Application of Feature Pyramid Network and Feature Fusion Single Shot Multibox Detector for Real-Time Prostate Capsule Detection

In the process of feature propagation, low-level convolution layers forward propagation network lack semantic information, and information loss occurs when fine-grained is transferred to higher-level convolution; therefore, multi-stage fusion networks are needed solve interaction between high-level layers. Based on a two-way feedback mechanism, we created new object detection called Feature Pyramid Network (FPN)-based Fusion Single Shot Multibox Detector (FFSSD). A bottom-up top-down architecture with lateral connections enhances detector’s ability extract features, then multi-scale maps utilized generate pyramid network. The results show that proposed mAP for prostate capsule image reaches 83.58%, providing real-time ability. context mechanism can transfer convolution, resulting after contains richer location information.