@INPROCEEDINGS{Publ2017-977, author = {Miguel Heredia Conde and Otmar Loffeld}, title = {A genetic algorithm for compressive sensing sparse recovery}, booktitle = {IEEE International Symposium on Signal Processing and Information Technology (ISSPIT)}, year = {2017}, pages = {106-111}, month = {Dezember}, abstract = {The advent of compressive sensing (CS) theory opened the possibility of linking the sensing effort, that is, the volume of data being produced by the sensor, to the amount of information it conveys, rather than to the desired sensor bandwidth, as traditional sampling theory suggests. Consequently, in the typical CS scenario, one ends up with a set of few measurements and the challenge is to recover a signal whose dimensionality is much higher than the number of measurements, typically under the assumption of being sparse. One faces, therefore, a constrained l0 minimization problem. Despite being ubiquitous in nature, finding the solution with lowest lo norm is known to be NP-hard. In this work, we propose mimicking the nature to approach a solution. More specifically, we design a genetic algorithm (GA) that, despite being based on the rules of evolution of biological systems, is fully tailored to our specific problem. Adopting the terminology from genetics, our chromosomes are representations of different support configurations, with an associated restricted-support temporal solution. The fitness of each chromosome is measured in terms of reprojection error of the associated solution. We deal with the sparsity requirement by means of a generalized crossover strategy based on support set overlap, rather than explicitly adding an l0 or l1 regularizer to the fitness function. We show that the proposed algorithm outperforms the generic multiobjective GA NSGA-II for solving the CS constrained l0 minimization in terms of l2 reconstruction error, at no cost in execution time.}, }