Deep Learning Approaches for Water-body Detection from Satellite Imagery

Abstract

Agriculture department of Government of India provides funds for digging wells and making farm ponds for improving farmers livelihood. The proper execution of the policies not been worked out due to the manual process. For the automation of the process, we are using the object detection algorithm Faster Region-based Convolutional Neural Network (F-RCNN) to detect the location of the wells and Farm ponds. The faster RCNN method introduced region of interest to improve the model speed. We have tried to address water body detection on satellite imagery with an AI approach to cross verify the utilization of funds is the challenge. The traditional object detection algorithms are low in accuracy, also uses coco image data set. Detecting water body has not been addressed by anybody. We have extended the algorithms for water bodies detection for the complete geographic region of agricultural land. Improved Faster R-CNN algorithm used in the detection of wells and farm ponds. More than 1000 Satellite imagery are used to train the model to detect the water body. The experimental results demonstrated promising accuracy on well and farm ponds detection. The accuracy of the faster RCNN algorithm is 90%, we have tested the accuracy of the results by GIS locations, and ground-truthing is achieved by our designed Android App.This app gives the latitude and longitude of well or farm ponds entered by the person in Jalgaon. Keywords : About four key words or phrases in alphabetical order, separated by commas.