Threshold estimation sample enhancement method for hyperspectral remote sensing image deep learning classification

The invention discloses a hyperspectral remote sensing image deep learning classification threshold estimation sample enhancement method, and the method comprises the steps: obtaining hyperspectral remote sensing image public data, and selecting a plurality of samples according to each type to form an initial training sample set; training the initial training sample set by using a twin network as a classifier, and predicting the hyperspectrum pixel by pixel to obtain a prediction result presented by a classification probability matrix; inputting the classification probability matrix into a threshold estimation module to adaptively estimate each type of threshold, and selecting samples with prediction probabilities higher than the corresponding type of threshold in the classification probability matrix as pseudo samples to obtain a pseudo sample set; and detecting and correcting the pseudo sample set to obtain a corrected pseudo sample set, putting the corrected pseudo sample set into the initial training samp