The　generation　of　image　texture　details　is　a　crucial　aspect　in　the　assessment　of　the　efficacy　of　image　super-resolution　reconstruction.　In　order　to　enhance　the　super-resolution　effect　of　real-world　degraded　images,　this　paper　proposes　a　method　for　image　super-resolution　reconstruction　with　an　attention　mechanism　in　generative　adversarial　networks.　Firstly,　a　multi-angle　degradation　modeling　process　is　introduced　for　real-world　image　degradation　scenarios　to　better　simulate　complex　real-world　degradation;　secondly,　the　structure　of　the　Residual-in-Residual　Dense　Block　(RRDB)　commonly　used　in　super-resolution　reconstruction　generative　adversarial　networks　was　optimized,　and　combined　with　the　dynamic　attention　mechanism,　a　Residual-in-Residual　Dense　Block　with　Dynamic　Channel　and　Spatial　Attention　Mechanism　(RRDB-DCSA)　is　proposed,　and　this　module　is　used　to　construct　a　generator　for　Generative　adversarial　networks　(GAN);　and　lastly,　locally　discriminative　learning　loss　is　introduced　to　reduce　the　artifacts　caused　by　the　GAN　network　with　the　aim　of　learning　features　associated　with　image　degradation.　The　experimental　results　demonstrate　that　the　proposed　algorithm　is　capable　of　reconstructing　the　detailed　components　of　an　image　while　simultaneously　suppressing　the　artifacts　present　in　scenes　featuring　people,　animals,　and　landscapes.　In　terms　of　subjective　assessment,　the　algorithm　exhibits　superior　performance　in　this　regard,　with　an　improvement　of　nearly　17　in　PSNR　and　SSIM,　respectively.　Additionally,　the　algorithm　exhibits　a　reduction　of　nearly　15　in　LPIPS　and　NIQE　when　compared　with　traditional　and　classical　algorithms　of　the　GAN　network.　©　2024　Copyright　held　by　the　owner/author(s).