Benchmarking of an affordable thermal camera for plant phenotyping

PEREYRA IRUJO, G.A.; AGUIRREZÁBAL, L.A.N.; FIORANI, F.; PIERUSCHKA, R.

Barcelona

Simposio; EPPN Plant Phenotyping Symposium; 2015

Universitat de Barcelona

High-throughput and non-invasive plant phenotyping devices are expected to greatly improve the speed and quality of plant breeding and research, by relieving the so-called ?phenotyping bottleneck?. In many cases these devices still have high initial costs and, therefore, access to this technology is usually restricted to a limited number of research institutes and private companies, mainly in developed countries. A way of widening the access to these tools is to develop affordable alternatives, which could be deployed with low initial and maintenance costs. Off-the-shelf consumer devices, such as pocket digital cameras or 3D sensors used in videogames, have usually been proposed as alternatives for plant phenotyping. However, their low cost might be associated with limited capabilities (i.e., precision, accuracy, automation), as compared to custom and/or sophisticated phenotyping equipment. Benchmarking of these devices is therefore necessary, in order to establish the range of conditions under which these affordable devices yield useful results.In this work we evaluated the FLIR One (FLIR Systems, USA), which is the first low-cost (ca. EUR 300), consumer-oriented thermal camera that has been released on the market. This device is attached to a smartphone (an Apple i-Phone 5 in the evaluated version), and is based on a low resolution (80 x 60 px) sensor. A comparison was made to a high spatial resolution (320 x 240 px) industrial thermal camera (FLIR A320). The thermal sensitivity of both sensors is similar (0.05°C), but while the accuracy of the FLIR A320 is reported to be ±2°C, that of the FLIR One is not supplied by the manufacturer. Cameras were tested under controlled environmental conditions by analyzing the measured temperatures of two rapeseed plants, which were imaged with both cameras every 30 seconds during 2 hours. The temperature of the background was simultaneously registered with a temperature sensor. Plants were first kept in the dark, and after 30 minutes they were illuminated with LED lights (700 μmol m-2 s-2 PAR), which increased the temperature of the background by 2.4°C and that of the leaves by about 1.9°C. The FLIR One camera showed very high fluctuations in the measured temperatures, which resulted in a very low accuracy in the measurement of the image background temperature (RMSE of 6.1°C vs. 1.0°C for the FLIR A320). As a result of this, the measured temperatures of the plant leaves were not correlated between the two cameras (R2 < 0.01). These fluctuation could, however, be corrected using the background temperature as a reference. After this correction, the measured temperatures of the leaves were highly correlated between both cameras (R2 > 0.95). These preliminary results show that it is possible to obtain non-invasive temperature data from plants using an affordable consumer device, with results that are similar to those obtained with a more expensive, scientific-grade camera. Further testing is required in order to evaluate the performance of this device.