Automatic Detection of Over-head Water Tanks from Satellite Images Using Faster-RCNN

Pattern recognition is pertinent field for detection of urban/manmade features from satellite imagery. Neural networks are best used in object detection for recognising patterns in imageries. Convolutional Neural Networks (CNNs) become way in solving object detection task based on deep learning concepts. This article demonstrates the usability of CNNs for detecting and mapping of small objects from the urban scenes. Identification and mapping of overhead water tanks from satellite imagery is a very important task especially during reconnaissance situation raised due to water contamination. Faster Region based CNN (Faster RCNN) has been used to detect and map the overhead water tanks in the urban scene from satellite imagery. The results from this study indicate that Faster RCNN gives affirmative accuracy towards detection of small objects from satellite imageries.