25results about How to "Reduce reconstruction error" patented technology

The invention relates to the technical field of artificial intelligence, in particular to an artificial intelligence-based enterprise digital management maturity evaluation method, which comprises the following steps of: performing sample generation by adopting a generative adversarial network algorithm based on a momentum optimization variant to realize data expansion; inputting the expanded enterprise management data into a feature extraction model to train the feature extraction model; inputting the enterprise management data after feature extraction into a feature dimension reduction model to train the feature dimension reduction model; inputting the dimension-reduced enterprise management data into a classifier model to train a classifier; and processing new sample data by using the trained model so as to evaluate the digital management maturity of the new sample data.

The invention relates to the field of computer vision, in particular to a decayed tooth intelligent identification method and system based on computer vision, and the method comprises the following steps: image collection, image segmentation, bottleneck feature extraction, reconstruction error optimization, diagnosis result generation, diagnosis result feedback, and decayed tooth development prediction. According to the method, a decoder part of U-Net and an encoder part of ConvNeXt are combined into a segmentation model, and a multi-label recall loss function is used for effective identification; a capsule graph network is introduced to serve as a bottleneck layer, and potential spatial relations and high-level dependence of local features in data are captured through a dynamic routing mechanism; and an MFO-SFR-HWACO algorithm is used to optimize the hyper-parameter configuration of the depth auto-encoder. The system comprises a data acquisition module, an image segmentation module, a bottleneck feature extraction module, a reconstruction error optimization module, a diagnosis result generation module, a decayed tooth development prediction module and an intelligent report generation module.

The present invention discloses an open set recognition method based on representation learning and diffusion model, which mainly consists of two stages: model training and model detection: the model training stage mainly learns the representation learning network and the diffusion network, wherein the representation learning network is trained using supervised contrast loss and cross entropy loss, and then the compact representation learned by the feature extraction module is input into the diffusion network for training; in the model detection stage, the representation learning network is used to obtain feature representation and pseudo-label, and then the diffusion network is used to perform biased reconstruction on the test set data, and finally the known class and the unknown class are identified according to the reconstruction error. The proposed method uses representation learning and diffusion model for open set recognition. Compared with the existing open set recognition method based on reconstruction, the present invention essentially enhances the bias of the model to the known class from multiple levels, so that the model can simultaneously obtain a larger reconstruction error of the unknown class and a better reconstruction error of the known class, thereby obtaining a better open set recognition performance.

The present invention is a subspace-based real-time reconstruction method for multi-parameter data of the Internet of Things. The method first utilizes the spatiotemporal correlation of IoT data and only collects part of the data for transmission, that is, only part of the sampled data is transmitted to the base station in each time slot, so as to reduce the energy consumption required for sensor nodes to collect and transmit data. Secondly, the multi-parameter data collected at continuous moments are organized into tensor forms, making full use of the spatiotemporal correlation between multi-parameter data and the inherent relationship between various types of data; to achieve real-time data reconstruction, the sliding window model is used to introduce historical data; based on Tucker decomposition, the spatial distribution subspace of the current time slot data is obtained from the historical data, and the spatiotemporal correlation and internal connection between the data are fully explored; the current time slot data is reconstructed in real time using the partial sampled data received in the current time slot and the constrained total variation in the spatiotemporal direction.

A video encoding and decoding method, apparatus, and medium are provided. An example method of video decoding includes receiving a video bitstream including a plurality of frames, the plurality of frames including a current frame; and identifying a primary reference frame for context initialization of the current frame. The method further includes identifying a reference frame of the current frame based on the primary reference frame, and reconstructing the current frame by applying a cross component sample offset (CCSO) mode using the reference frame.

The invention discloses a micro-service call chain anomaly detection method based on a Nested GNN and a VAE, and relates to the technical field of data processing. The method mainly comprises the steps of dividing an inner graph and an outer graph based on a call chain in a call chain data set; assigning an initial vector representation to nodes of each inner graph and each outer graph; a call chain reconstruction model based on the nested graph neural network and the variational auto-encoder is constructed, the call chain reconstruction model comprises an input layer, a graph structure layer, a sampling layer and a reconstruction layer, the graph structure layer comprises the nested graph neural network used as an encoder, and the reconstruction layer comprises a multi-layer perceptron used as a decoder; training the call chain reconstruction model by using the normal call chain data; to-be-detected call chain data are input into the call chain reconstruction model, if the reconstruction error exceeds a preset threshold range, the call chain is judged to be abnormal, and otherwise, the call chain is judged to be normal. According to the method, finer-grained service internal exceptions can be detected, and meanwhile, the accuracy of call chain exception detection can be improved.

