采用MSBN多智能体协同推理的智能农业车辆环境识别

Environment recognition of intelligent agricultural vehicles based on MSBN and multi-agent coordinative inference

为了解决智能农业车辆对所处复杂农田环境的识别信度定量分析困难的问题,提出了基于多连片贝叶斯网(MSBN)多智能体协同推理的目标识别算法.该方法把多智能体图像采集系统的局部信息表征在MSBN模型中,在观测不完备条件下,虽然单个智能体仅拥有目标的局部观测信息,但利用重叠子域信息的更新可以进行子网间消息的传播.利用MSBN局部推理和子网间信度通信的全局推理对多源信息进行融合,以提高识别性能.实验结果表明,与传统神经网络或BN方法相比,基于MSBN目标识别算法有效地对多源信息进行了补充,可以提高农业车辆在复杂环境进行识别的准确性.

In order to solve the problem existing in the agricultural environment recognition of intelligent vehicles , due to the difficulty of conducting quantitative analysis of the reliability of such recognition , a target recognition al-gorithm for multi-agent cooperative inference based on the multiply sectioned Bayesian network （ MSBN） has been proposed.This method characterizes local information of the multi-agent image acquiring system with MSBN model . In the circumstance of incomplete observations , although each single agent may only capture some local observation information from the target , the message propagation among subnets can be achieved by information update in the o -verlapping sub-domains.By combining the local inference and global inference of reliability communication between subnets in MSBN , the multi-source information was merged to enhance recognition performance .By comparing the traditional neural network and BN method , experimental results illustrate that , the target recognition algorithm based on MSBN can effectively supplement multi-source information , and thus, can improve the recognition accura-cy of agricultural vehicles in the complicated environment .