Feature Fusion Detection Network for Multi-scale Object Detection

Abstract

Small-scale-object usually occupies a small area in an image. The existing object detection methods are performing well for small-scale object detection in real scenes. This paper proposes, a Feature Fusion Detection Network (FFDN), for multi-scale objects detections. Firstly, the FFDN applies three feature maps in the improved VGG16 (Visual Geometry Group Network 16) with a proposed resolution expansion module to achieve the same resolution of the three feature maps. Then the FFDN fuses these three feature maps by a lightweight feature fusion method. Finally, it generates the feature pyramid by the fused feature map to achieve multi-scale object detection. In addition, we design a default box matching concession method which enables to train the real targets, and increases the number of positive samples. The experiments show that FFDN has better performance compared with the existing neural networks. It improves the recall rate for small-scale objects detection and the accuracy for large-scale object detection.