The application discloses a kind of space neighborhood reconstruction self-encoding multimodal fusion vector generation method, belong to geographic information science, computer vision and artificial intelligence cross technical field.The present mask self-encoding technique is only stopped at the technical limitation of grid level spatial unit, cannot process the heterogeneous modal vector of geographical entity with independent identity, the application proposes a kind of geographical entity-oriented multimodal mask self-encoding training framework: the multiple heterogeneous modal vectors of geographical entity with independent entity identification are randomly masked, the heterogeneous modal vector information of adjacent entity in the space neighborhood determined by k-ring algorithm is used, the masked modal is reconstructed by multimodal fusion network using the remaining modal vector of the entity not masked;With the weighted combination of reconstruction loss and contrast learning loss, the fusion network parameters are jointly optimized.The application also provides a joint optimization method and a cold start and gradual switching method.

The invention discloses a remote sensing image processing method based on deep learning, and relates to the technical field of remote sensing image processing, and the method comprises the steps: recognizing all buildings containing building features from a remote sensing image, marking the buildings as complete buildings and sheltered buildings, and enabling the sheltered buildings to correspond to the complete buildings based on similarity matching, the complete building is adopted as a template to assist in reconstructing the contour of the sheltered building; and through an accurate contour matching algorithm, the optimal corresponding relation between the sheltered building and the complete building can be ensured, so that the accuracy of contour reconstruction is improved. The contour of a complete building is used as a template, and reconstruction errors caused by contour information loss due to shielding can be reduced. The contour matching and splicing method is directly used, so that the calculation is simpler, the processing speed can be accelerated, and the overall efficiency can be improved. The problem of building shielding in the remote sensing image can be effectively solved, and the application range and practicability of remote sensing image analysis are improved.

This invention relates to the field of image analysis and discloses a method for analyzing railway freight car images. The method includes: acquiring an image of the freight car to be analyzed; reconstructing the freight car image using a trained reconstruction neural network to obtain a reconstructed image; calculating the reconstruction error between the reconstructed image and the freight car image; determining whether there are foreign objects in the freight car image based on the reconstruction error; identifying the foreign object region in the freight car image when foreign objects are present; querying the components to be detected in the freight car and locating the component region in the freight car image; performing fault detection on the component region to determine the fault type and key region of the faulty component; calculating the anomaly score of the key region; determining the abnormal region of the faulty component based on the anomaly score; and fusing information from the foreign object region, fault type, and abnormal region to obtain the analysis result of the freight car image. This invention can improve the real-time performance of railway freight car image analysis.

This invention discloses a generative recommendation system and method based on quantized vector retrieval and LLM, belonging to the technical field of computer information processing and artificial intelligence recommendation systems. The method includes: Step S1, extracting continuous collaborative features of users and items; Step S2, constructing and pre-training an AHPQ segmenter quantization module, mapping continuous collaborative features to discrete ID token sequences and optimizing the quantized codebook; Step S3, using the pre-trained codebook vector to collaboratively initialize the LLM's token embedding layer; Step S4, fine-tuning the LLM using a freeze-adapt strategy, and then generating user preference vectors through a collaborative semantic fusion module; Step S5, calculating the similarity between the user preference vector and the item database vector, and retrieving Top-K items based on cosine similarity as the recommendation result. This invention solves the problems of ID discretization difficulties and collaborative signal loss in LLM recommendation, preserving the high-order collaborative structure while leveraging the semantic reasoning capabilities of LLM, and simultaneously achieving efficient reasoning and cold-start generalization.

The application discloses a kind of based on LoRa's sea cable state remote wireless telemetry system and method, belong to cable laying state monitoring technical field, system includes being laid on several state monitoring nodes of sea cable, relay node and shipborne monitoring center.Adaptive different compensation strategies are taken according to the proportion of missing packet loss monitored by analysis computer: when mild packet loss, trend filling is executed using effective node change rate and spatial correlation;When moderate packet loss, confidence weight that attenuates dynamically with interruption time is introduced in reconstruction model;When severe packet loss, spatial physical correlation constraint matrix feature sub-block is extracted to perform order reduction calculation and combine spatial interpolation method to recover data.The application solves the problem of data loss caused by unstable wireless communication in extreme sea conditions through hierarchical compensation mechanism, ensuring that the full-line state distribution of sea cable can be reconstructed stably and accurately and the warning can be output under different packet loss conditions, significantly improving the robustness and construction safety of the sea cable monitoring system.

