Identification and prevention of deep green infection play an important role in high-quality production of tobacco leaf. However, at present, it is difficult to automatically and accurately evaluate the infection level, and especially prevent the disease at asymptomatic stage. In this study, a novel infection identification and prevention method for deep green tobacco infection severity based on portable near-infrared spectroscopy (NIR) technology and extreme learning machine (ELM) algorithm is proposed. Firstly, NIR data of deep green leaf infection at different levels was scanned with a portable NIR spectrometer directly from the fresh tobacco leaves without defoliation and any sample preparation procedures in the field. Then, the qualitative and quantitative models were both built to identify and prevent deep green tobacco infection by means of using ELM algorithm. The experimental results showed that the qualitative model can automatically identify if the tobacco leaf is infected and the quantitative model can accurately acquire the infection condition at asymptomatic stage. These methods are simple, rapid, precise and are helpful to make appropriate decisions in precisely controlling the disease in the field.