Detection of Prostate Capsule Based on Attentional Mechanism and Feature Fusion

The insufficient feature extraction ability of the network will lead to the decrease of the detection precision of the object detection. A new attention mechanism feature fusion network based on channels was proposed to solve that. We added Primary Component Analysis Squeeze-and-Excitation Networks (PCA-SENet) block to the backbone network VGG16 to increase the feature extraction ability, down-sampling and lateral connections were utilized to finish low-level and high-level feature fusion, new model called PCA-SENet Neighbour Single Shot Multibox (PNSSD) was proposed. The results showed that the proposed method had the mAP for prostate capsule image detection reached 82.02%, attention mechanism and feature fusion are helpful to improve the detection precision of object detection network.