Multi-objective Identification of UAV Based on Deep Residual Network

3rd International Conference on Automation, Control and Robotics Engineering

工业技术;计算机科学;无线电电子学

Qi, Wang Jia^1 ; Yang, Dai Ji^1 ; You, Zhai Jin^1 ; Jin, Ying^1

School of Information Engineering, Nanchang Hangkong University, Nanchang

330063, China^1

关键词: Cascade regions;    Deep convolutional neural networks;    Identification accuracy;    Learning Theory;    Network modeling;    Percentage points;    Region proposals;    Time complexity;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/428/1/012061/pdf
DOI  :  10.1088/1757-899X/428/1/012061

来源: IOP

【 摘 要 】

Beacuse of the classical RPN (Region Proposal Net) exists the defect of large computation and high time complexity when extracting targets candidate region, a search mode called CRPN (Cascade Region Proposal Network) mode was proposed to ameliorate it in this paper. In order to suppress the degradation phenomenon in deep convolutional neural network training, the residual learning theory was introduced, a novel Mu-ResNet (multi-strapdown deep residual network) was proposed. Combined the Mu-ResNet with CPRN, a network model for multi-target identification of UAV was designed and tested. Compared with the network model that combines ResNet with RPN, the identification accuracy was increased nearly 2 percentage points.