13results about How to "Low robustness" patented technology

This invention discloses a method and apparatus for real-time damage detection of plate structures based on a physical information spatiotemporal graph neural network in the field of structural health monitoring technology. The method includes: constructing a three-dimensional physical mesh model of the plate structure based on the acquired dimensional information and material properties of the plate structure to be monitored; generating a sensor and excitation source layout in the physical mesh model based on a pre-determined sensor placement range; arranging corresponding sensors and excitation sources on the plate structure to be monitored based on the sensor and excitation source layout; acquiring feedback signals received by the sensors after pulse signals are emitted to the plate structure by the excitation sources; and sending the feedback signals to a pre-trained physical information spatiotemporal graph neural network model to generate the location, area, major axis length, minor axis length, orientation angle, and severity index of each damage. This invention achieves low-cost and highly robust structural health monitoring.

The invention discloses an imaging method for passing through a random scattering medium in visible light based on a diffractive optical neural network, and the method comprises the following steps: 1, carrying out the optical coding of an input image through a coherent or low-coherent monochromatic plane wave, and enabling a coded light field to pass through a random phase diffuser for scattering; and step 2, continuous wavefront modulation of the scattered light field is obtained through a coherent or incoherent diffractive optical neural network, and finally, high-fidelity original image reconstruction is generated on an output plane. The invention finds that the coherent diffraction neural network obtains dynamic phase modulation by introducing a randomly generated diffuser in the training process, enhances the adaptive capacity of the network to the spatial coherence change of the light source, realizes the high-quality reconstruction of the scattering image under the visible light, has the robustness to the low-coherence light source, and is suitable for large-scale popularization and application. The method can be applied to low-power-consumption and real-time imaging of the dynamic scattering medium under natural light.

The invention discloses a rehabilitation training human body posture estimation method and system based on a conditional space-time diagram diffusion model, and belongs to the field of artificial intelligence rehabilitation medicine. The system comprises a multi-view image acquisition module, a two-dimensional attitude detection module, a rough three-dimensional reconstruction module, a standard action library module, a condition space-time diagram diffusion optimization module and an output module. The method comprises the following steps: acquiring a video through a low-cost multi-view camera, and obtaining a rough attitude sequence through two-dimensional detection and three-dimensional reconstruction; then, semantic features are matched and extracted on the basis of a standard action library and serve as conditions to guide a condition space-time diagram diffusion model to conduct space-time joint optimization on the rough sequence, and a high-quality and smooth three-dimensional posture sequence conforming to clinical semantics is generated. According to the method, a collaborative architecture of a lightweight front end and an intelligent rear end is constructed, the precision and smoothness of human body posture estimation in a rehabilitation scene are remarkably improved on the premise that the hardware cost is controllable, and a practical and reliable intelligent rehabilitation evaluation tool is provided for medical staff.

The application discloses a semi-supervised HRRP noise label filtering method based on early learning guidance, comprising the following steps: constructing an HRRP training data set and a deep neural network model; pre-training the deep neural network model; predicting the HRRP training data set by using the pre-trained network model and dividing the HRRP training data set into two data sets according to the confidence; constructing a known label data set and an unknown label data set to form a semi-supervised training data set and performing data enhancement; semi-supervised training the network model by using the enhanced data; generating the label of the unknown label data set by using the semi-supervised trained network model and fully supervised training the network model; and target recognition of the HRRP data by using the fully supervised trained network model. The application uses a generalized cross-entropy loss function in the early learning training stage of the model, proposes a data enhancement scheme suitable for HRRP, can better reduce the influence of noise labels on the model, and has better stability and implementability.

The application discloses a park enterprise inter-coordinated carbon reduction method, a storage medium and a computer program product, relates to the technical field of carbon management, and the method comprises the following steps: acquiring supply-side data and demand-side data reported by park enterprises, and matching a priority weight for each supply-side data; constructing a space-time coupling factor of potential supply pairs existing in the supply-side data and the demand-side data, and calculating an actual schedulable coordination amount of the potential supply pairs according to the space-time coupling factor; calculating an expected carbon emission reduction amount according to the actual schedulable coordination amount and the priority weight, and generating a candidate coordinated carbon reduction scheme based on the expected carbon emission reduction amount; inputting the candidate coordinated carbon reduction scheme into a pre-constructed digital twin simulation unit for verification to obtain a coordinated carbon reduction scheme, so as to improve the coordinated carbon reduction effect and the scheme execution success rate among park enterprises.

The application discloses a beam angle prediction method based on neural network and Kalman filtering fusion, and belongs to the technical field of millimeter wave communication.The method adopts a single-layer neural network without an activation function to learn the behavior of LSTM, extracts a weight matrix of the single-layer neural network as a state transition matrix of a Kalman filtering algorithm, and performs beam tracking by executing the Kalman filtering algorithm.The application fuses the strong robustness of the Kalman filtering algorithm on the basis of accurate prediction of the beam tracking angle; and since the LSTM has strong learning ability, dense observation of a communication target is not needed, the observation interval can be artificially set according to cost, the prediction cost is greatly reduced, and the performance of the beam tracking system is improved.

