Recursive classification of satellite imaging time-series: An application to water mapping, land cover classification and deforestation detection

A wide variety of applications of fundamental importance for security, environmental protection and urban development need access to accurate land cover monitoring and water mapping, for which the analysis of optical remote sensing imagery is key. Classification of time-series images, particularly with recursive methods, is of increasing interest in the current literature. Nevertheless, existing recursive approaches typically require large amounts of training data. This paper introduces a recursive classification framework that improves the decision-making process in multitemporal and multispectral land cover classification algorithms while requiring low computational cost and minimal supervision. The proposed approach allows the conversion of an instantaneous classifier into a recursive Bayesian classifier by using a probabilistic framework that is robust to non-informative image variations. Three experiments are conducted using Sentinel-2 data. The first one consists in the water mapping of an embankment dam in California (United States), the second one is a land cover classification experiment of the Charles river area in Boston (United States) and the last experiment addresses deforestation detection in the Amazon rainforest (Brazil). A classifier based on the Gaussian mixture model (GMM), a logistic regression (LR) classifier, and a spectral index classifier (SIC) are compared to their recursive counterparts. SICs are introduced to convert the NDWI, MNDWI and NDVI spectral indices into predictive probabilities. Two state-of-the-art deep learning-based models are also used as a benchmark for the water mapping experiment. Results show that the proposed method significantly increases the robustness of existing instantaneous classifiers in multitemporal settings. Our method also improves the performance of deep learning-based classifiers without the need for additional training data.