Spatial features with spectral properties enhance the quality of satellite image while mapping complex land cover. These features are integrated with the proposed classification approach for improving classification results. The ultimate objective of this investigation is to provide high-level features to the convolutionalneural network (CNN) for mapping flooded regions (before and after) using remote sensing data. Here, boundary-based segmentation is done to recognize the dimensions and scales of objects. Modeling a fully trained Convolutional network is feasible for training a huge amount of data in remote sensing studies. Finetuned CNN is utilized with slight modification for attaining classified Landsat images. Classification outcomes and confusion matrix are manipulated using B-CNN are compared with classifiers like SVM, random forest (RF)to compute B-CNN efficiency