The application discloses a wind power daily scene generation method based on a tensor self-organizing mapping neural network, and comprises the following steps: obtaining a daily scene data set of multiple wind farms; clustering the daily scene data set by using a tensor self-organizing mapping neural network; constructing a variational autoencoder for each cluster respectively to extract hidden features from the daily scene data; performing random simulation and sampling on the daily scene data of each cluster by using the hidden features to obtain a daily scene hidden variable data set; decoding and reconstructing the daily scene hidden variable data set to obtain reconstructed daily scene data of each cluster; and aggregating the reconstructed daily scene data of each cluster to obtain a reconstructed daily scene data set. After clustering the daily scene samples, the application respectively reduces the dimension and reconstructs the samples of each cluster, improves the accuracy of the reconstructed daily scene data, significantly reduces the spatial and temporal correlation error of the reconstruction by considering the time and space correlation of the scene data, and strengthens the feature expression ability of the generated scene.

The invention discloses an ultrahigh time resolution flow field reconstruction method based on time sequence-space joint coding ViT, belongs to the field of fluid mechanics flow field prediction and reconstruction, and aims to solve the problems of insufficient time sequence feature modeling, limited key region precision and poor dynamic consistency of traditional flow field reconstruction. The method comprises the following steps: firstly, collecting multi-time-period sparse pressure field data, carrying out time sequence construction, and carrying out key region mask weighting preprocessing; constructing a ViT model fusing time sequence-space two-dimensional position coding, and cooperatively extracting time sequence dependence features and space local features of the pressure field; secondly, designing a multi-component loss function containing key region weighted MSE, flow field continuity constraint and structural similarity, and ensuring numerical value consistency, dynamic rationality and visual structural integrity of a reconstructed flow field; and finally, improving the high-frequency flow field feature capture capability through output layer optimization combining hierarchical attention aggregation and convolutional projection. According to the method, complete ultrahigh time resolution flow field data can be reconstructed, and the method is suitable for high-precision flow field prediction and engineering fluid mechanics analysis scenes.

The present invention discloses a laser absorption spectrum signal denoising method based on improved local mean decomposition, comprising: obtaining a standard amplitude and a standard second harmonic signal, and obtaining the amplitude and the second harmonic signal of the laser absorption spectrum signal of the trace gas to be measured x ( k ) ; respectively for the second harmonic signal x ( k ) add two groups of white noise with equal size and opposite sign, perform local mean decomposition on the two groups of signals after adding white noise, and calculate the second harmonic signal according to the corresponding main factor component set x ( k ) corresponding to the main factor component set, from which useful main factor components are selected for reconstruction to obtain the second harmonic signal after noise reduction y ( k ), calculate the second harmonic signal y ( k ), and then correct the gas concentration corresponding to the trace gas to be measured according to the standard amplitude and the amplitude of the trace gas to be measured to obtain the corrected gas concentration of the trace gas to be measured. The present invention can improve the detection accuracy and the lower limit of measurement, and has a small amount of calculation, which is convenient for engineering application.

The invention relates to a fuel cell stack life prediction method and device and electronic equipment, and belongs to the technical field of fuel cells, and the method comprises the steps: obtaining to-be-measured life decline data of a fuel cell stack; carrying out self-adaptive noise complete set empirical mode decomposition on the to-be-measured life decline data to obtain to-be-measured intrinsic mode component data; based on a preset target autoregressive integral moving average model, predicting the intrinsic mode component data to be measured to obtain predicted intrinsic mode component data; and fusing the predicted intrinsic mode component data to obtain predicted life decline data. According to the invention, the intrinsic mode component of the to-be-measured life decline data is predicted through the data-based target autoregressive integral moving average model, and accurate prediction of the life of the fuel cell stack is realized.

