Hyperspectral Image Classification Method Combining Active Learning and Neighborhood Information

Abstract

The invention discloses a hyperspectral image classification method combining active learning and neighborhood information. The establishment of label samples in hyperspectral images requires field inspections and traditional methods only consider a single spectral information problem. The implementation steps are: (1) use the initial labeled sample set Xl to train the SVM classifier; (2) use the SVM classifier to select the q samples with the largest amount of information from the unlabeled sample set Xu and label them by experts; Put the q samples marked by experts into Xl; (4) retrain the SVM classifier with the updated Xl; (5) judge whether to exit the loop according to the stopping criterion; (6) use the trained SVM classifier to The test sample set is tested; (7) Use the neighborhood information of each sample in X1 to correct the test result to obtain the final classification result. The invention realizes the space-spectrum combination of hyperspectral images, and can obtain better classification results compared with other similar methods.

Description

technical field [0001] The invention belongs to the technical field of hyperspectral image processing methods and applications, and relates to a hyperspectral image classification method using active learning and neighborhood information at the same time, which can be used for map making, vegetation survey, ocean remote sensing, agricultural remote sensing, atmospheric research, and environmental monitoring. and other fields. Background technique [0002] Remote sensing is a comprehensive technology of earth observation developed in the 1960s. It refers to a technology for long-distance detection and perception of targets or natural phenomena without direct contact. As a new science of comprehensive detection technology, remote sensing has been developed for less than 50 years, while hyperspectral resolution remote sensing is even younger, with a history of less than 30 years. However, because they are based on modern physics, computer technology, mathematical methods and g...

AI Technical Summary

Problems solved by technology

[0008] However, the disadvantage of the traditional SVM method is that a large number of labeled samples are required to participate in the training when training the classifier. However, the more labeled samples, the better. Too many labeled samples will cause redundancy and increase the workload of calculation. ; Moreover, not every sample in a large number of labeled samples is useful for the training of the classifier

Images