27 results about "Stream network" patented technology

The application discloses a surveying-based binary structure small watershed division method and system, and particularly relates to the field of watershed division. Basic data processing of the application includes collecting and preprocessing topography, remote sensing and other data, establishing a basic database, converting vector contour line data into DEM and filling in depressions, generating a non-depression DEM, determining water flow direction by using a D8 algorithm, calculating cumulative flow, and generating a river network. The river channel is converted into a vector runoff network, the sub-watershed range is extracted and mapped, the outcrop point of underground river or karst spring is taken as a corrosion reference, the watershed outlet is determined in combination with the main river channel of a large river, and a complete correlation system is constructed. Finally, the development of the underground water system pipeline is determined through hydrogeological investigation, geophysical prospecting, drilling and pumping test, the results of comprehensive investigation, geophysical prospecting, drilling and pumping test are comprehensively considered to accurately correct the boundary of the underground water system, and the underground water system distribution map conforming to the actual conditions is formed in combination with remote sensing images and administrative division maps.

This invention discloses a dual-stream collaborative reconstruction method and system based on template frame initialization, belonging to the field of image processing technology. The method includes: recovering sparse point clouds from multi-view images and segmenting them to generate template clothing meshes and their UV unwrapping; initializing Gaussian elements and binding mesh patches using a nearest neighbor strategy, and extracting physical and Gaussian cue features to form a feature matrix; using a spectral domain low-frequency and spatial domain high-frequency dual-stream network to predict global deformation and local wrinkles respectively, and updating the clothing mesh; then jointly optimizing the human-clothing mesh through rendering and collision loss; and achieving high-fidelity clothing appearance reconstruction based on UV domain Gaussian representation and appearance refinement network, combined with differentiable rendering. This invention achieves high-fidelity and dynamically consistent 3D reconstruction of clothing images through template initialization and spectral-spatial dual-domain collaborative deformation prediction.

In an approach for efficient flood water analysis from spatio-temporal data fusion and statistics, a processor classifies regular waters by using cartographic data in a first location. A processor generates a water stream network including a watershed based on elevation data. A processor performs statistical analysis of spectral information from a multi-spectral satellite imagery over water bodies including the regular waters and flood waters. A processor correlates the spectral statistics of the multi-spectral satellite imagery to kinetic energy of the flood waters using machine learning techniques and physical modeling. A processor builds a learning model based on the correlation between the spectral statistics and the flood waters with the kinetic energy. A processor estimates kinetic energy of flood waters in a second location using the learning model. A processor evaluates a flooding risk for the second location based on the estimated flood waters kinetic energy.

The application discloses a non-intrusive load identification method and system based on environmental perception dimension reduction and double-flow confrontation. In view of the problems of limited computing power of edge and limited generalization ability of heterogeneous environment, the method collects power metering signals, intercepts high-frequency transient waveforms and calculates the robust standard deviation of quantized environmental noise; on this basis, multi-dimensional heterogeneous physical and nonlinear geometric features are extracted and reconstructed into one-dimensional scalar feature vectors; a dynamic objective function with a noise prior penalty term is constructed in combination with the noise standard deviation, and low-dimensional physical fingerprints decoupled across environments are adaptively selected; a double-flow network is further constructed, a static physical branch processes physical fingerprints to extract prior features, and a dynamic data branch inputs transient waveforms into a selective state space model, and through the field confrontation mechanism of internal hidden states, specific noise is stripped and high-dimensional time sequence shared features are extracted; and finally, the identification result is fused and output. The application effectively reduces the computing power consumption and improves the cross-domain generalization performance.

This invention discloses an anti-blurring dual-stream surface defect detection method and apparatus for industrial vision, relating to the fields of image processing and computer vision. The method acquires surface images of metal parts under high-speed motion environments; inputs a pre-constructed anti-blurring dual-stream feature enhancement network, which includes an adversarial detail recovery subnetwork and an inter-layer feature reshaping subnetwork set in parallel with a shared backbone; the backbone network includes a multi-scale guidance module, fusing residual dense connections, spatial-channel attention, and multi-scale convolution to enhance feature representation; outputs high-frequency detail images through the adversarial detail recovery subnetwork and pixel-level defect segmentation images through the inter-layer feature reshaping subnetwork; locates defect regions based on the segmented images and extracts image patches from the high-frequency images; finally, inputs the defect recognition subnetwork to output classification results; this invention improves the accuracy and robustness of defect detection in high-speed environments by collaboratively optimizing detail reconstruction and segmentation through a dual-stream network.

