynamic gating attention enhancement mechanism, and outputs the feature representation and weight matrix to the dynamic decision optimization layer; and the dynamic decision optimization layer realizes anomaly detection through a classifier taking a gradient perception dynamic smooth loss function as feedback, so that key problems such as multi-source heterogeneous data fusion, time sequence dynamic modeling and data label imbalance are solved, the anomaly detection accuracy and robustness of a training cluster are effectively improved, and the anomaly detection accuracy and robustness of the training cluster are improved. And a reliable technical support is provided for intelligent operation and maintenance of a complex training cluster.

The invention relates to the field of computer systems, and provides an intelligent identification method for flaws of a plush fabric based on multi-modal fusion. RGB, infrared and depth images during movement of the plush fabric are collected, multi-modal local features are extracted through a parallel encoder, and the multi-modal local features are fused into a joint feature map after semantic projection alignment; a lightweight YOLOv11 model is constructed, dynamic characteristics and a multi-scale mixed attention mechanism are combined, and flaw types, sizes and positions are recognized; setting a dynamic compensation mechanism to output defect labels and quality grades; according to the invention, through synchronous acquisition and feature alignment of multi-modal images, in combination with a lightweight YOLOv11 model and a multi-scale attention mechanism, multi-modal cooperative flaw detection of the plush fabric in a continuous moving state is realized.

The invention discloses an ultrasonic endoscope navigation system and method based on deep learning, and relates to the field of medical image analysis, and the system comprises a parallel encoder module which is used for extracting local features of an ultrasonic endoscope image through a CNN branch, and capturing global context information through a Transform branch; the channel attention fusion module is used for carrying out adaptive weighted fusion on the extracted features; a decoder module that generates an anatomical structure segmentation mask based on the fused feature; the time sequence processing module is used for receiving the frame-by-frame segmentation result output by the decoder module and realizing time sequence coherence analysis of the ultrasonic endoscope video through a bidirectional LSTM network; and the multi-modal fusion module is used for carrying out registration and fusion on the ultrasonic endoscope images. According to the scheme, a high-reliability artificial intelligence auxiliary tool can be provided for early screening of pancreatic cancer, meanwhile, the learning threshold and clinical application cost of the ultrasonic endoscope technology are reduced, and popularization of the ultrasonic endoscope technology in basic medical institutions is promoted.

The invention relates to a multivariable multi-step air conditioner load prediction model based on time sequence convolution and a double attention mechanism, and belongs to the technical field of air conditioner load prediction. The model adopts a parallel encoding structure, in an encoder, a time sequence convolution module is responsible for modeling local dependence and long-term trend in a time dimension, and a double attention mechanism module is used for modeling a dynamic dependence structure among multiple variables and a coupling relation between the variables and a target load from a variable dimension. The two structures are respectively subjected to feature extraction from a time dimension and a variable dimension, and are complementary to each other. In a decoder, a decoding module with a memory ability and a teacher mandatory strategy is designed, and continuous prediction from a historical multivariable sequence to a future target load is realized.

The invention discloses a data processing method and device of a data center network, equipment, a medium and a product, and relates to the technical field of data processing, and the method comprises the steps: obtaining network topology information and a data forwarding table rule corresponding to each equipment subnet in the data center network; performing parallel coding on each piece of network topology information and each data forwarding table rule based on a network parallel binary decision diagram to obtain a data forwarding path structure between subnets on each piece of equipment, the network parallel binary decision diagram being generated based on a lockless index and hierarchical cache; based on the path connectivity corresponding to the data forwarding path structure, constructing a data forwarding graph corresponding to the subnet on each device; and verifying the data reachability between the subnets on each device according to the data forwarding graph, and generating a data reachability verification result. According to the technical scheme, the concurrency limitation is overcome, high concurrency can be achieved, and the data plane verification efficiency of the data center network is improved.

