Improving land cover classification accuracy of Sentinel-2 images: a systematic review of articles between 2015 and 2021

The purpose of this work was to undertake a systematic assessment of the approaches used to improve the accuracy of land cover maps from Sentinel-2 satellite images when utilizing supervised cell–based classification, as reported in articles published between 2015 and 2021. The preferred reporting items for systematic reviews and meta-analyses (PRISMA) technique were utilized for this purpose. This involved searching for peer-reviewed articles relating to the review’s topic, which returned 551 articles. This was followed by sorting and filtering and, last, the exclusion and inclusion of articles based on specific criteria. This process resulted in nine articles, and their contents were examined from three perspectives: data preprocessing, classification model inputs, and classification techniques. Regardless of the differences like the targeted land cover classes, the number of training samples, and the classification model inputs, the results highlighted the importance of several factors in improving classification accuracy, including spatial resolution integration, data derivation (such as indices), and the selection of atmospheric correction and classification algorithms. All of these characteristics, however, are tied to the nature of the study area; that is, what is good for one area may not be acceptable for another. The study ends by summarizing the key conclusions and offering a workable strategy, as a general frame of reference, for classifying Sentinel-2 images in which the characteristics of the study region are carefully considered to achieve higher classification accuracy. This is based on the results and other pertinent references.