The　proliferation　of　multimodal　misinformation　on　social　media　has　become　a　critical　concern.　Although　detection　methods　have　advanced,　feature　representation　and　cross-modal　semantic　alignment　challenges　continue　to　hinder　the　effective　use　of　multimodal　data.　Therefore,　this　paper　proposes　an　IBWO-CASC　detection　model　that　integrates　an　improved　Beluga　Whale　Optimization　algorithm　with　cross-modal　attention　feature　fusion.　Firstly,　the　Beluga　Whale　Optimization　algorithm　is　enhanced　by　combining　adaptive　search　mechanisms　with　batch　parallel　strategies　in　the　feature　space.　Secondly,　a　feature　alignment　method　is　designed　based　on　supervised　contrastive　learning　to　establish　semantic　consistency.　Then,　the　model　incorporates　a　Cross-modal　Attention　Promotion　mechanism　and　global-local　interaction　learning　pattern.　Finally,　a　multi-task　learning　framework　is　built　based　on　classification　and　contrastive　objectives.　The　empirical　analysis　shows　that　the　proposed　IBWO-CASC　model　achieves　a　detection　accuracy　of　97.41%　on　our　self-constructed　multimodal　misinformation　dataset.　Compared　with　the　average　accuracy　of　the　existing　six　baseline　models,　the　accuracy　of　this　model　is　improved　by　4.09%.　Additionally,　it　demonstrates　enhanced　robustness　in　handling　complex　multimodal　scenarios.