A Hyperspectral Open Set Classification Method Jointly Densely Connected Network and Sample Distribution

Abstract

The invention belongs to the technical field of image processing, and discloses a hyperspectral open set classification method combining dense connection network and sample distribution; including data preprocessing, manual labeling of hyperspectral data, bilateral filtering, edge preservation and noise reduction, and principal component analysis Dimensionality reduction and maximum and minimum normalization processing; then use 1D / 2D dense connection network to extract spectral and spatial features of hyperspectral preprocessing data, use SoftMax classifier to obtain the probability value of input data relative to each known class, and take the maximum probability value The corresponding class is its predicted category; use the boxplot method to capture the abnormal classification probability value of the training data, obtain the abnormal value judgment threshold of each known class, and then judge the probability value corresponding to the predicted category of the input data: if If the probability value is greater than the outlier judgment threshold of the predicted category, the input data belongs to the predicted category, otherwise it belongs to the unknown category. The invention combines deep learning and a box-and-whisker method, and can reject unknown classes while classifying known classes.

Description

technical field [0001] The invention belongs to the technical field of image processing, and further relates to a hyperspectral open set classification method for joint densely connected networks and sample distribution in the technical field of hyperspectral image open set classification. The invention can not only classify the known classes appearing in the training process, but also reject the unknown classes not appearing in the training process. Background technique [0002] Hyperspectral classification technology is an important part of hyperspectral imaging technology. Its specific task is to use the spatial correlation of adjacent pixels and the characteristics of spectral information to classify the properties of substances corresponding to each pixel in the image. kind. This technology has great practical value in civilian and military fields, such as crop pest detection, geological exploration, environmental detection and battlefield camouflage target reconnaissa...

AI Technical Summary

Problems solved by technology

However, this method requires samples of unknown classes to adjust parameters, and the process of Weibull fitting is sensitive to the selection of abnormal points in training samples.

Therefore, in the actual use process, the robustness and generalization of this method are poor, and the performance of unknown target detection fluctuates greatly.

[0005] To sum up, the challenge of hyperspectral open set classification needs to be solved urgently; secondly, the current existing hyperspectral open set classification methods have problems of complex models, difficult parameter adjustment, and low accuracy. Further research is needed to achieve improvement or Propose a new open set classification method

Images