Object Detection and Segmentation Method for Multi-category Armored Targets Based on CNN

Abstract

A target detection method based on fisher discriminative Mask-RCNN is proposed to improve the ability of the object detection and segmentation of multi-category armored targets. The fisher discriminative layer is added to the classification branch and trained by imposing the Fisher discrimination criterion on loss function. Target segmentation is achieved by adding a mask branch network to generate binary target masks. Experimental results indicate that the proposed method can increase the mean Average Precision (mAP) by 1.3% on our dataset of four similar armored targets and achieve good segmentation results, which can better meet the need of military reconnaissance and other tasks.