Aiming at the problem caused by wide scale change in target detection

a first-level target detection method based on multi-level characteristic pyramid network is proposed.Based on single shot multibox detector（SSD）algorithm model

based on the characteristics of the feature fusion extracted by backbone network

connect into the design of an alternate base features of u-shaped module and feature fusion module

from the characteristics of each U module output scale characteristics of polymerization

used to construct a multi-level pyramid network used in the final target detection.This method is simulated on MS COCO data set and PASCAL VOC data set

and the experimental results show that this algorithm is superior to other mainstream target detection methods.