The invention discloses a dynamic compression coding method, which is applied to a sparse matrix, and comprises the following steps: S1, obtaining the sparse matrix, and generating a bitmap with the same dimension as the sparse matrix, a value in the bitmap being used for indicating whether an element at a corresponding position in the sparse matrix is zero; s2, a mark sequence is generated according to the bitmap, and each mark corresponds to multiple continuous elements in the sparse matrix and is used for representing whether each element in the multiple elements is zero or not; s3, extracting non-zero elements in the sparse matrix according to the mark sequence; and S4, combining the flag sequence with the non-zero elements to form a data stream after compression coding. The invention also discloses a dynamic compression coding system, a storage medium and a device. According to the dynamic compression coding method, the index storage cost can be reduced, the data access efficiency can be improved, the storage and bandwidth occupation can be reduced by supporting parallel coding and decoding, and finally the calculation throughput can be improved.

The invention relates to the technical field of load prediction, and provides a power load prediction method and system based on time sequence decomposition and an attention mechanism, and the method comprises the steps: carrying out the adaptive time sequence decomposition of an obtained original load sequence, calculating the sample entropy of each decomposed component, and carrying out the clustering; constructing a group of encoder and decoder networks for each piece of clustered data, performing parallel encoding to extract features, performing serial decoding reconstruction on the features from low frequency to high frequency, and outputting prediction data from low frequency to high frequency step by step; the weight is initialized based on the sample entropy, the trained weight is obtained through optimization in the encoder and decoder network training process, and the predicted value of the power load is obtained through weighted fusion. According to the method, adaptive time sequence decomposition, a weight mechanism guided by sample entropy and an attention-enhanced encoder-decoder structure are introduced, so that multi-component collaborative modeling and cross-scale dynamic prediction are realized, and the prediction accuracy and stability in complex load data and small sample scenes are effectively improved.

The invention discloses a real-time intrusion detection system and method based on deep learning, and aims to solve the problem that when the existing deep learning intrusion detection technology processes high-dimensional network flow data, the information fidelity and the real-time processing efficiency are difficult to consider at the same time. According to the system and the method, a heterogeneous feature decoupling and fusion network architecture is adopted, high-dimensional feature vectors are losslessly decomposed into feature subsets of static state, time sequence, graph topology association and the like, deep learning is carried out by a parallel coding subsystem, cross-modal fusion attention self-adaptive aggregation is carried out, and finally classification judgment is realized. According to the scheme, on the basis of keeping the integrity of high-dimensional data, the detection capability and the real-time processing efficiency of unknown attacks are remarkably improved, and the self-adaptability and the robustness are improved.

The invention provides a point cloud semantic segmentation method and system based on parallel coding feature fusion, and belongs to the technical field of computer vision, a second module, a third module, a fourth module and a fifth module in a pre-constructed feature encoder each comprise an SA unit and an improved PEFA unit, the improved PEFA unit is used for carrying out deeper feature extraction and aggregation on the point cloud data after feature aggregation and enhancing the scaling capability of a model, and comprises a grouping layer, a parallel coding layer, an attention layer and a residual link, heterogeneous information of the point cloud is effectively fused through a local geometric coding and local context coding parallel encoder, and the scaling capability of the model is improved. And the relevance between attention mechanism dynamic learning features is introduced, feature aggregation is efficiently carried out, the problem of insufficient feature extraction and fusion of a single encoder is solved, and the semantic segmentation precision in a complex scene is improved.

The invention provides a guide registration method, system and device for RGB and infrared image fusion and a storage medium, and belongs to the technical field of image enhancement based on computer vision. Firstly, input of a visible light image and an infrared image is obtained, main target objects in the images are recognized and detected, and a preliminary target detection frame is generated; then, taking the target detection frame as spatial prior information, and generating an RGB image segmentation result and an infrared image segmentation result; performing feature matching on two segmented image results through a newly-designed cross-modal feature matching engine strategy, firstly extracting coarse-grained and fine-grained features through two parallel encoders, then obtaining a homography matrix through a global matcher, distortion refining and matching sampling, and finally obtaining a cross-modal feature matching algorithm; and then homography transformation is carried out on the original infrared image to obtain a final registration image. The method is superior to an existing method in the aspects of registration accuracy and stability, and a robust solution is provided for registration of infrared and visible light images.

