Research of Crop Disease Based on Visible/Near Infrared Spectral Image Technology : A Review
Zhang De-rong,Fang Hui,He Yong
DOI: https://doi.org/10.3964/j.issn.1000-0593(2019)06-1748-09
2019-01-01
Spectroscopy and spectral analysis
Abstract:Crop disease is a major biological hazard in agriculture of China and causes serious interference to farming process, so a fast, accurate and efficient diagnosis method for crop disease is in pressing need. Compared to some common crop disease detection technologies (such as polymerase chain reaction technique, artificial sensory evaluation technique, and statistical method), which are time-consuming or can only be used to detect obvious disease spots, spectral technology has potential in rapid detection of crop diseases and has been studied extensively. This passage mainly focuses on the application of visible/near infrared spectroscopy technology in disease detection, discusses instruments involved in this technology, and analyzes research status of visible/near infrared spectroscopy in disease detection from cell, plant tissue, canopy and larger scale aspects. At present, most of researches on visible/near infrared spectroscopy related to plant diseases are based on plant leaves. Few researches are on smaller scale (from cell to microscale) or larger scale (from canopy to aeronautical/spaceflight remote sensing scale), especially when it comes to disease researches on single cell scale, which are only done in the field of animal cells and have no successful application of visible/near infrared technology. However, visible/near infrared technology has many successful application in researches which are on organ scale of plant leaves. Most of common crops and major diseases of common crops, and diseases caused by fungal and bacterial pathogens are involved in current researches of disease detection. These researches are studied usually in three ways: (1) automatic and rapid diagnosis of disease information based on computer image processing and pattern recognition technology, (2) judgement model spectral analysis for Region of Interest (ROI) extracted from hyperspectral images was established based on stoichiometric method, (3) spectral model of some physical and chemical parameters of leaves related to crop diseases was established to quantify the extent of disease. The main problem related to this scale is that the research is so fragmented, which means only one or a few kinds of diseases are studied, that models can only be used in very specific conditions and can't be used directly to make a full automatic judgment on field samples. What's more, there are few studies on direct monitoring of crop diseases or multi-spectral imaging of near ground whole plants and the classification methods adopted are similar with those of leaf scale data processing. In near ground canopy scale, three dimensional forms of plants become a new source of interference in the spectral model, and some passage showed that 2D image was used as disease detection data with a percentage deviation of 13%. Finally, according to the present situation of all aspects of researches, it is believed that visible /near infrared spectroscopy technology has a good application prospect in crop disease detection, but it is in the bottleneck period now. There exist some problems, including that unbalanced research content of plant disease detection, lack of systematisms caused by overabundance of disease species and insufficient cooperation of different subjects. According to those problems, this passage points out that visible/near infrared spectroscopy technology should pay more attention to the in-depth cooperation of multidisciplinary in the field of disease detection, and it is urgent to make breakthroughs in the related equipment and method model.