The present invention belongs to the technical field of gas turbine turbines, and specifically designs a variable input turbine surface physical field reconstruction method based on a pyramid neural network. A data set is obtained, data in the data set is extracted and preprocessed, and the data set is divided into training data and verification data; a pyramid network is constructed based on a convolutional network; a distributed training strategy is used to train the constructed pyramid network using the training data; the performance of the trained pyramid network is verified using the verification data; the normalized experimental data is input into the verified pyramid network, and the complete blade surface physical field can be reconstructed. The present invention uses a supervised learning method to estimate the complete blade surface pressure and temperature field from sparse data obtained from a limited number of fixed measurement points distributed at a given number of blade height sections, which can improve the information density of blade grid experiments and reduce the cost of experiments and numerical simulations.

The invention provides a computer tomography camera layout optimization method with a limited projection angle, and belongs to the technical field of chemiluminescence computer tomography. Comprising the following steps: S1, system modeling: establishing a mathematical relationship between a target flow field and a projection thereof; s2, reconstructing a flow field through an inversion algorithm; performing chromatography reconstruction by using a classic algebraic reconstruction technology ART algorithm; s3, a computer tomography camera layout optimization method of a limited projection angle; constructing a target function as an effective predictor of a reconstruction error; the maximization problem is globally optimized and solved by using a simulated annealing SA algorithm; and realizing an optimal solution by adopting a simulated annealing algorithm. By optimizing the pitch angle and the azimuth angle of the camera, capturing of effective projection information is remarkably increased. And camera layout optimization is carried out by adopting a simulated annealing algorithm, so that the system can flexibly adapt to a dynamic environment. According to the method, the morbidity of a weight matrix is effectively reduced, the calculation complexity is reduced, and the reliability of a measurement result is enhanced.

The invention discloses an industrial non-intrusive measurement data transmission method and system based on wavelet transform compression. The method comprises the following steps: designing a data transmission protocol for I format information; the power system collects the equipment type, the operation mode, the real-time power, the real-time current, the transient power and the transient current on the production line, and uniquely encodes the equipment type and the operation mode; the power system monitors the start-stop state of electric equipment on the production line; transmitting a remote measurement value and a remote signaling value to a monitoring system; directly transmitting real-time power and real-time current through a network; the transient power and the transient current are compressed and then transmitted; the start-stop state of the electric equipment is directly transmitted through a network; data encryption is carried out before transmission; and the monitoring system carries out data decryption, verification and error processing on the received data. According to the invention, effective compression transmission of transient power and transient current data is realized, and high requirements of a modern industrial automation system on real-time performance and accuracy are met.

The present invention belongs to the technical field of centrifugal pump fault detection, and discloses a method and system for modal reduction of cavitation flow field of a centrifugal pump based on dynamic modal decomposition. The method performs three-dimensional modeling on the centrifugal pump, and uses the impeller, volute, and inlet and outlet sections of the centrifugal pump to form a flow calculation domain; performs geometric modeling in Solidworks according to the size and performs geometric processing on the flow calculation domain modeling, and then imports the geometric model constructed in Solidworks into the ICEM software to perform mesh division of the flow calculation domain, and sequentially generates meshes of various components of the centrifugal pump; imports the divided meshes into the commercial software fluent for numerical simulation to obtain simulation data; changes the inlet pressure in fluent to obtain different cavitation states. The present invention has certain significance for quickly identifying cavitation in a centrifugal pump and preventing cavitation hazards caused by vaporization, thereby preventing centrifugal pumps from malfunctioning.

This invention discloses a virtual skeleton perspective method, device, and storage medium, belonging to the field of medical image-assisted and computer vision-based 3D human body reconstruction technology. Addressing the shortcomings of existing medical imaging methods, such as the lack of visualization of the internal skeleton during positioning, poor real-time performance of iterative SKEL parameter solving, insufficient ability to measure real-world scales, and low reconstruction accuracy in occluded scenes, this invention replaces iterative optimization with a learned SKEL parameter estimation model, rapidly obtaining SKEL parameters through a single forward computation. The preferred scheme uses depth as a geometric prior to modulate attention, and adds bone length prediction to achieve metric scale recovery without internal parameters. Simultaneously, it realizes 3D skeleton virtual perspective overlay, positioning guidance, and real-world spatial distance measurement. This invention significantly improves the accuracy and efficiency of image positioning, reduces the radiation burden on patients, and balances reconstruction robustness and real-time performance.

