The　existing　deep　learning-based　algorithms　for　point　cloud　semantic　segmentation　has　achieved　an　unprecedented　precision　in　accuracy　and　robustness　under　some　classical　situation.　However,　the　uneven　density　and　large　scale　of　point　clouds　make　it　difficult　for　segmentation　networks　to　handle　these　hard　scenes.　Therefore,　effective　downsampling　and　uniformization　of　the　point　cloud　spatial　distribution　are　meaningful　research　directions.　Along　the　way,　this　paper　firstly　proposes　a　new　sectoral　cylindrical　partitioning　(SCP)　strategy　to　reorganized　large　point　clouds,　and　secondly　designs　a　SCP　based　semantic　segmentation　network　(SCP-SSNet),　which　achieves　high-precision　segmentation　of　large-scale　non-uniformly　distributed　point　clouds.　SCP-SSNet　consists　of　three　modules:　PC-Embedding,　SCP-Homogenization,　and　Transf.-Fusing.　Specially,　the　point　cloud　embedding　(PC-Embedding)　module　is　designed　to　provide　long-range　contextual　information　for　the　subsequent　segmentation　network.　The　point　cloud　feature　sectoral　cylindrical　partitioning-based　point　cloud　homogenization　(SCP-Homogenization)　module　is　proposed　to　effectively　improve　the　problem　of　uneven　feature　distribution.　The　Transformer　based　feature　fusing　(Transf.-Fusing)　module　is　introduced　to　fuse　contextual　association　and　position　encoding,　achieving　high-precision　semantic　segmentation　of　point　clouds.　The　experimental　results　on　some　public　datasets　demonstrate　that　SCP-SSNet　can　efficiently　predict　a　more　accurate　segmentation　results　of　point　cloud　and　outperform　the　state　of-the-art　methods.　©　2023　IEEE.