Artificial neural networks for cloud masking of Sentinel-2 ocean images with noise and sunglint

Removal of cloud interference is a crucial step for the exploitation of the spectral information stored in optical satellite images. Several cloud masking approaches have been developed through time, based on direct interpretation of the spectral and temporal properties of clouds through thresholds. The problem has also been tackled by machine learning methods with artificial neural networks being among the most recent ones. Detection of bright non-cloud objects is one of the most difficult tasks in cloud masking applications since spectral information alone often proves inadequate for their separation from clouds. Scientific attention has recently been redrawn on self-organizing maps (SOMs) because of their unique ability to preserve topologic relations, added to the advantage of faster training time and more interpretative behavior compared to other types of artificial neural networks. This study evaluated a SOM for cloud masking Sentinel-2 images and proposed a fine-tuning methodology to separate clouds from bright land areas. The fine-tuning process which is based on the output of the non-fine-tuned network, at first directly locates the neurons that correspond to the misclassified pixels. Then, the incorrect labels of the neurons are altered without applying further training. The fine-tuning method follows a general procedure, thus its applicability is broad and not confined only in the field of cloud-masking. The network was trained on the largest publicly available spectral database for Sentinel-2 cloud masking applications and was tested on a truly independent database of Sentinel-2 cloud masks. It was evaluated both qualitatively and quantitatively with the interpretation of its behavior through multiple visualization techniques being a main part of the evaluation. It was shown that the fine-tuned SOM successfully recognized the bright non-cloud areas and outperformed the state-of-the-art algorithms: Sen2Cor and Fmask, as well as the version that was not fine-tuned.

Download Full-text

Soil salinity mapping utilizing sentinel-2 and neural networks

Indian Journal of Agricultural Research ◽

10.18805/ijare.a-316 ◽

2018 ◽

Author(s):

R.S. Morgan ◽

M. Abd El-Hady ◽

I.S. Rahim

Keyword(s):

Neural Networks ◽

Artificial Neural Networks ◽

Soil Salinity ◽

Vegetation Index ◽

High Spectral Resolution ◽

Coefficient Of Determination ◽

Infrared Band ◽

Artificial Neural ◽

Reflectance Data ◽

Sentinel 2

Soil salinity is the most important soil property that affects the agriculture productivity. Periodical monitoring of its status is considered a crucial factor in the selection of appropriate agricultural practices to attain a sustainable production. The availability of remote sensing data processed by a somewhat novel method such as artificial neural networks (ANN) offer a potential solution that could easily and affordably replace the in-site monitoring methods. The aim of this work is to use high spectral resolution Sentinel-2 (S2) data for soil salinity prediction utilizing neural networks. The study evaluated three approaches in processing the S2 data for inclusion in the artificial neural network for soil salinity prediction. These approaches included S2 spectral reflectance data, spectral indices and principal component analysis (PCA) of the S2 data. The results revealed that a combination of these approaches including the reflectance data of band 11(shortwave infrared band) of S2, the normalized differential vegetation index (NDVI) and the second PCA (dominated by the near infrared band) gave the best performance when used as input when designing the artificial neural networks to predict the soil salinity. The overall accuracy of this approach has a coefficient of determination (R2) of 0.94 between the actual and predicted soil salinity.

Download Full-text