The embodiment of the invention provides a multi-node UWB relative positioning method and system based on a liquid neural network. The method comprises the steps that observation data between known node pairs are collected in real time through UWB; obtaining a multi-frame time sequence feature sequence corresponding to a moment based on observation data between known node pairs acquired in real time; inputting the multi-frame time sequence feature sequence into a liquid neural network model based on an ordinary differential equation, and outputting a prediction distance result between known node pairs; based on the real distance result and the predicted distance result, optimizing the liquid neural network model by using a loss function to obtain a trained liquid neural network model; using the trained liquid neural network model to obtain a prediction distance result between the unknown node pairs; and obtaining the relative position of the unknown node at the current moment based on the predicted distance result between the unknown node pairs and the relative position of the unknown node at the previous moment. The method can give consideration to the real-time performance and robustness of positioning.

The invention provides an over-break and under-break monitoring method, device and equipment for a tunnel, a medium and a product, and relates to the field of tunnel construction. Comprising the following steps: acquiring real-time point cloud data of a to-be-monitored tunnel by adopting a first preset acquisition unit; a second preset acquisition unit is adopted to acquire contour data of the tunnel to be monitored; performing data fusion processing according to the real-time point cloud data and the contour data to obtain a real-time tunnel three-dimensional model of the tunnel to be monitored; obtaining a design three-dimensional model of the tunnel to be monitored; according to the real-time tunnel three-dimensional model and the design three-dimensional model, determining over-break and under-break information of the to-be-monitored tunnel; and displaying the over-break and under-break information on a preset display interface. The technical problem that in the prior art, due to the fact that the modern sensing technology is poor in environmental adaptability, accurate and real-time monitoring of back break is difficult to achieve, and consequently the back break monitoring efficiency of a tunnel is low is solved.

This invention discloses a UAV image transmission signal classification method based on EMA-ResNet50. This method generates a grayscale time-frequency image from the UAV image transmission signal through short-time Fourier transform, introduces an EMA attention module, and preserves complete channel information without channel dimensionality reduction through feature grouping, parallel sub-networks, and cross-spatial learning, enhancing the ability to extract multi-scale features and pixel-level relationships. Furthermore, the EMA is embedded into a ResNet50 Bottleneck structure to construct an EMA-ResNet50 model. Experiments show that this method achieves a recognition accuracy of 99.81% across a distance of 20–150m on the DroneRFa dataset, and still reaches 94.26% at a low signal-to-noise ratio of -15dB. It also achieves 100% accuracy in distinguishing interference signals such as WiFi and Bluetooth, making it suitable for scenarios such as low-altitude security and airport airspace protection.

The invention relates to the technical field of quantum machine learning, in particular to a random forest classification method based on quantum-classical double-feature branch fusion. The problems of parameter solidification and precision limitation of a classic module are solved. According to the technical scheme, the method comprises the following steps that S1, data preprocessing and feature engineering are executed; s2, constructing dynamic quantum feature mapping; s3, constructing a quantum decision tree; and S4, constructing quantum-classical soft voting integration. Compared with the prior art, the method has the advantages that quantum resource consumption is greatly reduced, and adaptability is improved. The over-fitting risk is significantly inhibited, and the generalization ability is enhanced. And the classification precision and the robustness are improved. And the training efficiency and the universality are obviously enhanced. The engineering landing performance is high, and the synergistic advantage is prominent.

The application discloses a kind of robust fast secret channel construction methods for amplitude using CSS spread spectrum modulation, mainly solve the problems, such as the limited transmission distance and capacity of secret channel in the existing low-power wide-area network environment, steps include: the modulation method design of secret channel;Signal frame structure design and frame detection;The demodulation method design of secret channel;Analysis is carried out to the information loss in the demodulation process of secret signal.The application designs the LoRa physical layer secret channel modulated by amplitude using CSS spread spectrum modulation method, designs the modulation and demodulation method of secret channel, analyzes the information loss in the demodulation process of secret signal, constructs the robust fast secret channel, widens the construction scene of secret channel, and the operation method is simple and clear, has practical value.

The invention relates to an intelligent alignment compensation method for assembly of a liquid crystal panel and a backlight module, and aims to solve the problems of dynamic pose deviation and weight mismatch caused by material batch difference, nonlinear deformation and environmental factors in a high-precision fitting process. According to the core technology, an industrial camera array and a multi-source sensor are used for collecting multi-dimensional micro texture and environment data, a fingerprint spectrum of intrinsic deformation characteristics of a material is established in a fusion mode, spectrum matching and weight self-adaptive calibration are carried out in combination with real-time working conditions and visual feedback, and six-degree-of-freedom micro dynamic correction is achieved through FPGA acceleration driving control. According to the scheme, the sub-pixel-level precision in the attaching process of the liquid crystal panel and the backlight module is effectively improved, the nonlinear deformation trend of materials is restrained, the control instruction response speed is optimized, continuous and accurate calibration of self-adaptive compensation weights among batches is guaranteed, and the process consistency and the automation level are improved.

This application relates to a road surface elevation prediction method, apparatus, electronic device, and storage medium. The road surface elevation prediction method includes: acquiring a current frame monocular image and a historical frame monocular image of the road surface ahead of a vehicle; extracting features from the current frame monocular image and the historical frame monocular image to obtain first image features and second image features; mapping the first image features to the bird's-eye view BEV space based on real-time extrinsic parameters to generate a current frame BEV feature map; performing temporal-spatial alignment processing on the second image features to obtain a historical frame BEV feature map that is temporally-spatial aligned with the current frame BEV feature map; calculating the correlation between the BEV voxel features corresponding to the current frame BEV feature map and the BEV voxel features corresponding to the historical frame BEV feature map to obtain correlation features; and obtaining a road surface elevation prediction result based on the correlation features. This application can effectively improve the perception and adaptation capabilities of autonomous vehicles to complex road environments.