37results about How to "Reduce the false alarm rate of detection" patented technology

The invention discloses a fault detection, diagnosis and performance evaluation method for a redundant aileron actuator. According to the method, fault detection, diagnosis, evaluation and real-time detection of the actuator are performed by means of an input order signal, an output displacement signal, a force motor current signal and aerodynamic loading data of the actuator; the fault detection is realized by a two-stage neural network, a first neural network is used as a system observer and is matched with actual output to acquire a residual error, and a second neural network outputs a self-adaptive threshold value synchronously; the fault detection is realized by the system observer and a force motor current observer; a time domain feature is extracted from a residual error signal and output to a self-organizing mapping neural network, and a minimum quantization error is acquired and normalized to a health degree, so that the actuator performance is evaluated; and on the basis of fault detection, the aerodynamic loading data is introduced, by means of a specific input order spectrum, the system observer and the neural network with the self-adaptive threshold value are trained, and the real-time fault detection is realized.

The invention discloses an unmanned aerial vehicle video small-object detecting method based on super-resolution reconstruction. The unmanned aerial vehicle video small-object detecting method comprises the steps of selecting one input image as a reference frame, and selecting three consecutive image frames for performing sub-pixel displacement estimation on the reference frame; then placing the displacement estimation result of the four image frames into a high-resolution image grid; and estimating pixels which are lost in the high-resolution image grid, thereby obtaining an objective image with a relatively high resolution. Afterwards, an objective template is extracted from the objective image, and the characteristic of the objective image is calculated. Then the reconstructed objective image is divided for obtaining a plurality of objective area blocks. Characteristic extraction and characteristic identification are successively performed on all objective area blocks, thereby finishing preliminary detection for the object. Afterwards, false object elimination is performed, thereby obtaining a final detection result.

Embodiments of the present invention provide a method for detecting power transmission towers based on synthetic aperture radar images, by processing high-resolution synthetic aperture radar (Synthetic Aperture Radar, SAR) images to obtain a region of interest (Region of Interest, ROI), and Calculate the semi-variance feature of the corresponding area, and use the neural network to detect the transmission tower. The present invention does not require that the image must be an expensive full-polarization SAR image, and for the same SAR image, especially when the outline of the high-voltage transmission tower on the image is not clear and there are many false targets, the present invention can detect More transmission tower targets, and the detection false alarm rate is lower than the existing technology.

The invention discloses a hyperspectral remote sensing small target detection method based on multiple aperture information processing. The hyperspectral remote sensing small target detection method based on the multiple aperture information processing comprises data observation, information processing and target detection. In the data observation, distributed parallel partitioned information processing performed on hyperspectral remote sensing data through a multiple aperture asynchronous mapping mechanism is simulated, partitioned data collection is performed on the hyperspectral remote sensing through a data collection chip, multiple aperture information obtaining through a fly visual system and a mapping mechanism are simulated, and the hyperspectral remote sensing spectrum data is decomposed and recombined according to an ommatidium mapping function to form into a spectrum data cube which is about local ground objects and construct a distributed parallel partitioned information processing mode. In the information processing, a cartridge system of the fly visual system is simulated to achieve integration of redundant compression and super sensitivity. In the object detection, a self-adaption mechanism of optic high order nerve cells of the fly visual system is simulated and an anomaly detection result is obtained through a self-adaption small target detection algorithm in combination with judgment of whole situation anomaly and local anomaly.

The invention discloses a graph-domain multi-scale infrared weak and small target detection method based on local gradient trilaterals. The method is as follows: converting an infrared image containing weak and small targets into a local gradient trilateral represented by a node and its edge weight relationship Graph signal; Secondly, according to the multi-scale transformation of the graph Laplacian matrix, the local gradient trilateral graph signal is multi-scale decomposed to obtain the low-frequency sub-bands of the graph signal at different scales; thirdly, according to the edge weight of the graph signal The relationship locally weights the high-frequency sub-bands of each scale, and takes the median value as the new central node coefficient, and then performs multiplicative fusion on the high-frequency sub-bands after local weighting; finally, the high-frequency sub-bands after multiplicative fusion The sub-band is segmented by adaptive threshold, the spatial position of the target is determined, and the detection result is output.

The present invention provides a hyperspectral urban water body detection method based on adaptive sample selection. In the preprocessing of hyperspectral near-infrared band images, the noise band images are removed through the quality evaluation SSIM method, and the noise is further eliminated by using two average operations. , can obtain a more stable mean image of the near-infrared spectrum, compared with the traditional single-band threshold segmentation method, there is no need to manually select the image to be segmented; the invention uses the unsupervised threshold segmentation method to extract the suspected water body area, and then through the supervised Feature learning and classifier training, remove building shadows, building roof asphalt and other ground objects that are similar to water in the near-infrared spectrum from the suspected water body area; therefore, the present invention combines the unsupervised threshold segmentation method with supervised feature learning Combined with the classifier training method, it has the self-adaptive ability of the urban observation scene, and can realize the "real-time collection and real-time processing" of the measured scene data, and the false alarm is low.