Detection is one of the important tasks in image processing. How to improve the accuracy of object location is a difficult problem in the process of detection. On the basis of the SSD algorithm, this paper proposes a rough location and object class combined with selective search algorithm. Then, the paper searches the area of object by using an improved selective search algorithm to generate an object candidate frame. Finally, the paper adopts a boundary judgment method to obtain the accurate position of object and complete the detection from coarse to fine. In this paper, a lot of experiments are carried out based on 22531 images of PASCAL VOC data sets. The results show that the algorithm has a location error of 7.6 centers and a mean coverage rate of 90.8%. Compared with the other algorithms, this algorithm has lower center location error and higher coverage rate, improving the location accuracy of object detection by 20%~30%.