The invention provides a large model dynamic adaptation and collaborative extraction method based on a unified framework. The method comprises the steps of preprocessing to-be-processed multi-modal data; dynamically identifying a modal type and carrying out adaptive coding to obtain a basic feature; the semantic anchor points are combined to generate fusion features through a parallel Transform encoder, feature distortion is synchronously monitored, and error correction is carried out; realizing cross-modal information interaction based on a cross-attention mechanism of an anchor point weight enhancement matrix, and generating a cross-modal feature vector; determining a differential fusion weight according to the type of the relationship between the modes, and obtaining a global fusion feature; semantic consistency is detected, and the weight is dynamically adjusted; semantic information is converted, fine-grained data is supplemented, and a preliminary result is formed; verifying and complementing attributes based on the knowledge graph; and generating a multi-granularity extraction result containing the global result and the fine-granularity grounding information. According to the method, efficient adaptation, accurate fusion and high-quality information extraction of multi-modal data can be realized.

The embodiment of the invention provides a time sequence prediction method and system based on multi-source feature fusion. The method comprises the following steps: receiving first time sequence data and second time sequence data; encoding the first time series data through an independent first encoder to generate a first encoding representation, the first encoding being a query vector; performing parallel encoding on the second time sequence data through an independent second encoder to generate a second encoding representation, the second encoding being a key vector and a value vector; inputting the first code and the second code into a cross-modal attention mechanism for fusion, and generating fusion feature representation; based on the fused feature representation, a target prediction time series is generated by a decoder. According to the embodiment of the invention, through independent feature extraction, mutual interference of multi-source information in a feature extraction stage is effectively avoided; through an intelligent fusion mechanism, the feature interaction capability exceeding that of a traditional fusion method is realized, and then the robustness and precision of a prediction result are enhanced.

The invention relates to the technical field of artificial intelligence, and discloses a separated multi-mode large language model service system and a first lexical element generation method. The system comprises a preprocessor, an encoder instance, an instance interaction layer and a pre-filling instance, the preprocessor is configured to analyze the question and answer request to obtain original data of multiple modals; the encoder instance is used for carrying out parallel encoding on the original data of each mode to generate a subsequence of the corresponding mode; the instance interaction layer is configured to send a sub-sequence of each mode generated by the encoder instance to the pre-filling instance; the pre-filling instance comprises a large language model trunk and is configured to perform asynchronous pre-filling on each sub-sequence of each mode by taking the sub-sequence as granularity so as to generate a first lexical element of an answer text corresponding to the question and answer request. By adopting the system, the response speed of the online question and answer service system can be improved, and the user experience is improved.

The invention discloses a camouflage target detection method based on multi-scale feature fusion and interference suppression, and the method comprises the steps: firstly constructing and training a camouflage target detection model which comprises a preprocessing network module, a difference feature interaction module, a detail feature fusion module, a boundary sensing module and an interference filtering module; and inputting a to-be-detected image into the detection model, wherein the output of the detection model is the camouflage target image obtained by detection. A progressive feature extraction architecture is adopted, in the initial feature extraction stage, a network constructs basic feature representation through multi-scale convolution operation, and a primary feature graph with global semantic information is generated. In the second stage, edge information is introduced to assist feature extraction, and in the third stage, interference filtering and refining processing is carried out on a feature map through IFM. According to the method, multi-scale features and interference suppression are fused, and the detection performance of the camouflage target is remarkably improved through parallel coding and multi-scale feature fusion in combination with edge feature information and an interference filtering module.

The invention discloses a hierarchical category refinement method and system for a remote sensing image semantic segmentation sample, and the method comprises the steps: (1), extracting feature information from a remote sensing image map and a coarse-grained label map, and obtaining image features and label features; step (2), performing enhanced fusion and grouping normalization processing on the image features and the label features to obtain fusion feature information; (3) the fused feature information is decoded and then preliminarily segmented, and a refined second-level semantic tag graph and a preliminary third-level semantic tag graph are predicted and output; and (4) further refining the three-level semantic tag graph by using a pre-trained visual-language large model guided semantic neural network model. The system comprises a double-flow parallel coding module, a bidirectional cross attention fusion module, a decoding module and a visual-language large model guided semantic graph neural network module. The method can solve the problems that in the prior art, remote sensing image label granularity is rough, and fine segmentation is difficult.

