A hyperspectral image classification algorithm for few shot contexts
Gabor filter is a common spatial feature extraction technique.In order to address the problem of sparse labelled samples in hyperspectral image classification,the algorithm in this paper generates a large number of multiple views by setting multiple 3D-Gabor filters in different directions.The algorithm generates multiple graph connections on the basis of the multi-view data to achieve label propagation,and fuses the classification results of multiple graph labels propagated to obtain the predicted label results.The superpixel principal component analysis(Super PCA)algorithm is a simple but very effective unsupervised feature extraction method,where the prediction results are weighted and fused with the classifier incorporating Super PCA to obtain more accurate classification results.Simulations of this algorithm on three datasets show that traditional hyperspectral image classification algorithms using Gabor filters are computationally intensive and time-consuming,whereas this algorithm can reduce computational and time consumption while ensuring accuracy and cost savings.
few shothyperspectral image classification3D-Gabor filtermultiviewlabel propagationsuperpixel segmentationsemi-supervised learningactive learning
万方数据
维普
