eduzhai > Applied Sciences > Engineering >

Remote Sensing Image Scene Classification with Deep Neural Networks in JPEG 2000 Compressed Domain

  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: To reduce the storage requirements, remote sensing (RS) images are usuallystored in compressed format. Existing scene classification approaches usingdeep neural networks (DNNs) require to fully decompress the images, which is acomputationally demanding task in operational applications. To address thisissue, in this paper we propose a novel approach to achieve sceneclassification in JPEG 2000 compressed RS images. The proposed approachconsists of two main steps: i) approximation of the finer resolution sub-bandsof reversible biorthogonal wavelet filters used in JPEG 2000; and ii)characterization of the high-level semantic content of approximated waveletsub-bands and scene classification based on the learnt descriptors. This isachieved by taking codestreams associated with the coarsest resolution waveletsub-band as input to approximate finer resolution sub-bands using a number oftransposed convolutional layers. Then, a series of convolutional layers modelsthe high-level semantic content of the approximated wavelet sub-band. Thus, theproposed approach models the multiresolution paradigm given in the JPEG 2000compression algorithm in an end-to-end trainable unified neural network. In theclassification stage, the proposed approach takes only the coarsest resolutionwavelet sub-bands as input, thereby reducing the time required to applydecoding. Experimental results performed on two benchmark aerial image archivesdemonstrate that the proposed approach significantly reduces the computationaltime with similar classification accuracies when compared to traditional RSscene classification approaches (which requires full image decompression).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×