The invention relates to the technical field of network transmission, in particular to a parallel coding and decoding method based on a multi-channel network TCP-UDP (Transmission Control Protocol-User Datagram Protocol). According to the invention, by constructing a multi-path parallel transmission mechanism of a transmission control protocol and a user datagram protocol, the transmission performance of each path can be dynamically evaluated and quantified according to the round-trip time and packet loss rate of the path at the beginning of session establishment, and the initial data distribution proportion is scientifically calculated, so that the initial optimization utilization of network bandwidth resources is realized; in the data transmission process, the actual sending rhythm and the bandwidth utilization condition of each link are continuously monitored, and the evolution trend of the bandwidth load is accurately identified through a time sequence prediction model, so that the path weight is dynamically modulated, the distribution proportion of the data packet is corrected in real time, and the data stream can be continuously and uniformly transmitted on the optimal path. The problems of transmission bottleneck and delay caused by performance fluctuation of a single path are effectively solved, and the reliability, stability and efficiency of data transmission are greatly improved.

The application discloses a kind of parallel computing methods of FPGA acceleration for Key-Value storage system, method includes: triggering compaction operation, the SSTable of key range repetition is transferred to FPGA memory;Using FPGA parallel decoding and sorting these SSTable, obtain ordered and non-repeated key-value pair set;Using FPGA parallel coding this set, form new SSTable, and these SSTable are directly written from FPGA memory to SSD, bypass CPU, and the invalid key-value pair in log file is recycled by parallel garbage collection operation.The application parallel computing method of FPGA acceleration for Key-Value storage system, using key-value separation method reduces the write amplification problem of key-value storage, and provides comprehensive optimization for the calculation load and data transmission delay in key-value storage, reduces the pressure of CPU.

This invention discloses a knowledge- and data-driven time series forecasting method and system, belonging to the field of industrial time series forecasting technology. To address the technical problems of existing technologies when processing strongly non-stationary data, such as key signal attention dilution, systematic underestimation of peak amplitude, incomplete capture of multi-scale dynamic patterns, and difficulty in embedding physical domain knowledge, this invention constructs standardized multi-dimensional time series data, performs heterogeneous feature grouping and parallel encoding to obtain multiple sets of feature representations, then extracts and integrates features from multiple sources to obtain intermediate state features, and finally performs multi-branch prediction and adaptive weighted fusion based on the intermediate state features to obtain the final time series forecast result. This invention can achieve active focusing on abrupt events in industrial sequences and accurate characterization of extreme value amplitudes, significantly improving prediction accuracy, physical consistency, and robustness to extreme conditions.

This invention provides a lossless compression and decoding method for depth images. Lossless compression includes: dividing sequential data into multiple independent image blocks according to predetermined partitioning rules; selecting a corresponding thread model based on the characteristics of the depth image; calling a thread based on the thread model and using a reversible encoding algorithm to independently encode each image block; wherein, at the start of encoding for each image block, the encoder state is reset so that the encoding of each image block does not depend on the data of the preceding image block; generating metadata containing position index and compressed data length information and assembling it with the encoded image block for output. Lossless decoding includes: reading the metadata and decoding at least one image block based on the metadata. Through the above methods, parallel encoding and decoding are achieved; precise positioning via physical offsets and random access and on-demand region decoding are supported, improving the processing response of high frame rate depth streams under low computing power environments while ensuring lossless restoration.

The present application relates to the field of substation wiring, especially to a wiring system and method for skid-mounted electric control integrated substation, the method comprising: constructing time data sequences of cabinet temperature, cabinet temperature and humidity outside the substation, and comprehensive hidden features of the last time step, and passing them through a multivariate attention mechanism and a parallel LSTM encoding model to generate encoding features; passing the time data sequences through a water vapor mass conservation equation to generate physical prediction features; and passing the encoding features and the physical prediction features through a fusion mapping layer, a time attention mechanism and a control model to generate dehumidifier working power and heater working power. The present application realizes intelligent control of the dehumidifier and the heater in the anti-condensation wiring of the skid-mounted electric control integrated substation, and ensures appropriate heat dissipation and dehumidification effects.

