18results about How to "Increase dependence" patented technology

The invention belongs to the technical field of remote sensing image processing, and particularly relates to an open vocabulary and remote sensing image interaction-oriented multi-modal semantic segmentation method. According to the method, a double-branch text encoder is introduced, and scene-level semantics are extracted from various text data to realize dynamic cross-modal fusion. The text derivative features dynamically interact with visual embedding through a proposed text guide visual semantic fusion module, and domain perception feature refinement and human interpretable decisions are achieved. Besides, text integration opens up a way for opening word semantic segmentation, so that the system can recognize and classify invisible categories through natural language description, and strict constraints of a predefined category classification method are overcome.

This invention relates to the field of multimodal sentiment analysis technology, and in particular to a method and system for multimodal sentiment analysis based on sequential translation under uncertain and missing modalities. The method includes: acquiring multimodal sentiment data, including visual, textual, and auditory modalities; constructing a multimodal sentiment analysis model based on sequential translation, firstly by extracting self-attention encoding features of each modality using a Transformer encoder; and then designing a bidirectional sequential translation architecture centered on the textual modality. This invention improves the utilization of textual modal features in the feature fusion process, thereby enhancing the quality of multimodal fusion features and providing strong support for subsequent complex sentiment analysis.

The invention relates to the technical field of neural network acceleration, and relates to a CPU-FPGA-oriented graph neural network training acceleration method and system. The method comprises the steps that a CPU stores complete graph feature data and distributes different feature data to storage modules of different FPGAs through data channels according to a load balancing algorithm; partitioning the GNN network model by the CPU, and matching a partitioned model operator with a corresponding FPGA (Field Programmable Gate Array); the CPU completes issuing and sampling of a partition model according to a sampler algorithm, and sends the partition model to the FPGA; reading feature data after the FPGA is started, completing a calculation task by a load execution module, and writing a result back to a CPU memory; the CPU completes data updating and weight updating; and resending the updated weight data to the corresponding FPGA, and carrying out iterative training. According to the invention, operation of the graph neural network on a CPU-FPGA platform is successfully realized, and the overall acceleration effect is excellent.

The invention discloses a Beidou base station data transmission and detection system, which relates to the technical field of wireless communication networks, and is characterized in that a link stability score, a link prediction state and a self-adaptive weight mechanism driven by a network environment disturbance level are introduced into a Beidou base station multi-link data return scene; path selection does not only depend on transmission parameters at a single moment, but comprehensively considers fluctuation characteristics and future trends of the parameters in a time dimension, so that link availability misjudgment caused by instantaneous interference in a complex electromagnetic environment is remarkably reduced; according to the method, firstly, a normalized stability score is constructed by using statistics such as a trend reference, deviation amplitude and deviation frequency, so that the system can identify which links are more stable in a long-time window, and blind switching caused by short-term false improvement is avoided.

The application discloses an ultrablack coating with wide-angle absorption and a preparation method and application thereof, and belongs to the technical field of ultrablack coatings and preparation thereof. The application solves the problem that the existing ultrablack coating cannot present extremely low scattering and high absorption under large-angle incident light. In the application, carbon black is used as an absorbent, a high-temperature-resistant silica sol solution is used as a binder, and a one-step sintering method is adopted to form a stable structure and construct an ultrablack coating with a forest-like light trap array structure. The forest-like array structure on the surface of the ultrablack coating significantly changes the refractive index of the coating surface, thereby reducing the surface optical reflection; the surface forest-like light trap structure is closely arranged, can make the light incident from each angle to reflect back and forth in the structure, and then maximally absorbs the incident light, realizes stable high light absorption under wide angles, makes the reflectivity of the ultrablack coating not obviously change with the angle of incident light, and improves the deep space exploration precision of a space optical system.

The invention provides a hybrid deep learning automobile sales prediction method fused with multi-source features and application, and belongs to the technical field of hybrid deep learning fused with multi-source features. The method comprises the following steps: 1, inputting subjective evaluation data into a semi-supervised feature generation layer, and extracting features in the subjective evaluation data through a method based on semi-supervised learning feature clustering; 2, inputting static non-time-series data and the subjective evaluation type data and the time series data in the step 1 into a parallel feature extraction layer, and performing feature extraction; 3, performing deep fusion on the output in the step 2, and inputting the merged feature vector into a fusion optimization layer; step 4; and using an Adam optimization algorithm to train the models formed in the steps 1-3. According to the method, combined feature extraction and fusion of non-time-sequence feature mixed data of time sequence features, static attributes and subjective evaluation classes are realized, so that the automobile sales volume prediction precision is optimized.

