Network training method based on Parallel Vision theory for object detection in battlefield environment

Abstract

Visual perception and understanding in complex battlefield environments have few data samples, which cannot meet the need of deep network model training in variable and complex environments. In this paper, based on parallel vision theory, deep learning network training for object detection is conducted using artificial virtual battlefield environment. Virtual battlefield environment data are collected for experimental training and validation analysis, and further training tests are conducted in real scene data. The experiments verified that the deep network model training with virtual-real interaction can effectively solve the data deficient problem of object detection in the battlefield.