A method by which an encoding apparatus encodes an image may comprise the steps of: identifying a plurality of segments from an image; encoding the plurality of segments in parallel; storing substreams generated through parallel encoding in a plurality of buffers; while the substreams are stored in the plurality of buffers, if the current state corresponds to a preset output condition, acquiring substream packets having the predetermined size from the plurality of buffers; and arranging the substream packets in a predetermined order and outputting a bitstream in which the substream packets are arranged.

The invention discloses an LDPC partial parallel encoder and encoding method under a DVB-S2 standard, and the encoder comprises an ROM control module which is used for selecting a matrix column weight according to an input code rate, controlling a counter to count to a corresponding ROM address, and selecting a corresponding storage unit at the same time, so as to generate a read address signal and a read enable signal of the corresponding ROM; the ROM storage module is used for reading corresponding parity check matrix information according to the input information of the ROM control module; the check calculation module is used for completing iterative calculation of an intermediate value variable according to the parity check matrix information read from the ROM storage module, calculating a check bit at the same time, and finally outputting the check bit to the de-encapsulation module; the packaging module is used for splicing the input 8-bit data into 360-bit data and outputting the 360-bit data to the ROM control module and the check calculation module; and the de-encapsulation module is used for disassembling the output data into 8-bit data so as to keep the bit width of the 8-bit data consistent with the bit width of the input data. The method has the advantages of low resource consumption and high throughput, and can be used for satellite communication.

Various implementations involve methods for encoding a media image. A video encoder encodes media images in a video stream by dividing the media images into blocks of pixels and encoding numerical coefficients associated with the blocks to generate a bitstream. As part of the encoding process, the video encoder estimates the bitrate required to encode a set of blocks of the media image. The video encoder produces more accurate bitrates for encoding the blocks by performing the bitrate estimation based on various context-based parameters of the coefficients, rather than just the coefficients themselves. This context-based bitrate estimation method can improve the visual quality of the encoded video stream.Furthermore, the video encoder encodes blocks using various simplifying assumptions to eliminate inter-block dependencies without unduly reducing visual quality. As a result, each block can be encoded independently and in parallel with the encoding of other blocks.

The invention discloses a method and a system for detecting rice bacterial leaf blight based on unmanned aerial vehicle spatial spectrum fusion. The method comprises the following steps: firstly, constructing a rice bacterial leaf blight detection data set; a rice bacterial leaf blight severity grading detection double-branch model is constructed, and the model adopts a spectrum-space double-branch parallel coding and fusion framework and comprises a spectrum branch module, a space branch module and a cross attention fusion mechanism module. And inputting the complete image of the rice field into the trained rice bacterial leaf blight severity grading detection double-branch model, and obtaining severity grading detection results of the bacterial leaf blight in different rice growth periods of the target field. According to the invention, deep complementation and dynamic information balance of spectrum and spatial features of the multispectral unmanned aerial vehicle are realized, so that the accuracy of disease detection and the generalization ability of the model are effectively improved.

This invention discloses an adaptive end-to-end network learning-driven polarization descattering depth estimation method. It proposes a physics-driven end-to-end network to simultaneously acquire clear reconstructed images and high-precision depth images in scattering environments. The method first acquires linearly polarized images of the scene in four directions, extracting light intensity and linear polarization feature maps through decoupling calculations. Then, a dual-task network including a polarization-guided fusion module is constructed. Utilizing the physical sensitivity of DOLP to scattering particle concentration and optical path length, it transforms these into a spatial attention mask, physically constraining and enhancing the shallow texture of the intensity map in the feature space. Finally, dehazing and depth estimation branches are encoded and decoded in parallel. This invention overcomes the intensity-depth ambiguity of traditional RGB images, effectively suppressing the interference of non-uniform fog on geometric perception and visual texture, and achieving high-precision scene depth estimation and high-fidelity dehazing imaging under adverse weather conditions.