The invention provides a dual-optical-path fire smoke detection smoke chamber optimization system and method, and relates to the technical field of fire detection. The method specifically comprises the following steps: synchronously acquiring light intensity signals of a main light path and an auxiliary light path, calculating an instantaneous ratio and a change rate of the main light path, and inputting into a weight analyzer to obtain a weight coefficient so as to generate a comprehensive light intensity signal; and extracting time sequence features based on the second time window, and outputting a fire probability and a confidence coefficient in combination with a pre-trained fire analysis model so as to optimize an alarm threshold value of the dual-optical-path fire smoke detection system. According to the invention, the fusion weight of the dual-optical-path signal can be dynamically adjusted, a virtual adjustable scattering angle is designed at the algorithm level, and the fire alarm fixed threshold is dynamically adjusted by comprehensively analyzing the fire probability and the detection confidence coefficient, so that the detection sensitivity of the system to different smog and the anti-interference capability in a complex environment are improved, and the system has high practicability. And meanwhile, the false alarm probability is effectively reduced.

The invention relates to the technical field of automobile pedal testing, and discloses an intelligent networked automobile pedal testing device which comprises a shell and a base, the base is arranged at the bottom of the shell, a fixing support is fixedly connected to the surface of the shell, a groove is formed in the top of the fixing support, and a supporting rod is fixedly connected to the groove in the top of the fixing support. A pedal plate is fixedly connected to the bottom end of the supporting rod, and a sensor is fixedly connected to the upper surface of the pedal plate, so that the problem that a test result is lack of objectivity due to the influence of subjective consciousness and individual differences of testers during manual testing is avoided, and the problems of fatigue, inattention and the like in the testing process are solved; according to the invention, the test result is not accurate due to the fact that the test device needs to be manually treaded for testing, manual operation and errors are reduced, the labor cost and risks are reduced, pressure can be flexibly adjusted for testing, real use conditions can be stably simulated, and the accuracy and reliability of the test result are ensured.

The invention discloses a vehicle lane changing intention recognition method, which comprises the following steps: acquiring own vehicle information, lane line information and target vehicle information based on an own vehicle sensing system, and determining the category of a judgment signal used for recognizing the lane changing intention of a target vehicle according to the own vehicle information and the target information. And after the category of the used judgment signal is determined, obtaining the value of the judgment signal corresponding to the category according to the target vehicle information and the lane line information, and determining the lane changing intention of the target vehicle according to the value. Namely, when the lane changing intention is identified, the own vehicle information, the lane information and the target vehicle information are comprehensively considered, and the information on which the lane changing intention identification depends is increased, so that lane changing intention identification errors caused by inaccurate sensed target vehicle information are avoided, and the stability and robustness of cruise are improved.

The invention discloses a novel automatic connecting piece for shield tunnel segments and a machining method, and relates to the field of shield tunnel engineering.The connecting piece comprises a C assembly pre-buried in concrete of a first segment, and the surface of the C assembly is flush with the surface of the concrete; the O assembly is pre-buried in concrete of the second segment, and a horn-shaped guide hole is formed between the end of the O assembly and the surface of the concrete; the component I is assembled in the component C and is used for being inserted into the component O when the duct pieces are assembled; wherein the component C comprises a main body frame for accommodating the component I, and a sliding mechanism is arranged in the main body frame; and the component I can move in the joint plane through the sliding mechanism. By arranging the C assembly with the guiding and sliding functions, the I assembly serving as the middle connecting body and the O assembly with the locking function, the problem that an existing connecting piece is complex in functional structure and difficult to accurately match can be solved, and therefore the efficiency and accuracy of segment splicing are improved.

The application discloses a question and answer data processing method and device, and a specific embodiment comprises the following steps: in response to a question and answer request, determining a request field, calling a knowledge base to obtain corresponding implication relationship words and synonymous relationship words; performing word segmentation processing on a sentence in the question and answer request, and then removing implied segmented words in the sentence after the word segmentation processing based on the implication relationship words to obtain a to-be-searched sentence; calling a preset corpus, performing similarity calculation on the to-be-searched sentence through the synonymous relationship words to obtain a sentence with the highest similarity and outputting the sentence. Thus, the application can solve the problem that the similarity calculation result is deviated due to redundant information and field synonymous words in a sentence in the prior art, and overcome the shortcoming of the labeling requirement of a large amount of data in a supervised mode.

The aim of the proposed solution is to introduce additional available knowledge about the geometry of the object to be detected into the training of the object detection system beforehand. The aim of the invention is to improve robustness and reduce data requirements compared to a pure data driven scheme. These belong, exactly for industrial applications, to the major obstacle of using AI schemes in critical and dynamic environments, such as in workshops and in autonomous systems. In another aspect, knowledge of shapes that should be detected in these scenes by an object detection scheme has generally been defined and readily available. According to the invention, knowledge representation which is widely used industrially can be used, so that the application of a machine learning system in the industrial field is further simplified.

