10 results about "Spectral distance" patented technology

This invention discloses a hyperspectral target detection method based on adversarial autoencoders and attention mechanisms, comprising: performing coarse detection on an input hyperspectral image; obtaining background samples as clean as possible as training data by adjusting the binarization threshold of the coarse detection result image; training a model based on adversarial autoencoders and attention mechanisms using the prepared training data to obtain a model that can accurately reconstruct the background of the hyperspectral image; inputting a test hyperspectral image into the network for reconstruction; calculating the spectral distance between the reconstructed result and the original hyperspectral image pixel by pixel to obtain a distance map; and applying background suppression to the distance map to obtain the final detection result image. This invention introduces a spectral attention mechanism to assign weights to the spectral dimensions, accelerating the training and convergence speed of the network; and overcomes the problem of imbalanced positive and negative samples in hyperspectral images by selecting only background samples as training data to train the image reconstruction model based on adversarial autoencoders.

A method of Raman spectroscopy for determining concentrations of at least one target component included in a medium including a given combination of multiple components includes: for each component of the medium, providing a reference spectrum of the component; based on Raman peaks included in the reference spectra, identifying Raman peaks that occur within less than a predetermined minimum spectral distance from each other as disturbing peaks; for each component, determining a component spectrum by eliminating each Raman peak included in the reference spectrum of the respective component identified as a disturbing peak; based on the component spectra, determining synthetic spectra of samples of the medium including different concentrations of the components; and based on the synthetic spectra, determining and providing a model for determining concentrations of each target component based on measured spectra of samples of the medium.

This application provides a method and system for adaptive process control in laser cutting, belonging to the field of laser cutting technology. The method involves: acquiring and preprocessing multimodal data of the material to be cut; extracting multidimensional feature vectors characterizing material properties; inputting these vectors into a deep neural network classification model; outputting a preliminary probability distribution of the material type; combining this distribution with electromagnetic and visual features as independent evidence; and using D-S evidence theory for fusion reasoning to determine the final material type. Then, based on a Kalman filter algorithm, spectral distance and geometric contour data are fused to obtain a final thickness estimate. Anomaly detection is performed using the final material type and final thickness estimate, triggering anomaly handling. Based on the final material type and thickness estimate, a preset database is queried to obtain a set of basic process parameters, which are then fine-tuned using a parameter optimization algorithm and configured into the CNC system of the laser cutting machine. This application achieves adaptive control of laser cutting through multimodal fusion recognition.

The invention relates to a system (100) for object detection in road traffic for a vehicle, comprising: an object detection system (101) for detecting individual objects OBJ N is installed and set up in the vehicle's environment; at least one optical sensor (102) for capturing spectral image data BD of the environment with the individual objects OBJ N is designed and configured, a gaze direction detection unit (103) designed and configured to detect the current gaze direction BR of a driver of the vehicle; an evaluation unit (104) designed and configured to detect those objects OBJ n * of the objects OBJ N to identify those that are arranged one behind the other in the current line of sight BR of the driver, for the objects arranged one behind the other OBJ n * based on the spectral image data BD spectral distances SD i = SD i(OBJ n *) of the surfaces of these objects depicted in the image data OBJ n * to determine, and provided that at least one of the spectral distances SD i for the objects arranged one behind the other OBJ n * If the value is less than a predefined limit value GW, a warning signal WS1 = WS1 (SD) is triggered. i , GW) to generate, and an output unit (105) to output a warning triggered by the warning signal WS.

A loss function optimization system for a 3D medical image segmentation deep learning model, according to the present invention, comprises: an image input unit for receiving an input of a 3D medical image; a masking unit for masking a partial region of the input 3D medical image; a deep learning model unit for performing masked image modeling mechanism deep learning on the masked 3D medical image by using a deep learning model; a pixel area reconstruction unit that calculates a pixel-based loss function during learning of the deep learning model unit and performs backpropagation; a frequency domain reconstruction unit that calculates a frequency-based loss function during learning of the deep learning model unit and performs backpropagation; a loss function optimization unit that calculates a spectral distance scaling loss function in which the pixel-based loss function and the frequency-based loss function are combined, and performs backpropagation on the deep learning model of the deep learning model unit; and a pre-trained model generation unit for generating a pre-trained model that has been trained through the spectral distance scaling loss function.

