5 results about "Wavelet filter bank" patented technology

The invention relates to the technical field of tobacco production data management, in particular to a cigarette quality index feature extraction method based on improved empirical wavelet transform. The method comprises the following steps: collecting and preprocessing original signals of cigarette quality indexes; fourier transform is carried out on a time domain signal obtained after preprocessing to obtain a single-side amplitude frequency spectrum, tracks of local minimum values are detected and tracked in a scale space of the single-side amplitude frequency spectrum, and the length of each track is calculated; according to the track length distribution, an adaptive threshold algorithm is adopted to screen out effective spectrum boundary points; constructing an empirical wavelet filter bank based on the boundary points, decomposing the signal to obtain a plurality of modal components, and screening out a key modal component according to a preset rule; and finally, extracting statistical characteristics of the key modal components as quality index characteristics to be output. According to the method, the dependence of a traditional method on a fixed modal number and an experience frequency band can be overcome, the accuracy and stability of feature extraction are remarkably improved, and effective support is provided for accurate control of cigarette quality.

The invention belongs to the technical field of image processing, and particularly discloses a fruit detection image denoising method based on graph signal processing, which comprises the following steps: acquiring an original fruit image, and constructing a multi-layer non-Euclidean graph structure model of the original fruit image; constructing a tight support biorthogonal graph wavelet filter bank, wherein the filter bank comprises an analysis filter bank and a comprehensive filter bank; performing multi-scale decomposition, filtering and inverse transformation reconstruction on the multi-layer non-Euclidean diagram structure model by using the tight support biorthogonal diagram wavelet filter bank to obtain a de-noised fruit image; the image preprocessing algorithm can effectively suppress noise caused by factors such as external and sudden unstable factors, uneven illumination, equipment vibration and the like of transmission channel characteristics when the industrial camera collects fruit images, and has a good denoising effect on multiplicative noise in the fruit images. And normal judgment of foreign matter detection of a visual system on a production line can be guaranteed.

The application discloses a hyperspectral anomaly detection method fusing graph attention and beta wavelet graph network, which regards the pixel of a hyperspectral image as a graph node, regards spectral features as node attributes, establishes edge connection through a K nearest neighbor algorithm, and constructs a graph data structure containing spatial and spectral information. Firstly, the graph attention network is used to dynamically aggregate neighborhood information and enhance spatial context features. Secondly, a beta wavelet filter bank is constructed based on a graph Laplacian matrix, multi-scale frequency domain filtering is performed on node features, and the right shift phenomenon of spectral energy caused by an abnormal target is effectively captured. Finally, multi-scale filtering features are fused and mapped into an abnormal probability, and pixel-level anomaly detection is realized. Through the joint mining of nonlinear spatial correlation and unique spectral anomaly features of hyperspectral data by the graph neural network, the accuracy and robustness of anomaly detection in a complex scene are significantly improved.

The invention discloses an industrial furnace working condition mode recognition system based on image recognition. The system comprises an image acquisition module, an edge processing module, an image texture analysis module, a data association module and a working condition recognition module. The image acquisition module can acquire and preprocess source image data in the furnace through a near-infrared camera and a polarization filter, and then construct a thermal radiation distribution diagram reflecting material component differences; and the edge processing module can perform frequency domain decomposition of scale and direction on source image data through a Log-Gabor wavelet filter bank algorithm to obtain a pixel phase value. According to the invention, the image acquisition module adopts the combination of a near-infrared camera and a polarization filter, an image data basis is provided for subsequent texture analysis, and the edge processing module can extract closed continuous edge lines formed at the initial stage of crusting, so that the detection sensitivity of fine textures is improved, and the detection precision is improved. And meanwhile, a periodic structure of a crusting area and random textures generated by material flowing can be effectively distinguished.

The invention discloses a nuclear power pump bearing anomaly detection method, system, medium and equipment based on an empirical wavelet network, and the method comprises the steps: obtaining a vibration signal of a nuclear power pump bearing; preprocessing the vibration signals, performing spectral analysis on the preprocessed vibration signals, determining a plurality of characteristic frequencies based on local maximum detection, adaptively dividing a plurality of frequency bands according to characteristic frequency intervals, and constructing empirical wavelet filter banks corresponding to the frequency bands; constructing an empirical wavelet network, performing sub-band decomposition on the training samples of the training set by using the empirical wavelet filter bank, inputting each sub-band signal into the empirical wavelet network to extract sub-band features and perform signal reconstruction, accumulating all sub-band reconstruction signals to obtain an overall reconstruction signal, and performing reconstruction on the overall reconstruction signal; in the training process, minimizing a reconstruction error is taken as an optimization target; and inputting a test sample into the trained empirical wavelet network, calculating a reconstruction error of the empirical wavelet network, and determining the health state of the bearing according to a comparison result of the reconstruction error and a threshold value.