Research On Fast Recognition And Location Technology Of Multispectral Target

In small target recognition and tracking problems,targets often lack effective spatial structure information due to their long distance or small volume.When the target is located in a complex background,traditional recognition methods that only utilize single features such as color,texture,and shape are difficult to obtain satisfactory results,so a more robust method is needed.A multispectral image is a three-dimensional data cube composed of multiple single spectral images.Unlike traditional wideband images,multispectral images have more complete spatial and spectral information.Due to the unique spectral features of different objects,using multispectral image data to identify and track small targets can make up for the shortcomings of traditional target recognition and tracking methods.However,traditional imaging spectrometers have a slower speed in obtaining data and cannot meet real-time requirements.In recent years,with the maturity of snapshot imaging spectrometers,it has gradually become possible to apply multispectral imaging technology to small target recognition and tracking problems.This article involves the research on fast recognition and positioning technology for multispectral targets,and the main work is as follows:(1)Multispectral image preprocessingThis chapter first elaborates on the imaging principle and data characteristics of the eight spectral segment snapshot multispectral camera.In response to the problem of spatial resolution loss in the original spectral data,a multispectral de mosaic algorithm based on an improved guided filter is proposed.This algorithm utilizes the spatial and spectral correlation of pixels to estimate unsumpleted pixels through interpolation and obtain a three-dimensional data cube;To eliminate random noise during image acquisition,four methods are used: derivative operation,smooth filtering,multivariate scattering correction,and variable standardization;To verify the accuracy of data obtained by the snapshot multispectral camera,its spectral response function,energy utilization efficiency,and spectral matching were tested in a laboratory environment.(2)Research on small target recognition and location algorithm based on Spatial-Spectrum information combinationIn the research process of multispectral small target recognition algorithms,the main focus is on the rapid recognition of small target in complex backgrounds.This chapter proposes a small target recognition algorithm based on spatial spectral anomaly coefficients,which combines the characteristics of the collected data.The algorithm uses multi-scale local contrast algorithm and global anomaly detection algorithm to calculate spatial anomaly coefficients and spectral anomaly coefficients,and combines them through multiplication to obtain spatial spectral anomaly coefficients.The target points are then filtered using threshold segmentation;Subsequently,three different target positioning methods were elaborated in detail,and after considering the positioning accuracy and speed,the grayscale centroid method was selected as the target positioning algorithm in this paper;Finally,the superiority of the proposed algorithm was verified through simulation and comparative experiments.(3)Research on moving target tracking algorithm based on spectral featuresThis chapter introduces the principles of generating class tracking algorithms and discriminating class tracking algorithms,and selects correlation filtering as the basic framework for tracking.During the implementation of the tracking algorithm,a multidimensional spatial spectral gradient histogram is used to extract target features,and a position filter and a scale filter are used to estimate the position and scale information of the target,respectively.Simulation experiments show that the tracking accuracy of this algorithm can reach more than 96%,which can meet the needs of practical situations.