摘要
提出了一种基于BP神经网络和D-S证据推理的多传感器数据融合图像目标识别算法,利用数据融合的思想对来自目标的多幅图像进行空间域融合处理。首先提取图像的Hu不变矩作为待识别目标图像的特征,尔后针对DS证据理论基本概率指派函数构造困难的问题,用BP神经网络对目标的初步识别结果构造基本概率指派函数,最后用D empster组合规则对BP网络的初步识别结果进行决策级数据融合,完成了三维飞机图像目标的识别仿真。仿真结果表明了融合识别方法的有效性和鲁棒性,识别率达到100%。
：This paper presents an image recognition algorithm based on BP Neural Network （BPNN） and Dempster - Shafer Evidence Theory. Firstly, modified Hu invariant moments is used as the feature of the image, and BPNN is applied to identify the target. Secondly, basic belief assignment function is constructed through the output of the BPNN, and then Dempster fusion rule is adopted to finish the decision data fusion. Lastly, recognition of 3 - D airplane image are completed. The simulation results show that this fusion method is effective and robust, and the recognition rate is 100%.
出处
《计算机仿真》
CSCD
2005年第11期184-186,197,共4页
Computer Simulation
基金
国家自然科学基金项目(60372085)
关键词
证据理论
多传感器数据融合
神经网络
图像识别
Evidence reasoning
Multi -sensor data fusion
Neural network
Image recognition
