K-means clustering-based data compression scheme for wireless imaging sensor networks

—Image-based embedded wireless sensor networks (WSNs) can be a useful tool in various environmental monitoring applications to unobtrusively observe biological phenomena. Our prior deployments of an embedded wireless imaging system at the James Reserve have already shown its feasibility and usefulness. However, we argue that data compression schemes employed in prior systems can be improved to provide higher image transfer rates per node, or lower the energy costs of wireless communication. In this paper, we develop an image compression scheme using K-means clustering on low-power embedded devices for image-based WSNs. Specifically, we use the similarity of pixel colors to group pixels and compress the original image. Using 100 000 images collected from our pilot deployments at the James Reserve, we study the applicability and impact of the proposed K-means clustering-based compression algorithm. Our results suggest that the cost of running K-means learning on a wireless sensor node may outweigh the benefit of data compression, but offloading the learning step and only performing the compression can provide significant energy gains. Specifically, our evaluations with real-world data sets show that our proposed scheme reduces power usage by ∼49%, when sending image updates from a bird nest periodically every 15 min.