The present application relates to photovoltaic power generation fault diagnosis technical field, especially in photovoltaic power generation abnormal diagnosis method and system of multi-source data fusion, method is: S1, obtains meteorological data, photovoltaic power generation data, photovoltaic equipment state information, photovoltaic panel image, historical power generation anomaly diagnosis record;S2, the data obtained is preprocessed;S3, constructs photovoltaic power generation prediction model, predicts the theoretical power generation of photovoltaic;S4, set power generation anomaly deviation threshold, compare the theoretical power generation with actual power generation, detect whether photovoltaic power generation is abnormal;S5, according to photovoltaic power generation abnormal detection result, comprehensive analysis meteorological data, photovoltaic equipment state and photovoltaic panel image, analyze photovoltaic power generation abnormal reason.The present application realizes the accurate modeling and real-time abnormal perception of photovoltaic power generation behavior, significantly improves the intelligent operation and maintenance efficiency of photovoltaic power station, reduces the artificial inspection cost, enhances the economy and sustainability of photovoltaic system.

The invention relates to the technical field of artificial intelligence, and provides a scene text recognition method and device, equipment and a medium. The method comprises the steps of obtaining a to-be-recognized scene text image; inputting the scene text image to be recognized into a scene text recognition model, wherein the scene text recognition model comprises a visual converter encoder and a decoder based on a Mama model; extracting a visual feature sequence of the to-be-recognized scene text image through a visual converter encoder; and decoding the visual feature sequence through a decoder based on a Mama model to generate a text sequence corresponding to the scene text image to be recognized. According to the method, the visual converter encoder is combined with the decoder based on the Mama model, the scene text recognition precision is guaranteed, meanwhile, efficient sequence decoding is achieved through the linear calculation complexity of the Mama decoder, reasoning delay is remarkably reduced, and the method is particularly suitable for the edge deployment environment with limited calculation resources.

A lithium ion battery state of health prediction method and system based on quantum gate coupling. The cycle operation data of the lithium ion battery is acquired, the health characteristics are constructed, and the time series samples are constructed. A quantum enhanced gate recurrent neural network is constructed, a plurality of parameterized quantum circuits are configured in at least one gate module, and a quantum gate coupling mechanism is established between different gate modules. The quantum feature representation output by the plurality of parameterized quantum circuits is fused with the inter-gate coupling information, and the gate function value is calculated. The gate function value is used to update the recurrent state, and finally the SOH prediction value is output. The application enhances the single gate expression capability through the structured combination of multiple quantum circuits, realizes information interaction and collaborative calculation through inter-gate coupling, and improves the nonlinear modeling capability of the model for the battery degradation process and the cross-cell generalization capability.

This invention relates to a method and system for scanning and superimposing correction of dominant noise source distribution in passive source multichannel surface wave analysis. The method first determines the stable frequency band and velocity scanning range. Then, within the stable frequency band, at each scanning velocity, a frequency-azimuth domain energy spectrum scan is performed, incorporating the azimuth of the two-dimensional observation station. Next, based on the energy distribution of the frequency-azimuth domain dispersion spectrum, the azimuth range of the dominant noise source distribution corresponding to each velocity is selected, resulting in the final azimuth range of the dominant noise source distribution. Finally, the dispersion spectrum after azimuth correction of the dominant noise source distribution is obtained. This invention considers the azimuth of the observation station pair and can be applied to any two-dimensional observation array. Furthermore, through the reasonable determination of the stable frequency band and velocity scanning range, the analysis of the dominant noise source azimuth range, and the weighted superposition of the azimuth-corrected dispersion spectrum, the calculation quality of the dispersion spectrum under non-uniform noise source distribution conditions is effectively improved, and it has high fault tolerance.

The application relates to the technical field of artificial intelligence, in particular to a multi-scale causal shunt prediction method and system for an online soft measurement of a cut tobacco process. Multivariate process data and outlet material moisture data of the cut tobacco process are acquired; the data is subjected to time synchronization, missing value processing, abnormal value processing and standardization preprocessing, and a sliding time window sample is constructed; input features are divided into local path features and global path features along a channel dimension, the local path features are subjected to local mapping, and the global path features are subjected to attention calculation with a causal mask; the shunt modeling result is input into a multi-scale causal coding backbone network to extract hierarchical features, and the different hierarchical features are subjected to time alignment and cross-layer fusion; local transient disturbance and high-frequency change information are extracted through a high-frequency residual branch; the main road fusion features and the high-frequency compensation features are jointly input into a prediction head to output an outlet material moisture prediction result. The application aims to solve the problem of how to improve the online soft measurement precision of the cut tobacco process and reduce the calculation overhead.

The application discloses an ecosystem productivity estimation method based on LSTM prediction of optimal vegetation temperature, and relates to the technical field of ecological information, comprising: obtaining historical meteorological, topographical and future climate data; constructing a spatial neighborhood of a target pixel, quantifying climate connection costs based on topography and wind field and generating anisotropic spatial features; solving a dynamic spatiotemporal confidence factor representing microclimate complexity; inputting the anisotropic spatial features and historical air temperature into a long short-term memory network, using the confidence factor to regulate fusion weights through a dynamic gating mechanism, and predicting optimal vegetation temperature; constructing a decoupling model containing thermal kinetic energy and a nonlinear thermal stress term, and calculating productivity based on the predicted value and future air temperature; the application can accurately depict microclimate heterogeneity and vegetation thermal adaptation mechanism, and effectively improve the estimation accuracy of ecosystem productivity under complex climate scenarios.