摘要
针对水下成像的特殊性以及成像环境的复杂性,构造了基于区域矩的仿射变换不变量,以克服水下不确定因素给目标识别带来的困难。此外针对传统的BP神经网络存在收敛速度慢以及容易陷入局部极小值的缺点,引入粒子群算法对神经网络的学习训练进行优化。为了验证所提方法的有效性,对四类水下目标进行了特征提取以及神经网络识别实验。结果表明改进后的神经网络收敛速度快,并且获得了较高的识别准确率。
Effective feature extraction is one of the key elements to underwater target recognition. The affine invariants is constructed based on region moments in order to eliminate the negative effects, which are brought by the particularity and complexity of imaging environment. Aiming at the drawbacks of traditional BP neural network, such as converging slowly and tending to get into the local minimizer, the particle swarm algorithm is introduced into the training of neural networks. The affine invariant features of four different objects are extracted and selected as the input of the trained neural network. The experimental results show that the extracted features and the improved neural network can result in fast convergence rate and high accuracy.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2008年第1期171-175,共5页
Systems Engineering and Electronics
关键词
目标识别
特征提取
神经网络
粒子群算法
target recognition
feature extraction
neural network
particle swarm algorithm
