Object　detection　methods　have　demonstrated　remarkable　performance　on　conventional　datasets.　However,　their　susceptibility　to　degraded　visibility　in　hazy　weather　remains　a　significant　limitation,　e.g.,　in　maritime　object　detection　applications.　The　importance　of　dehazing　preprocessing　becomes　evident　when　considering　the　detrimental　impact　of　haze　on　object　detection　accuracy.　In　this　paper,　we　introduce　a　novel　framework　called　TD-YOLO,　which　is　short　of　utilizing　Taylor-attention　for　dehazing　to　enhance　the　detection　performance　of　YOLO,　which　incorporates　dehazing　processing　to　enhance　the　visibility　of　hazy　images　before　the　detection　task.　The　dehazing　network　we　employ　is　based　on　the　Transformer　architecture,　addressing　the　limitations　of　CNNs　in　capturing　long-range　dependencies　and　non-local　self-similarity.　However,　transformers　are　computationally　intensive,　particularly　on　high-resolution,　large-scale　images.　Drawing　inspiration　from　MB-Taylorformer,　Taylor　expansion　is　utilized　to　reduce　the　complexity　of　softmax　in　the　Transformer.　Additionally,　we　propose　a　novel　dataset　for　maritime　detection　in　hazy　conditions,　named　MDHD,　comprising　pairs　of　hazy　images　and　their　corresponding　haze-free　counterparts.　Furthermore,　we　meticulously　annotated　the　images　to　identify　ship　targets　across　9　distinct　categories:　Cargo　Ships,　Fishing　Vessels,　Sailing　Vessels,　Pleasure　Craft,　Speedboats,　Guard　Vessels,　Other　Ships,　Buoy,　and　Helicopters.　This　dataset　serves　as　a　foundation　for　research　in　dehazing　and　detection　of　maritime　scenes.　Experimental　results　demonstrate　a　significant　improvement　in　detection　accuracy　under　hazy　weather　conditions　using　our　proposed　framework.