一種具有漸進(jìn)學(xué)習(xí)能力的融合方法

運(yùn)行在動態(tài)與未知環(huán)境下的多傳感器系統(tǒng)往往會面臨環(huán)境與自身結(jié)構(gòu)的漸進(jìn)式變化,導(dǎo)致一般的具有學(xué)習(xí)能力的融合方法很難適用. 本文提出了一種具有漸進(jìn)學(xué)習(xí)能力的融合方法,它具有良好的自適應(yīng)性和魯棒性. 該方法由一種名為接受域加權(quán)回歸(Receptive FieldWeighted Regression) 的漸進(jìn)式學(xué)習(xí)算法和加權(quán)平均的融合算法組成.最后以三個攝像機(jī)聯(lián)合定位作為研究對象,對該方法進(jìn)行了仿真,驗(yàn)證了其有效性,同時還和基于BP 神經(jīng)網(wǎng)絡(luò)的融合方法進(jìn)行了比較.
關(guān)鍵詞: 　傳感器融合; 漸進(jìn)式學(xué)習(xí)算法; 接受域加權(quán)回歸
Abstract : 　The multisensor systems under the dynamic and unknown environment often encounter the incremental modification of environment and its configuration. This results in the fact that the fusion methods with learning ability cannot be suitable any more under this condition. In this paper ,a new fusion method with incremental learning ability is proposed. This method utilizes an incre2 mental learning algorithm called Receptive Field Weighted Regression (RFWR) ,and weighted average is used as the fusion strategy , thus it is more adaptive and robust than previous ones. The problem of three cameras positioning is taken into account and the corre2 sponding simulation is implemented. Simulation results verify the effectiveness of this method. Comparison with BP neural network2 based fusion method is also provided.
Key words : 　sensor fusion ;incremental learning algorithm;RFWR