This invention relates to a multimodal fusion intelligent closed-loop control method and system, comprising the following steps: (1) Multimodal information acquisition and fusion recognition: The state data of the controlled object is acquired through a multimodal perception module, wherein the state data includes at least one of visual image data, auditory audio data, and auxiliary sensor data; the state data is weighted and fused, and the fusion algorithm can be implemented by calling a cloud API. The multimodal fusion intelligent closed-loop control method: System-level architecture innovation: Constructing an integrated core architecture of "multimodal perception - real-time judgment - unified scheduling - human-machine collaboration - multi-channel execution - cross-modal closed-loop correction", adopting a miniaturized integrated intelligent control terminal form, the core algorithm can be implemented by calling a cloud API, compatible with a variety of existing mainstream network connection methods, forming a hierarchical difference with existing single-point execution patents, completely avoiding conflicts, and forming an irreplaceable system-level barrier.

This invention relates to an online monitoring and dosing control method for flocculant morphology based on optical flow and machine learning. The method comprises the following steps: S1. Acquiring and preprocessing the raw video stream; S2. Extracting static morphological feature vectors and dynamic rheological feature vectors in parallel using a dual-stream network, then adaptively weighting and fusing them through a feature fusion module to obtain a fused feature vector. The dynamic rheological feature vector is obtained sequentially through dense optical flow calculation, bubble interference suppression, and dynamic feature encoding; S3. Constructing and training a prediction model, inputting the fused feature vector obtained in S2 into the trained prediction model, and outputting the current predicted flocculation degree value; S4. Constructing and training a feedforward prediction module and a feedback correction module, calculating the feedforward dosing acceleration rate and feedback correction rate of the flocculant, and generating a final control signal to adjust the flocculant dosage. This method not only provides high-precision and robust real-time assessment of the flocculation state but also achieves advanced and stable dosing control based on prediction information.

The application discloses a bird collision with a building injury identification method based on a double-flow network and multi-modal fusion, which comprises the following steps: first, collecting multi-source data such as bird autopsy physiological characteristics and environmental parameters, and constructing and processing to obtain a standardized training sample set; then, a double-flow network model containing a video stream and a historical data stream is constructed, the video stream branch extracts space-time dynamic characteristics through EfficientNet and Bi-GRU, the historical data stream branch excavates parameter correlation through TabTransformer, and the characteristics are fused through cross attention and a gating mechanism; finally, the model is refined through a multilayer perceptron, the temperature is adjusted to output a probability distribution through Softmax, the model is optimized in combination with a focus loss with a labeled smoothing, and a layered output determination result is obtained. The application improves the identification accuracy and the result reliability, and provides technical support for bird protection decisions.

The application discloses a general image fusion method based on a task customization mixed adapter network, which comprises the following steps: a general image fusion network uses a pre-trained Vision Transformer as a backbone network to extract multi-source image features in a frozen double-flow network structure, obtains global attention through window cycle movement, and inputs a subsequent task customization mixed adapter network; in a training process, the task customization mixed adapter network customizes fusion prompts according to current sample features and a task to which a current sample belongs; a fusion layer obtains locally biased fusion features according to task customization prompts and source coding of multiple source images; the locally biased fusion features are returned to each source network backbone branch, the fusion features are injected into backbone features, and the features are reconstructed into a fusion image; the value of a fusion image loss and mutual information regularization is calculated respectively, the sum of the two values is obtained, and then back propagation is performed to train the network; and various fusion images are obtained by scaling and shifting the fusion prompts.

The application discloses a ship shafting dynamic load identification method based on a physical information double-flow network, and belongs to the field of dynamic monitoring and state identification of a ship propulsion shafting. The method comprises the following steps: data acquisition and preprocessing; constructing a physical information frequency domain mask matrix based on prior physical characteristic frequencies; dynamic load identification model construction and double-flow feature extraction; orthogonal constraint; feature fusion; time sequence reasoning; load regression; joint training; and load inversion. Through the organic fusion of the physical information mask, the double-flow orthogonal decoupling network and the LSTM time sequence reasoning, the two major problems of weak anti-same-frequency interference ability and missing physical logic of the prior art are fundamentally solved.

The application discloses a method and device for configuring a drainage network, electronic equipment and a storage medium. The method comprises the following steps: configuring a unique network address segment for a drainage module and a drainage target respectively, configuring a physical address for the drainage module, configuring a data packet marking strategy and a routing strategy according to the network address segment and the physical address of the drainage module and the network address segment of the drainage target, and attracting a data packet between the drainage module and the drainage target according to the data packet marking strategy and the routing strategy. According to the application, the data packet marking strategy and the routing strategy are configured according to the network address segment and the physical address of the drainage module and the network address segment of the drainage target, the data packet is attracted between the drainage module and the drainage target according to the above strategy, the restriction that the drainage object and the drainage target must be used in pairs is eliminated, the drainage target can be shared by multiple drainage objects, and the utilization rate of the security resource is effectively improved.

