We present a data compression and dimensionality reduction scheme for datafusion and aggregation applications to prevent data congestion and reduceenergy consumption at network connecting points such as cluster heads andgateways. Our in-network approach can be easily tuned to analyze the datatemporal or spatial correlation using an unsupervised neural network scheme,namely the autoencoders. In particular, our algorithm extracts intrinsic datafeatures from previously collected historical samples to transform the raw datainto a low dimensional representation. Moreover, the proposed frameworkprovides an error bound guarantee mechanism. We evaluate the proposed solutionusing real-world data sets and compare it with traditional methods for temporaland spatial data compression. The experimental validation reveals that ourapproach outperforms several existing wireless sensor network's datacompression methods in terms of compression efficiency and signalreconstruction.
展开▼
