analysis of haze detection and haze removal algorithms on urban and rural land cover on sentinel 2 imagery

Abstract

The development of increasingly advanced technology has a positive impact, especially in mapping using satellite imagery. The existence of clouds and fog that is always there makes it difficult to get a clean image (clear scenes). Haze will be very easy to form in vegetated and non-vegetated areas even though they have different characteristics and formation processes. Haze Removal is a method that has been commonly used in optical images for fog removal so as to facilitate identification on the image. Haze detection is an important step in haze removal by determining the range of haze values using several algorithms such as Haze Optimized Transform (HOT) and Supervised Haze Transform (SHT). This study aims to analyze the use of the best fog detection method on Sentinel-2 imagery in rural (Semarang Regency) and urban (Semarang City) areas which have different land cover characteristics and to test the Dark Object Subtraction (DOS) method for land cover classification in imagery Sentinel-2.

The results of this study are that in urban areas the accuracy value obtained by both the HOT and SHT methods has the same value in the haze class, which is 80.65%, while for the clean class the SHT method is superior to the HOT method with a value of 85.06% and 79. ,87%. In rural areas, the accuracy value in the haze class, the HOT method, is superior to the SHT method with a value of 82.31% and 80.95%, while for the clean class the SHT method is superior to the HOT method with a value of 94.77% and 90.85%.  Haze removal processing using the DOS method on Sentinel-2 for urban and rural areas can increase the accuracy of interpretation of land cover classification in supervised classification with the Maximum Likelihood method. In the urban area (Semarang City) before the supervised classification was carried out, there were quite a lot of misclassifications in the class of built-up land and water bodies in foggy areas. In general, the increase in accuracy before processing was 82.67% to 94.67% for both HOT-DOS and SHT-DOS methods. In rural areas (Semarang Regency) before the supervised classification was carried out there was a misclassification of small vegetation classes in foggy areas. In general, the increase in the accuracy value before processing was 85.53% to 96.05% for both HOT-DOS and SHT-DOS methods