Detection of Harran plain soil salinity with deep learning

Soil salinity occurs in the arid and semi-arid climatic regions by the dissolution of salts formed by the combination of anions and cations in the structure of the soil, by mixing with groundwater. These salts formed dissolve with high ground water to the surface of the soil and accumulate on the surface of the soil as a result of evaporation of the ground water. This affects plant growth negatively and decreases the yield. Harran Plain, one of Turkey's largest agricultural plain, was aimed to identify salinity problem with remote sensing techniques for ensuring sustainability of agricultural land management. NDSI, SI, SII and plant index NDVI, which are the most used salinity indices in the literature, were used for the determination of salinity in Harran Plain. Convolutional Neural Networks, a deep learning method, has been added to the image as a separate spectral band, in addition to the 5 bands of the image, in the classification process made with U-NET architecture. It has been the SII (93.78%) salinity index combination that gave the best classification accuracy in 300 iterations.