The application discloses a kind of sparse sampling conditions based on optimization kriging full-band electromagnetic environment map construction method, belong to modern wireless communication network and cognitive electronic countermeasure field.First, the electromagnetic power received by each sparse sampling point in target area is obtained and converted into received signal strength, while the radiation source coordinates are obtained, the deterministic trend component of each sampling point is extracted using the logarithmic distance path loss model, and the residual component is further obtained;Then, combined with particle swarm optimization algorithm and leave-one-out cross-validation, the optimal core parameters of the semi-variogram function are obtained, and further combined with the residual component of the sampling point, the residual space interpolation of the target area grid is carried out;Finally, the interpolation result is superimposed with the deterministic trend component of the target area grid, and the full-band three-dimensional electromagnetic environment map of the target area is obtained.The application effectively breaks through the bottleneck of traditional interpolation algorithm ignoring the radio propagation law, and significantly improves the damage reconstruction robustness under the condition of extreme sparse sampling.

The invention provides an image super-resolution method based on a novel network convolution architecture, and aims to improve the spatial resolution and detail recovery capability of an image. According to the method, a super-resolution generative adversarial network is taken as a baseline model, and the network is improved and optimized by using a novel network convolution architecture. In the improved network, batch normalization in the original network is replaced by a new layer normalization module, so that the network can perform independent normalization on a single sample, the dependence on statistical information of the whole batch is avoided, and the feature extraction capability is enhanced. In addition, a new module is introduced in the method, the structures of the generator and the discriminator are further optimized, and the performance of the network during high-resolution image generation is improved. According to the method, the paired image data sets are adopted for training, the model obtained after improved network training can effectively reconstruct the high-quality image, the definition and detail performance of the image are remarkably improved, and good adaptability and wide application prospects are shown.

The invention belongs to the field of remote sensing image processing, and relates to a multimodal remote sensing image spectrum unmixing method based on combination of a digital surface model (DSM) and a hyperspectrum. The invention aims to solve the problems that the existing method does not consider that DSM and hyperspectral data are actually different-dimension information from different sensors in the same scene and does not model joint imaging of the DSM and the hyperspectral data in a model imaging process, so that the spectral unmixing precision of a remote sensing image is poor. The method comprises the following steps: acquiring multi-modal remote sensing images with the same coverage geographic area and the same resolution; the multi-modal remote sensing image comprises a hyperspectral remote sensing image and DSM remote sensing image data; based on DSM remote sensing image data, geometric component information of ground feature types is obtained; constructing a DSM and hyperspectrum combination-based multi-modal remote sensing image spectrum unmixing model; and solving the DSM and hyperspectrum combination-based multi-modal remote sensing image spectrum unmixing model by adopting an alternating direction multiplier method to obtain abundance information.

The application provides a differential absorption laser radar signal denoising method and a computer readable medium. The application performs multi-signal space-time average processing on a differential absorption laser radar measured signal sequence to obtain a space-time average laser radar signal sequence; acquires a to-be-decomposed radar signal sequence, obtains a plurality of to-be-decomposed radar signal forward and negative noise sequences through the to-be-decomposed radar signal sequence and a plurality of white noise sequences, and obtains a plurality of layer intrinsic mode signal sequences and a residual signal sequence through empirical mode decomposition calculation; divides the plurality of layer intrinsic mode signal sequences into each layer high-frequency and low-frequency intrinsic mode signal sequences, and respectively obtains reconstructed high-frequency and low-frequency intrinsic mode signal sequences through filtering processing; and obtains a denoised differential absorption laser radar signal sequence through residual signal sequence reconstruction calculation. The application can realize global adaptive filtering of a signal, and ensures that the signal has a high signal-to-noise ratio, a high space-time resolution, a low reconstruction error and a low distortion degree.

The application discloses a kind of sensor signal active fault-tolerant control system for strong electromagnetic pulse environment, belong to digital control fault-tolerant technical field.System includes the pre-processing module, signal identification module, pseudo-cluster separation module and signal reconstruction module connected in turn.Pretreatment module carries out FIR low-pass filtering to original sensor signal;Signal identification module generates abnormal flag by sliding window difference and energy mutation monitoring;Pseudo-cluster separation module carries out pseudo-cluster to continuous abnormal flag by finite state machine, and outputs interference section mask;Signal reconstruction module reconstructs signal to interference section based on mask, using local autoregressive prediction model.The scheme can identify and reconstruct sensor signal distorted due to strong electromagnetic pulse interference in real time and efficiently, significantly improve the reliability and robustness of digital control system such as aero-engine in complex electromagnetic environment, and is particularly suitable for FPGA platform implementation.