GPU Accelerated Automated Feature Extraction From Satellite Images

The availability of large volumes of remote sensing data insists on higher degree of automation in featureextraction, making it a need of thehour. Fusingdata from multiple sources, such as panchromatic,hyperspectraland LiDAR sensors, enhances the probability of identifying and extracting features such asbuildings, vegetation or bodies of water by using a combination of spectral and elevation characteristics.Utilizing theaforementioned featuresin remote sensing is impracticable in the absence ofautomation.Whileefforts are underway to reduce human intervention in data processing, this attempt alone may notsuffice. Thehuge quantum of data that needs to be processed entailsaccelerated processing to be enabled.GPUs, which were originally designed to provide efficient visualization,arebeing massively employed forcomputation intensive parallel processing environments. Image processing in general and hence automatedfeatureextraction, is highly computation intensive, where performance improvements have a direct impacton societal needs. In this context, an algorithm has been formulated for automated feature extraction froma panchromatic or multispectral image based on image processing techniques.Two Laplacian of Guassian(LoG)masks were applied on the image individually followed by detection of zero crossing points andextracting the pixels based on their standard deviationwiththe surrounding pixels. The two extractedimages with different LoG masks were combined together which resulted in an image withthe extractedfeatures and edges.Finally the user is at liberty to apply the image smoothing step depending on the noisecontent in the extracted image.The image ispassed through a hybrid median filter toremove the salt andpepper noise from the image.This paper discusses theaforesaidalgorithmforautomated featureextraction, necessity of deployment of GPUs for thesame;system-level challenges and quantifies thebenefits of integrating GPUs in such environment. The results demonstrate that substantial enhancement inperformance margin can be achieved with thebestutilization of GPU resources and an efficientparallelization strategy.Performance results in comparison with the conventional computing scenario haveprovided a speedup of20x, on realization of this parallelizing strategy