In　the　field　of　computer　vision,　person　re-identification　is　a　popular　research　topic,　one　of　its　difficulties　is　how　we　can　enhance　the　recognition　effect　of　the　algorithm　in　scenes　containing　varieties　of　similar　pedestrian　images.　Traditional　metric　learning　algorithms　are　not　effective　in　recognizing　pedestrian　samples　with　high　similarity.　We　propose　an　adaptive　weighted　cascaded　metric　learning　mechanism　in　this　paper.　The　algorithm　performs　multiple　metric　learning　in　a　cascade　manner　to　enhance　the　differentiation　between　features.　The　results　of　each　level　of　metric　learning　are　weighted　and　fused　using　an　adaptive　weighting　mechanism,　which　ensures　that　some　important　pedestrian　identity　information　contained　in　the　original　or　shallow　features　will　not　be　lost,　and　also　avoids　the　overfitting　phenomenon　to　a　certain　extent.　Our　method　achieves　good　recognition　results　in　VIPeR,　GRID　and　CUHK01.　©　2024　IEEE.