Object detection from images based on MFF-RPN and multi-scale CNN

Abstract

In this paper, an object detection model based on MFF-RPN and Multi-scale CNN is proposed. Firstly, the region proposal network based on multi-level feature fusion (MFF-RPN) is presented to extract the candidate proposals. Secondly, a convolutional neural network (CNN) with different scale convolution kernels is conducted to extract features adaptively. Finally, multi-task loss is employed to establish a complex mapping between image object features and object detection mode. The experimental results prove that the proposed algorithm gets better classification performance and higher detection accuracy.