Performance Evaluation of Jaccard-Dice Coefficient on Building Segmentation from High Resolution Satellite Images

In remote sensing applications, segmentation of input satellite images according to semantic information and estimating the category each pixel from a given set tags are great importance for automatic tracking task. It is important in situations such as building detection high resolution images, city planning, environmental preparation, disaster management. Buildings metropolitan areas crowded messy, so high-resolution satellites need be automated detect buildings. Segmentation with deep learning technology has been widely considered area research. The Fully Convolutional Network (FCN) model, popular used based on pixel-level images. U-Net model developed biomedical image modified our study, its performances during training, accuracy testing were compared by using customized loss functions Dice Coefficient Jaccard Index measurements. score was obtained 84% lost 70%. addition, increased 87% Batch Normalization (BN) method instead Dropout model.