This application discloses a method and related apparatus for generating high spatiotemporal resolution images based on a spatiotemporal fusion algorithm, relating to the field of remote sensing image fusion technology. The method includes: for any target raster pixel, initial screening is performed according to reflectance differences within a moving window centered on the position of the target raster pixel; then, a final set of similar raster pixels is obtained by screening based on spectral distance, spatial distance, temporal distance, and corresponding thresholds; when calculating the spectral distance, different combinations of coefficient constants are used according to the different bands of each similar raster pixel to obtain a more accurate spectral distance calculation result; then, based on the weight matrix calculated therefrom, remote sensing images with high temporal and low spatial resolution and high spatial and low temporal resolution are effectively fused, solving the shortcomings of existing remote sensing products in terms of spatiotemporal resolution, especially the limitations of low-cost remote sensing products, which often have high temporal resolution but low spatial resolution, or high spatial resolution but low temporal resolution.

This invention provides a method, apparatus, processor, and storage medium for object detection based on lidar. The method further includes: transmitting signal light emitted by a laser to a target object via optical devices and an optical phased array transceiver; detecting end-face reflection signals and target reflection signals, wherein the end-face reflection signal is formed by reflection from the optical phased array transceiver and optical devices, and the target reflection signal is formed by reflection from the target object; determining the spectral distance between the end-face reflection frequency and the target reflection frequency based on the end-face reflection signal and the target reflection signal; adjusting the spectral distance if it is less than a distance threshold; determining the passband frequency and stopband frequency of a filter based on the adjusted spectral distance; filtering out the end-face reflection signal from the target reflection signal according to the passband frequency and stopband frequency; and detecting the position of the target object based on the filtered target reflection signal. This invention solves the technical problem of inaccurate target object position information detection.

The invention belongs to the technical field of marine environment monitoring and prediction, and discloses a non-linear sea wave reconstruction and prediction method and system based on coherent radar, and the method comprises the steps: obtaining sea wave data through the coherent radar; processing the sea wave data to obtain the radial speed of the sea surface; constructing a propagation operator matrix based on a linear wave theory; introducing a second-order nonlinear correction term to correct the linear wave theoretical model; solving a nonlinear equation containing the second-order nonlinear correction term to obtain a model coefficient; and reconstructing and predicting the wave height time calendar based on the model coefficient. The sea surface elevation is predicted by using the coherent radar, so that more reliable sea wave data can be obtained; after the radial velocity of the sea waves is obtained through a spectral distance method, band-pass filtering is carried out on the radial velocity, and therefore the more accurate radial velocity of the sea waves is obtained. By introducing a second-order nonlinear term, the method has higher reconstruction and prediction capabilities on nonlinear sea waves, and finally, integration of sea wave reconstruction and prediction is realized.

This invention discloses a hyperspectral image classification method based on semantically constrained dual-graph complementary learning. It acquires hyperspectral images of ground object samples, forming a labeled sample set and an unlabeled test sample set. For the labeled sample set, an intra-class sample weight modulation mechanism is introduced to adaptively adjust the edge weights of intra-class sample connections, constructing intra-class similarity and homogeneity graphs and intra-class dissimilarity and homogeneity graphs. For the global sample set, a global similarity graph is constructed based on a spatial-spectral distance metric. The labeled sample set and the global sample set are combined to obtain two augmented datasets corresponding to intra-class similarity and homogeneity graphs and intra-class dissimilarity and homogeneity graphs, respectively. A dual-branch GCN framework is proposed, with the main branch and auxiliary branch trained on the augmented datasets based on the intra-class similarity graph structure and the intra-class dissimilarity graph structure, respectively. The auxiliary branch assists in optimizing the main branch for extracting ground object features. During the testing phase, the main branch is used to predict the category of unknown samples. This invention significantly improves the accuracy, generalization ability, and robustness of hyperspectral image classification.

The invention relates to the technical field of image processing, and provides an outdoor image highlight suppression method and system based on multispectral dynamic compensation, and the method comprises the steps: obtaining a second-order mixed partial derivative according to the radiation brightness of a current spectral band; obtaining a wavelength attenuation coefficient and a spectral distance between the current spectral band and a reference band; obtaining a sensitivity score of the current spectral band according to the image space domain, the second-order mixed partial derivative, the wavelength attenuation coefficient and the spectral distance; and selecting an optimal reference wave band according to the sensitivity score, and performing highlight suppression on the outdoor image based on the optimal reference wave band. The optimal reference band is selected according to the sensitivity score, a multispectral band sensitivity analysis and dynamic reference band selection mechanism is constructed, the spectrum-space joint sensitivity function is introduced for the first time, the reference band is adaptively selected in combination with the platform height, the limitation of fixed band reference in a traditional method is solved, and the accuracy of the method is improved. And spectrum consistency recovery can be realized under the condition of strong light change.