The Improvement And Application Of The Extreme Learning Machine Algorithm

As the development of science and technology,the effective processing of information becomes particularly important as people are facing increasing amounts of data,and machine learning technology turns into an important tool.As one learning technique of machine learning,Extreme Learning Machines(ELM),with its simple theory and easy implementation,has attracted widespread attention and is applied to various fields including imbalanced data learning,noise and missing data learning,feature extraction,face recognition,remote sensing image and so on.Despite its good performance in many studies,ELM still has some shortcomings.For example,serious impacts can be found on the classification accuracy due to irregular distributions of data,redundant information,the noises etc.In the classification of hyperspectral remote sensing images,the geometric characteristics between data samples and the discriminative information contained in the data are not fully considered,making the ELM learning insufficient and thus affecting the generalization ability of ELM.Aiming to deal with the aforementioned two issues,the paper has conducted studies on ELM,and main research achievements are presented as follows: 1)Three data phenomena of noises,outliers,and irregular distributions can cause bad impact on classification as ELM is applied.On account of this phenomenon,this paper combine rough set theory with model of ELM and put forward Rough Extreme learning machine(RELM)by depicting relationships between data.By comparison experiments on UCI data sets,compared with ELM,NFELM,and RAFELM algorithms,it is proved that the proposed algorithm has a better classification effect.2)The Algorithm of ELM failed to utilize discriminant information contained in the data preferably in the classification of hyperspectral remote sensing images,thereby limiting the classification performance of ELM.Therefore,a discriminative information extreme learning machine(IELM)is put forward.The geometric features and discriminant information contained in the data samples is introduced into the ELM model with the IELM algorithm,which enhance classification performance and generalization ability at a certain level.compared with ELM,NFELM,and RAFELM algorithms,the hyperspectral remote sensing image calssification results proved that the proposed algorithm has a better classification effect.