3 results about "Frequency dependence" patented technology

The present invention provides a target simulation method, device and storage medium, wherein the method comprises: firstly obtaining the first RCS data of the whole space corresponding to the first frequency of the target to be simulated; then determining the strong scattering area of ​​the target to be simulated according to the fluctuation characteristics of the first RCS data, the strong scattering area being the area where the RCS value is greater than the preset threshold; then performing scattering center diagnosis on the strong scattering area to obtain the diagnostic structure of the strong scattering area and the visible angular domain range corresponding to the diagnostic structure; finally adjusting the structure of the diagnostic structure, and performing an equivalence evaluation on the scattering characteristics of the adjusted structure and the diagnostic structure before the adjustment according to the visible angular domain range and the corresponding frequency; if the evaluation equivalence is high, the adjusted and optimized structure is determined as the simulation structure of the target to be simulated within this visible angular domain range. This scheme can simulate the local scattering characteristics of the target in detail, and the simulated scattering characteristics have good fluctuation characteristics and frequency dependence, and the overall fidelity is high.

The application discloses a kind of mechanical fault diagnosis methods based on intra-sequence-inter-sequence information fusion network in the field of deep learning and intelligent manufacturing technology, comprising the following steps: 1) original vibration signal is first converted into frequency spectrum sequence by fast Fourier transform (FFT), as unified input.Signal is split into sub-sample of specific length, and a certain number of sub-samples are selected in order to construct K-Nearest Neighbor (KNN) graph, Radius graph and Path graph respectively;2) these sequences are processed using hybrid 1D-CNN-Bi-GRU feature extractor to capture local spectral morphological features and long-range frequency dependence, and obtain feature vector;3) the structural association between sub-samples is encoded by three kinds of mapping methods, and the neighborhood information is aggregated on these graphs using GATv2 module with dynamic attention, learning topological representation, while introducing dimension alignment to ensure feature consistency;4) using attention network, and, dynamic weighting and fusion are carried out, and residual connection is added, to generate interpretable representation, which can quantify the dependence degree of the model on spectral features and structural features, for final classification;The application solves the problem that existing intra-sequence dynamic features and inter-sequence structural relationships are difficult to consider and do not have self-adaptive and interpretable fusion capabilities.

This invention discloses a method for predicting attribute scattering center parameters based on automatic radar component segmentation and a phase-coordinated network. First, the method utilizes a PointNet++ network with multi-scale grouping and normal vector features to automatically segment complex target point clouds into basic geometric components, and achieves physical diversity of ray data based on the segmentation results. Second, a phase-coordinated network is constructed, endowing the network with electromagnetic interference sensing capabilities through explicit phase encoding. Distributed covariance pooling is used to accurately capture the spatial topological shape distribution of rays. A multi-task decoupled prediction head and a physically constrained sensing loss function are used to achieve high-precision prediction of the attribute scattering center position, field value, amplitude, frequency dependence factor, and length distribution parameters. This invention significantly improves the modeling accuracy and physical consistency of the electromagnetic scattering characteristics of complex targets, and its inference speed is faster than the traditional ESPRIT algorithm, making it suitable for real-time radar target recognition and RCS reconstruction.