We address energy-efficient connected target coverage in wireless sensor networks (WSNs), seeking the smallest active subset of sensors that covers all targets and remains connected to the sink. We propose a Redundancy-Aware Island Genetic Algorithm (RA-IGA). It combines a redundancy-aware mutation with a lightweight deterministic coverage-repair step that aims to activate as few additional sensors as needed to restore feasibility. It also uses a heterogeneous three-island model with periodic elite migration to maintain diversity and improve final quality under the same budget. RA-IGA is benchmarked against the improved genetic algorithm (IGA) and the modified marine predators algorithm (MMPA) across grid and random deployments while varying network size, target count, and field dimensions (up to N = 400 , K = 200, L = 500 ). RA-IGA consistently selects the fewest active sensors, reducing the active set by 5%–24% vs. IGA and 48%–56% vs. MMPA, with tighter dispersion over 20 seeds. A Friedman test with Nemenyi post-hoc confirms p< 0.001 . Because fewer actives generally reduce per-round energy under matched packet and model assumptions, these results suggest longer network lifetime. Ablations indicate that redundancy-aware mutation and repair drive sparsity while preserving feasibility. They also show that the heterogeneous island model helps escape single-population local optima, yielding better final solutions.