Research On Classification Algorithm Of Hyperspectral Image Based On Weighted Nearest Neighbor Space Representation

Hyperspectral images(HSIs)technology is an integrated optical remote sensing technology that combines image and spectral detection.The technology detects the ground object by receiving the ground object reflection from the optical system and acquiring the spectral information of the ground target pixel.At present,the classification of HSIs for ground material information is a frontier research topic of integrated image processing and remote sensing imaging technology.At the same time,how to improve the recognition ability and processing efficiency of spectral similar regions has always been the research focus of HSIs classification.Therefore,this paper proposed two HSIs classification methods combining spectral and spatial information,and validates the effectiveness of the proposed method using real HSIs data.The main work of this paper is summarized as follows:(1)The traditional k-Nearest Neighbor(k-NN)method relies on Euclidean distance as a measure to predict the label of a test sample.However,when each test sample is judged by class,the weight distribution of the distance information tends to affect the classification accuracy.Therefore,this paper proposed a HSIs classification method that combines weighted nearest neighbors and sparse representations.The method includes the following steps: First,the neighborhoods around the test pixels are combined according to the window size to form a joint region of test pixels.Obtain the Euclidean distance between pixels.Secondly,the Gaussian weighting function is used to assign weight information to the distance information of pixel.Finally,the reconstructed residuals of the sparse representation are combined to form a new decision function.The experimental results show that the proposed method has higher classification accuracy.(2)There are often unavoidable miscalibrations in the calibration of HSIs objects,resulting in a small number of uncertain samples(Noisy label)in the training set.Therefore,in order to reduce the effect of uncertain samples on supervised classification,this paper proposed a method based on weighted nearest neighbors for peak density HSI classification.The specific steps of the method are as follows: First,an adaptive homogenous region corresponding to the training sample is obtained by using an Entropy Rate Superpixel(ERS)algorithm.Secondly,the Euclidean distance between each pixel in the homogeneous region is calculated,and the Gaussian weighting function is used to weight the distance information.Next,the error-calibrated samples are detected and removed by the peak density clustering algorithm.Finally,the support vector machine is used to evaluate the effectiveness of the method.The experimental results show that the proposed method can effectively detect and remove uncertain samples and improve the performance of supervised classifiers.In summary,this paper proposed two classification methods respectively based on spectral-spatial joint and noisy label conditions for the label prediction of test samples and the noisy label of training samples,and proved the superiority of the proposed methods by hyperspectral analysis.New ideas are provided in the field of HSI classification research.