There are hundreds of sources of heavy metal pollution,including the industries of coal,natural gas,paper,and mining.Toxic heavy metals,such as mercury,cadmium and lead,in air,soil,and water are global problems that are a growing threat to humanity.Rice is an important food crop in world,the rice polluted with heavy metal is seriously harmful to people's health.There are many methods to detect the heavy metal,such as inductively coupled plasma-mass spectrometry(ICP-MS),inductively coupled plasma-atomic emission spectrometer(ICP-AES),inductively coupled plasma optical emission spectrometry(ICP-OES),atomic absorption spectrometry(AAS),X ray fluorescence spectrometry(XRF),atomic fluorescence spectrometry(AFS)and so on.Although there are many advantages in the above technologies respectively,they are time-consuming,high-cost and sometimes require considerable analytic skill.Nowadays,as near infrared spectroscopy(NIR)responds to molecular energy transitions associated with hydrogen bonds of organic,while inorganic salts are not expected to directly influence NIR spectra.To our interest,several studies have described useful NIR calibrations for minerals analysis.NIRspectra with supposed NIR-transparent minerals may be due to the association of cations with organic or hydratedinorganic molecules.Thus,in order to develop the fast detective technology on heavy metal polluted rice leaves,NIR was combined with pattern recognition to discriminate the mercury,cadmium and lead in polluted rice leaves.The rice was grown in paddy field polluted by mercury,cadmium and lead,the concentration of which was 1.5,1and 500mg/kg respectively.After 50days growth,the absorbance of near infrared spectroscopy of back of flagleaf was detected with Nicolet Nexus 870(Thermo Corporation USA)and the data was collected with the softwareof Omnic 7.0.The acquired spectra of leaves with different heavy metal treatments were firstly pretreated withwavelet transform and then input in pattern recognition models of back propagation neural network(BPNN)andradical basis function neural network(RBFNN).It was shown that the rice could grow,blossom and bear fruit in mercury,cadmium and lead polluted paddyfield,the concentration of which was 1.5,1and 500mg/kg respectively.The spectra of rice leaves were firstlypretreated with wavelet db2function at 0-5level,and then calculated with back propagation neural network(BPNN)and radical basis function neural network(RBFNN)model.It was shown that the pretreatment of db2function at 3level combined with RBFNN model was best.And the correct classification rates of rice in mercury,cadmium and lead polluted soil and control soil were 95.5%,81.8%,91.3%and 100.0%respectively.Our results indicated that it should be feasible to develop useful calibration models for the prediction of heavymetal in rice leaves.The performance of RBFNN model was best in the prediction of heavy metal(mercury,cadmium and lead)polluted rice leaves.It has also provided a basis of NIR on the recognition of heavy metalpolluted rice,and then ensure the safety of plant environment.