The accurate alignment between visual features and semantic concepts is a challenging problem in zero-shot object detection. To address that, this paper proposes a new algorithm using attributes based category similarity. An unsupervised learning method is utilized to evaluate and adjust an attribute table, which helps to establish a better synergy between visual and semantic domains. Based on that, the similarity between categories are exploited to simultaneously detect and recognize novel concept instances with visual attributes. Experimental results on the MS-COCO dataset show that the proposed algorithm can achieve the highest mean average precision (15.34%), compared with the state-of-art algorithms.