High-resolution remote sensing image classification based on active learning

With the development of remote sensing technology, high-resolution remote sensing images have become the mainstream. However, how to accurately classify the ground objects in high-resolution remote sensing images has been a long-term research direction. In traditional image recognition technologies, whether it is SVM or CNN, there are requirements for the number of images in the training set. In recent years, neural networks have become the mainstream of image classification. However, remote sensing images are often affected by weather and other factors, leading to a limited number of available images for a specific area. Active learning can achieve the expected classification performance with a limited number of training samples. First, the images are segmented into superpixels, and supervised learning is used to determine the categories of superpixels. Active learning is then applied to process the superpixels with uncertain classifications. Classification is achieved by calculating the similarity between superpixels. Compared with traditional classification methods that do not use active learning strategies, the overall classification accuracy and Kappa index are improved when using active learning strategies.