A New Semisupervised-Entropy Framework of Hyperspectral Image Classification Based on Random Forest

Table 2

The process of semisupervised classification of random forest based on weighted entropy algorithm.

---

Algorithm: Semi-supervised Classification of Random Forest Based on Weighted Entropy

---

Input: Hyper-spectral remote sensing image data; training sample set; the category set corresponding to the training sample.

Output: The results of classification images and confusion matrix;

method:

  Random forest with probability output is used to determine the expected value of the category of ground objects with the largest number of votes;

  Judging the type of the samples according to the results of output;

  Judging the accuracy of the output results according to the classified ground object classification data;

  Then the weighted entropy algorithm is used to give the category with the highest weighted return weight of the class predicted by the model;

  The ground objects which account for 5% or 10% of the total sample increase are selected and added to the training sample to form a new training sample, and then the prediction classification is carried out again;

  The unlabeled label pixel in the input hyper-spectral image is converted into the tag pixel according to the uncertainty evaluation value;

  A new tag is added to the original training set and a new training set is constructed;

  Run iteratively until termination requirements are met or unlabeled training samples run out.

  Using the remaining samples to test the performance of the classifier, that is, to evaluate the accuracy of the classifier;

  In order to test the classification performance and the universality of the classifier, the hyper-spectral remote sensing images of Hyperion and the hyper-spectral remote sensing images of AVIRIS type are classified, and the accuracy is evaluated.

---