Spectral imaging for plant health

Over the past decades, agricultural sciences relied principally on reflectance (in the visible, VIS, 0.4–0.7 μm, near-infrared, NIR, 0.7–1.3 μm and short wave-infrared, SWIR, 1.3–2.5 μm regions), thermal (in the thermal infrared, TIR 7.0–20.0 μm region) and fluorescence (at 0.68 and 0.74 μm wavelengths) sensors. 

Depending on the application, these sensors can be used for microscopic observation (i.e., laboratory spectroscopy or hyperspectral microscopy), ground or proximal sensing, airborne, and satellite remote sensing. The main applications in plant health detection are based on the spectral wavelengths ranging from 400 to 2,500 nm, since reflectance in the VIS, NIR and SWIR is primarily influenced by photosynthetic pigments, cell structure and water content, respectively.

A promise for the future

The research community has made important steps in identifying Spectral Reflectance Indexes (SRIs), a combination of different wavelengths with high discriminative power for specific biotic and abiotic stress factors. 

Despite the high potential of spectral imaging techniques for effective disease detection, considerable barriers hinder the adoption of these technologies in typical farming operations. In this article, we dive deeper into the technological challenges and attempt to highlight the opportunities from adopting “superhuman” vision in the farms of the future. 

The challenges

As promising as it may sound, spectroscopy and its application in real farms is still a topic that is limited from a technological and agronomical point of view. 

  • Multi and hyper-spectral imaging offers high spectral resolution, but lag behind in pixel resolution when capturing a scene. 
  • Image frame rate is another big issue (<< 1Hz), especially for hyper-spectral imaging systems, compromising their deployment in moving platforms
  • Spectral imaging is greatly affected by ambient light changes that are inevitable in outdoor farming applications. 
  • Spectral discriminance is not directly linked to identifying biotic and abiotic stress factors because reflectance changes are attributed to anomalies in the leaf pigments, cell structure and water content of plants. In simple terms, different enemies can cause the same spectral response. 
  • Satellite and UAV spectral imaging is a well-established technology in agriculture remote sensing, but is not suitable for disease detection on the plant level because of the relatively low pixel resolution for the application. 

Source: Eden Library | Grape vine Monochromatic-28/JUL/2020-v1

The emergence of Artificial Intelligence (AI) and machine learning can fill this gap, especially in terms of data processing and symptom recognition. AI techniques can reduce the amount of processed data, extract the most discriminant features and prioritize on visual patterns that are not comprehensible to humans.


Thumbnail image source: