Identification Of Citrus Huanglongbing By Near Infrared Spectroscopy Based On Extreme Learning Machine

The early identification of citrus greening disease is significant to the citrus industry.The near infrared（NIR）spectroscopy has the advantages of convenience,rapidness and pollution-free in identifying citrus huanglongbing.In order to solve the issue of low accuracy and poor stability of traditional detection methods,this paper proposed two kinds of near-infrared spectral models for citrus huanglongbing.The model combined spectral data preprocessing,wavelength screening,optimized deep extreme learning machine and other methods,and carried out experimental verification on spectral data with different resolutions.The main research contents are summarized as follows:（1）In this paper,A model of least angle regression algorithm combined with kernel extreme learning machine was proposed to identify citrus huanglongbing.In this model,the spectral data were firstly unitized by the L2 norm,and the acquired spectral data were preprocessed by the mexican hat wavelet transform.Then,the spectral characteristic wavelengths were screened by the least Angle regression（LAR）algorithm.Finally,the filtered spectral data were classified by the nuclear extreme learning machine（KELM）.The results showed that the LAR-KELM_（RBF）model was more accurate and stable than ELM,SWELM,BP_{（2 layers）},KELM_（RBF）and SVM models,and could be applied to the detection and identification of citrus huanglongbing.（2）This paper proposed a sparrow search optimized deep extreme learning machine to detect citrus greening disease.First,the Deep Extreme Learning Machine（DELM）is constructed by combining the Extreme Learning Machine and Auto Encoder（AE）.Then,the Sparrow Search Algorithm（SSA）was adopted to optimize the input weights and biasing in the Deep Extreme Learning Machine,to compensate for the influence of random input weights and biasing on the network,and improve the stability and prediction performance of the model.In order to verify the effectiveness of this method,different resolutions and different proportions of data sets were used for experiments and compared with BP_{（2 layers）},SVM,ELM,SWELM,KELM_（RBF）and LAR-KELM_（RBF）.And the experimental results show that SSA-DELM has higher classification accuracy and stronger stability than the contrast method.It shows that this method can be applied to the early diagnosis of citrus greening disease by near infrared spectroscopy and has good application value.