Beyond the Visual Spectrum: Early Detection of Plant Disease using Sensors

Plant disease is one of the primary causes of economic loss in the agricultural industry. For ages, experts' visual scouting was the only way to identify the early onset of plant disease. As one Chinese proverb goes, “the best fertilizer is the gardener’s shadow.” However, as grow operations are getting bigger visual scouting may not be a scalable solution for disease management. But help is on the horizon. Recent advances in real-time object identification using algorithms like YOLO and the explosion of machine learning applications triggered by opensource tools like tensor flow has resulted in a deluge of plant disease classification algorithms [1,2,3,4]. PlantVillage dataset, for example, with a carefully curated set of more than 50,000 plant disease images, has become the de facto benchmark for testing new algorithms for their disease classification accuracy.

Several commercial solutions are trying to tap into plant pathogenesis (progression of plant disease over time) hype by mounting cameras on a multitude of mobile platforms such as Drones (Taranis), Satellites (GeoAgro), and even on rails inside greenhouses (iUNU). Their slick marketing videos promise everything from plant stress detection, yield prediction, biomass estimation, vegetation classification, to even identification of the type of fungus or bacteria that is infecting a plant. Still, the reality is that their dashboards offer only plant growth rate monitoring by tracking canopy growth over time. Below is a screenshot from iUNU’s timelapse comparison that shows the plant growth rate.

Though cameras can revolutionize plant disease detection, they are not very reliable at the early asymptomatic stage. The below image shows various bacterial infection stages in a plant, starting from the asymptomatic phase to the late symptomatic phase.

Pathogen Infection progress [5]

For effective pest and disease management, detecting plant stress during the asymptomatic stage offers the greatest reward. Some of the currently available techniques for plant disease detection at various pathogenesis stage is shown below.

Detection Technology and Pathogenesis Stage [6]

Volatile Organic Compounds (VOC) based detection can be the earliest biomarker for plant stress when attacked by a herbivorous. One of the earliest papers that explored VOC profile difference between healthy peanuts against ones infected with white mold has shown that methyl salicylate and 3-octanone VOC concentrations were higher than normal in infected ones[7].

VOC is a catch-all term for any molecule that has transformed into the gaseous state under ordinary room temperature, i.e., they have high vapor pressure at room temperature. The smell of fresh-cut grass, new car smell, scent emanating from the newly painted varnish on a table are all due to molecules that evaporated at room temperature and slowly floated on to your olfactory nerve endings. Plants use VOCs, similar to human pheromones, as a medium for plant-to-plant and plant-to-insect communication [6]. Some of the common triggers for plant VOC are shown below.

Plant VOC Triggers

The plant produced VOC called Green Leaf Volatiles (GLV) has opened up the possibility of asymptomatic detection of plant disease using VOC sensors. VOC sensors can either target an individual VOC molecule or serve as an untargeted, broad range VOC sensor that provides a global perspective of the plant under stress. Since the type of VOC molecule generated by a plant depends on many factors such as the strain of the plant, type of infection, etc., a targeted VOC sensor has minimal application as a biomarker for early onset of plant disease. A broad range VOC sensor, on the other hand, is effective in handling VOC molecule variations. Still, their output is noisy due to its inability to differentiate between GLVs, the cause of GLV, or even VOC generated from non-biological sources such as pesticides, paints, etc.

Metal Oxide (MOx) sensors are a broad range VOC sensor that changes its electrical resistance due to REDOX (Reduction/Oxidation) reaction with VOC molecules present around it when the MOx layer is heated.

MOx VOC Sensor

As mentioned previously, the output from these sensors is highly noisy. This necessitates a multivariate data analysis by bringing in other triggers for VOCs such as CO2, PAR (photoactive radiation) sensor, and soil water data to isolate VOC's effect generated by herbivorous infections. Additionally, to account for VOCs' non-biological sources, they would require baseline settings collected daily using manually entered events such as watering, pesticide application, etc. Below is a rendering of our new VOC wireless solution that would help with broad range VOC detection. We added a slew of sensors in addition to MOx based VOC sensors such as CO2, PAR, Temperature, Humidity, Pressure, soil EC and soil VWC to help with the multivariate analysis to isolate herbivorous triggered GLV.

All-in-one Wireless Sensor Module with VOC, Pressure, Humidity, Temperature, CO2, PAR sensing

References

1. Matese, A.; Di Gennaro, S.F. Practical Applications of A Multisensor UAV Platform Based on Multispectral, Thermal, and RGB High-Resolution Images in Precision Viticulture. Agriculture 2018
2. Singh AK, Ganapathysubramanian B, Sarkar S, Singh A. Deep learning for plant stress phenotyping: trends and future perspectives. Trends Plant Sci. 2018;23:883–98.
3. Nagasubramanian, K., Jones, S., Singh, A.K. et al. Plant disease identification using explainable 3D deep learning on hyperspectral images. Plant Methods 15, 98 (2019)
4. Knauer U, Matros A, Petrovic T, Zanker T, Scott ES, Seiffert U. Improved classification accuracy of powdery mildew infection levels of wine grapes by spatial-spectral analysis of hyperspectral images. Plant Methods. 2017;13:47
5. Reverchon, Sylvie & Muskhelishvili, Georgi & Nasser, William. (, 2016). Virulence Program of a Bacterial Plant Pathogen: The Dickeya Model. 10.1016/bs.pmbts.2016.05.005.
6. Martinelli, Federico & Scalenghe, Riccardo & Davino, Salvatore & Panno, Stefano & Scuderi, Giuseppe & Ruisi, Paolo & Villa, Paolo & Stroppiana, Daniela & Boschetti, Mirco & Goulart, Luiz & Davis, CristinaE & Dandekar, AbhayaM. (2014). Advanced methods of plant disease detection. A review. Agronomy for Sustainable Development. 35. 1–25.
7. Cardoza YJ, Alborn HT, Tumlinson JH (2002) In vivo volatile emissions from peanuts plants induced by simultaneous fungal infection and insect damage. J Chem Ecol 28:161–173.