This invention discloses a polarimetric SAR image classification method based on a physically driven dual-stream Mamba network, primarily addressing the problem of poor classification performance in complex scenes using existing methods. The method includes: 1) preprocessing the input polarimetric synthetic aperture radar data and adaptively compensating for polarimetric azimuth offset using a learnable polarimetric calibration layer; 2) constructing a physical-semantic dual-stream feature extraction network to simultaneously acquire scattering mechanism features and deep semantic features, and using a physically guided attention mechanism to achieve dual-stream feature fusion; 3) performing physically-aware sequential rearrangement of the fused features based on feature similarity and spatial distance, modeling the sequence features using the Mamba module, and obtaining the category prediction result for the center pixel; 4) using the trained classification model to obtain the classification result of the image under test. This invention effectively improves the accuracy and robustness of polarimetric SAR image classification in complex scenes by combining physical driving and sequence modeling.

This invention discloses a processor hardware acceleration control system based on an event data stream network, relating to the field of embedded microcontrollers. The system includes: multiple peripheral modules; an event stream network comprising multiple hardware-level event transmission channels for event triggering and data stream routing among the peripheral modules; and a pipelined DMA module connected to the event data stream network, used to automatically execute data transfer operations from a first peripheral module to a second peripheral module based on pre-stored index information after receiving an event trigger signal, and to trigger the second peripheral module to perform functional operations based on the received data via the event stream network upon completion. The index information represents the mapping relationship between the trigger event source of each peripheral module and the corresponding data transfer task. This invention solves the technical problem in the prior art where the CPU's execution of trivial peripheral tasks leads to excessive CPU resource consumption and difficulty in meeting real-time performance requirements.

The application discloses a bimodal target detection method and system, and belongs to the field of computer vision. The method comprises the following steps: acquiring a first modality image and a second modality image that are spatially registered in a scene to be detected; extracting initial features through a double-flow network, recalibrating the initial features by using a multi-scale receptive field attention sub-network to fuse multi-scale context information, and obtaining enhanced features; extracting multi-network level feature maps based on the enhanced features, decoupling and fusing features for each level, decomposing the features into shared feature components and specific feature components, and recombining the features; and finally outputting a detection result according to the fused feature map. Through early context perception and an active decoupling mechanism, the application effectively avoids inter-modality interference and improves the detection accuracy and robustness in a complex scene.

The application belongs to the technical field of deep learning, and relates to an audio and video parsing method based on noise label learning, which comprises the following steps: preprocessing original audio and video to obtain a segment-level input sequence; constructing a mutual learning noise-resistant double-flow network; training the mutual learning noise-resistant double-flow network according to a training set; comparing the validation set indicators of two sub-networks in the trained mutual learning noise-resistant double-flow network, and taking the sub-network with the larger validation set indicator as an audio and video parsing model; and parsing through the audio and video parsing model according to a test set to obtain a video prediction result. The mutual learning noise-resistant double-flow network is composed of two sub-networks with the same structure but different initializations, a cross filtering mechanism is executed according to the clean masks generated by the two sub-networks during the training of the mutual learning noise-resistant double-flow network, and the dynamic confidence ratio is gradually reduced through a cosine strategy, so that the problems of high pseudo-label noise rate and easy overfitting noise in the existing audio and video parsing task are solved.

The application relates to the technical field of Internet of Things communication, in particular to a device instruction issuing control method and system. The method comprises the following steps: performing confidence calibration on the original severity of a to-be-issued instruction through a temperature scaling algorithm, dividing the to-be-issued instruction into different priority levels and binding corresponding SLA indexes; performing real-time parameter telemetry of a multi-physical link and block confirmation protocol negotiation; combining an analytic hierarchy process and a self-adaptive gated double-flow network, fusing measured telemetry data and timing prediction data, dynamically calculating the comprehensive quality score of each candidate channel and adding a feature label; performing frame cutting based on a maximum transmission unit of a link, and performing forward deduction and scheduling distribution by using a multi-level reflection scheduling algorithm; and triggering multi-stage cascading disaster recovery switching and downstream non-core service fuse degradation control when the network is limited. The application can effectively guarantee low-latency and high-reliability transmission of high-priority core control instructions, and ensure safe and stable operation of the equipment under an extreme network environment.