The present invention discloses an ultrasonic endoscopic navigation system and method based on deep learning, which relates to the field of medical image analysis. The system includes: a parallel encoder module for extracting local features of ultrasonic endoscopic images through CNN branches and capturing global context information through Transformer branches; a channel attention fusion module for adaptively weighted fusion of the extracted features; a decoder module for generating anatomical structure segmentation masks based on fused features; a timing processing module for receiving the frame-by-frame segmentation results output by the decoder module and realizing temporal continuity analysis of ultrasonic endoscopic videos through a bidirectional LSTM network; and a multimodal fusion module for aligning and fusing ultrasonic endoscopic images. This application scheme can provide a highly reliable artificial intelligence auxiliary tool for early screening of pancreatic cancer, while reducing the learning threshold and clinical application cost of ultrasonic endoscopic technology, and promoting its popularization in primary medical institutions.

The present invention relates to a self-attention multi-scale encoding and decoding intrinsic image decomposition network, method and application, and belongs to the field of image recognition technology. The network uses a parallel encoding and decoding strategy, that is, two sets of identical but independent encoders and decoders are used to generate images of reflectance maps and illumination maps. Each network includes three parts: an encoder, a decoder and a skip connection. The encoder and decoder are both composed of multiple self-attention modules and downsampling or upsampling modules. The adaptive window self-attention proposed in this method can adaptively adjust the window according to the image content to strengthen the network's attention to different types of features and reduce the resource usage of the model; the channel enhancement self-attention module enhances the effective information in the feature map by expanding the self-attention calculation between different dimensions.

The present invention discloses a rate control method for inter-frame parallel coding based on virtual frame synchronization. A video frame to be encoded is divided into multiple virtual subframes; and then the multiple virtual subframes are combined into a virtual frame. The encoding core is scheduled based on the virtual frame, and each virtual subframe within the virtual frame is scheduled to different encoding cores for parallel encoding. According to the encoding status information of the current virtual frame, target bits are allocated to each virtual subframe within the next virtual frame to be encoded. The present invention performs an inter-frame parallel encoding process and a rate control process based on the virtual frame. The rate control process of the latter virtual frame can use the encoding status information of the adjacent previous virtual frame that has been completed encoded, so that the rate control is delay-free.

The application discloses a synchronization method of a multi-core video encoder. Each encoding core of the video encoder corresponds to a piece of encoding instance reconstructed frame state information. A piece of encoding instance reconstructed frame state information in an invalid state is configured according to information of a video frame to be encoded. As the encoding is performed, the encoding core updates an image block row number of a reconstructed image in the corresponding encoding instance reconstructed frame state information in real time. In the process of inter-frame parallel encoding, each encoding core does not need to be synchronized with other encoding cores, or the encoding core corresponding to a video frame M(N) only needs to be synchronized with the encoding core corresponding to a video frame M(N-1). When the video frame M(N) is encoded, the encoding core only needs to be synchronized with the encoding state of the video frame M(N-1) at most, so that the synchronization relationship and the synchronization operation among the encoding cores are greatly simplified.

The application discloses an FSE encoding method, device, equipment and storage medium, comprising: obtaining a to-be-encoded symbol sequence corresponding to to-be-encoded data, and determining a to-be-encoded symbol with no dependency relationship between a symbol state and other symbol states from the to-be-encoded symbol sequence to obtain a target to-be-encoded symbol; splitting the to-be-encoded symbol sequence based on a position of the target to-be-encoded symbol in the to-be-encoded symbol sequence to obtain a plurality of to-be-encoded subsequences corresponding to the to-be-encoded symbol sequence; and separately encoding the plurality of to-be-encoded subsequences to obtain the to-be-encoded data after encoding. The application splits the to-be-encoded symbol sequence by taking a symbol with unique state as a splitting point, and then obtains a plurality of subsequences, which can realize parallel encoding of the subsequences compared with serial encoding of the entire to-be-encoded symbol sequence after splitting, thereby improving the encoding rate.