We focus on wireless sensor networks (WSNs) that perform data collection with the objective of obtaining the whole dataset at the sink (as opposed to a function of the dataset). In this case, energy-efficient data collection requires the use of data aggregation. Whereas many data aggregation schemes have been investigated, they either compromise the fidelity of the recovered data or require complicated in-network compressions. In this paper, we propose a novel data aggregation scheme that exploits compressed sensing (CS) to achieve both recovery fidelity and energy efficiency in WSNs with arbitrary topology. We make use of diffusion wavelets to find a sparse basis that characterizes the spatial (and temporal) correlations well on arbitrary WSNs, which enables straightforward CS-based data aggregation as well as high-fidelity data recovery at the sink. Based on this scheme, we investigate the minimum-energy compressed data aggregation problem. We first prove its NP-completeness, and then propose a mixed integer programming formulation along with a greedy heuristic to solve it. We evaluate our scheme by extensive simulations on both real datasets and synthetic datasets. We demonstrate that our compressed data aggregation scheme is capable of delivering data to the sink with high fidelity while achieving significant energy saving.