The application relates to the technical field of logistics network planning, and discloses an underground logistics network planning method and system based on artificial intelligence. The method comprises the following steps: each node predicts a local future state through a space-time graph convolutional neural network; an offline trained multi-agent reinforcement learning model is used to generate an optimal local action sequence; and space-time conflicts are eliminated through a communication-free deterministic protocol. The system comprises a distributed state sensing and prediction module, a multi-agent reinforcement learning decision module, a space-time conflict elimination module and a hierarchical task allocation module. The application realizes decentralized and high-robustness autonomous collaborative planning, and significantly improves the operation efficiency and safety of the underground logistics network in a weak communication environment.

Systems and methods for identifying correlations for adaptive noise reduction. The system obtains a set of network performance metrics measured during a period of time from a sensor module communicatively coupled to a set of sensors deployed at remote devices. The system may input the set of network performance metrics into a trained model to obtain an optimal set of network performance metrics to be measured during a next period of time, wherein the model is configured to determine a next set of metrics to be measured based on a previous set of metrics. The system identifies a second set of sensors for usage and may generate one or more commands configured to effectuate activation of one or more sensors of the second set of sensors, and deactivation of one or more sensors of the first set of sensors.

The application discloses an immersive operation and maintenance tool calling method based on hand and whole body action fusion recognition, and belongs to the fields of human-computer interaction, virtual reality (VR) and pattern recognition. The method comprises six steps: (1) a mixed collection environment is constructed, multi-participant operation data is collected based on "intention driving", and time stamp alignment and action classification are performed; (2) a self-centered coordinate system is established, whole body data is mirror corrected and normalized, and is uniformly mapped to a first person perspective; (3) a differential sliding window is adopted to construct an offline data set and an online ring buffer, data enhancement and real-time action segmentation are realized; (4) four types of feature streams of joints, bones, speed are decoupled, input into an asymmetric double flow network, and hand fine features and body noise resistant features are respectively extracted; (5) a double granularity gated residual fusion mechanism is adopted to dynamically calculate complementary weights, four flow feature classifications are combined, and tool calling intentions are output; and (6) based on a PC-VR architecture, a sliding window voting and cooling strategy are used to confirm the intention, a tool model is generated and is adsorbed to the hand. Compared with traditional menus and static gestures, the method significantly reduces operation interruption and cognitive load, captures continuous dynamic semantics, fuses heterogeneous skeleton data, and improves the immersion, efficiency and robustness of maintenance training.

This application discloses an inventory allocation method and related apparatus, relating to the field of information processing. It involves acquiring attribute data from multiple inventory batches and predicted data from multiple future time periods. The attribute data includes at least available inventory and cost information, while the predicted data includes at least predicted demand and sales price information. A directed flow network is constructed, comprising source nodes, batch nodes, time nodes, and sink nodes. The capacity of the directed flow path is the predicted demand, and the cost is the negative of the unit expected revenue. The network is solved using a minimum-cost maximum flow algorithm to determine the inventory batch combinations corresponding to each time node. A sales order is determined based on a revenue algorithm, and sales recommendations are executed. This invention transforms the multi-source, inter-period inventory allocation problem into a globally optimized network flow model, solving the problem of local optima in existing technologies. This improves inventory turnover efficiency while maximizing total revenue.

The application discloses a bridge support visual detection system and method based on artificial intelligence, and relates to the technical field of intelligent detection.The system comprises the following steps: collecting bridge support data and performing pretreatment on the bridge support data to obtain an image sequence and a load event time axis; the bridge support data comprises real-time video stream under the action of dynamic vehicle load, static images, real-time displacement distance of the surface of the bridge support, and temperature and humidity of the bridge support; a space-time double-flow network is constructed based on the image sequence and the load event time axis, time feature maps and space feature maps are extracted and fused; the fused features are subjected to time sequence enhancement, decoding and up-sampling to generate a dynamic mask matrix.The application improves the precision of deformation separation, realizes seamless connection from damage identification to mechanical property evaluation, can intuitively show the influence of damage on the bearing capacity and stability of the bridge support, and improves the scientificity and effectiveness of bridge management decision.

This invention proposes an intelligent segmentation method for reticular fiber stained pathological images and its application. Addressing the limitations of existing technologies in identifying reticular fiber structure disruptions to define tumor boundaries and artifacts caused by sliding windows, this invention employs adaptive sliding window segmentation and multi-stage pre-filtering to extract effective image blocks. The input is a dual-stream network, where structural and visual flows are fused through cross-attention based on tissue physical size mapping receptive fields to achieve collaborative verification of visual abnormalities and structural disruptions. Model training utilizes a dynamic boundary-aware loss term with physical scale constraints for joint optimization. Finally, a spatial weight fusion mechanism combined with a graph model incorporating physical priors eliminates breakage artifacts and smooths global boundaries. This invention is primarily used for high-fidelity lesion segmentation of pituitary neuroendocrine tumors.

