Semisupervised classification of remote sensing images using efficient neighborhood learning method
Engineering Applications of Artificial Intelligence
Efficiency of a classification model can be enhanced with more number of accurately labeled samples, which are difficult to obtain in remote sensing imagery. To mitigate this issue, semisupervised learning methodologies can be suitably used that exploit both unlabeled and labeled samples in the learning process and lead to the performance improvement of a classification model. With this reasoning, the present article proposes a self-learning semisupervised classification model using a neural network (NN) as the base classifier. The model uses an efficient neighborhood information learning method for the NN to overcome the demerits of existing conventional approaches. It is very crucial and challenging to find the true and the most relevant neighborhood information of unlabeled samples. Using two different approaches, we propose the generation of similarity matrixes for extracting neighborhood information that eventually improve the learning process of NN. The first method considers mutual neighborhood information and the second method uses the class-map of unlabeled samples. Class labels of the unlabeled samples are predicted by a classifier, i.e., trained with the available labeled samples. Finally, the collaborative neighborhood information is derived from these two matrixes and used for the development of the proposed semisupervised classification model. Experimental demonstration on three multispectral and one hyperspectral remote sensing images justified the superiority of the proposed model compared to the existing state-of-the-art methods. For comparative analysis, various performance measures, such as overall accuracy, kappa coefficient, precision, recall, dispersion score, β, and Davies–Bouldin (DB) scores are used.
Kothari, Neeta S. and Meher, Saroj K., "Semisupervised classification of remote sensing images using efficient neighborhood learning method" (2020). Journal Articles. 334.