This invention relates to the field of materials testing technology, specifically disclosing a method and system for detecting materials features by fusing spatiotemporal visual features and process parameters. The method includes acquiring video stream data of the polymer to be tested under controlled operating conditions, simultaneously acquiring process parameters; performing spatiotemporal processing on the video stream data based on a preset dual-stream network, outputting spatial feature vectors and temporal feature vectors; obtaining a material flowability index and a material viscosity characterization index based on the spatial and temporal feature vectors, and determining the state identification result of the polymer. This invention combines fluid mechanics principles with a computer vision deep learning model to achieve non-contact, real-time, and automated identification and monitoring of material viscosity and flowability. The original input is a high frame rate video stream; the spatial appearance information and temporal motion information are processed separately through a dual-stream network, and the identification results are fused to obtain material features. The method boasts high identification efficiency and high accuracy.

This invention discloses a method for small-disturbance stability analysis of grid-connected and off-grid data centers, belonging to the field of stability analysis technology. The method includes the following steps: S1, constructing the data center power supply system; S2, constructing the dynamic admittance matrix of the DC network and the transfer function of the converter; S3, determining the reconstructed admittance matrix and the reconstructed transfer function; S4, determining the dominant oscillation mode based on the reconstructed transfer function. This invention equates the dynamic differences between different converters to the series and parallel impedance forms of the connected network. Therefore, all converters have the same transfer function, and the system retains its original dynamic characteristics. Through the principle of similarity transformation, the full-order state-space model is equivalent to multiple decoupled low-order system equations, thereby achieving efficient and accurate calculation of small-disturbance stability analysis for grid-connected and off-grid data centers.

The application discloses a gait emotion recognition method and system based on a double-flow network, and comprises the following steps: performing feature extraction on an obtained walking video to obtain three-dimensional gait data of skeleton points; inputting the three-dimensional gait data into a pre-trained double-flow network model, wherein the double-flow network model comprises a global capturing module, a local capturing module and a feature fusion module; extracting global space-time features of the three-dimensional gait data by using the global capturing module; extracting local space-time features of the three-dimensional gait data by using the local capturing module; and fusing the global space-time features and the local space-time features by using the feature fusion module to output a predicted emotion recognition type; the gait emotion recognition method and system combine the global space-time features and the local space-time features, and improve the emotion recognition accuracy for gait data.

The application belongs to the field of resource allocation, and discloses a planning method and device for an offshore integrated energy island, comprising the following steps: collecting multi-source heterogeneous operation data and marine environment data of the offshore integrated energy island, and calculating the credible weight of each data source by constructing a cross-source consistency entropy model; establishing a physical mechanism model, and performing parameter inversion based on a weighted least deviation criterion to determine a device parameter set reflecting the actual operation environment; constructing a multi-modal energy flow network covering electric energy, chemical energy and thermal energy based on the device parameter set, introducing wind speed and significant wave height as exogenous disturbances into the multi-modal energy flow network, and establishing an uncertain performance flow propagation model; taking the capacity of wind power, energy storage, electrolytic hydrogen production, ammonia production and a sending channel as decision variables, constructing a comprehensive objective function considering construction cost, operation cost and risk loss, and solving to obtain an optimal capacity configuration scheme, so as to plan the offshore integrated energy island according to the optimal capacity configuration scheme.

The application discloses a kind of spatio-temporal double-flow network model processing methods for dim weak space target detection, belong to visual target detection technology and astronomical information technical field.The method uses static-dynamic double-flow network architecture, and continuous optical image sequence is handled jointly;Single frame image is processed using static flow network, and morphological features of fine granularity are extracted by tree-like hierarchical feature aggregation and deformable convolution;Parallelly, the whole sequence is processed using dynamic flow network, and motion features are extracted based on 3D Swin Transformer using spatio-temporal window attention mechanism;Morphological features and motion features are weighted and adaptively fused by spatio-temporal parallel attention module;The features after fusion are input into the detection head based on key points, and the center heat map, boundary box size and center offset of spatial target are output, to complete end-to-end target detection.The application effectively improves the joint representation ability of model to dim target morphological structure and time sequence motion pattern, significantly improves the detection accuracy and robustness in low signal-to-noise ratio and complex dynamic background.Under the condition of about 18.58M parameter quantity of model size and about 96.00 GFLOPs of calculation amount, single frame inference speed can reach about 32.41 milliseconds, which balances high detection performance and low calculation overhead, and is suitable for real-time or near real-time processing requirements in actual space observation task.