Using a single-board computer as a low-cost instrument for SPAD value estimation through colour images and chlorophyll-related spectral indices

The leaf chlorophyll content is a major indicator of plant stress. Therefore, it is often used for the evaluation of crop status to adjust agricultural management to ensure high quality yield while concurrently applying water and agrochemicals in a sustainable manner. Since laboratory procedures for their assessment are time-consuming and destructive, nondestructive methods have been developed recently based on known vegetation spectral response characteristics. In addition to various vegetation indices derived from remotely sensed data, hand-held sensors such as SPAD-502 are currently widely used for in-field sampling to gain precise information for decision-making in terms of best-fitting agricultural management. However, the costs of such commercial devices can be limiting for farmers. The low-cost alternatives that have been developed recently exploit widely accessible digital cameras with sensors sensitive to the visible region of the electromagnetic spectrum. Digital numbers extracted from colour images in RGB channels serve as the input for broadband "chlorophyll index" calculations. Major constraints regarding digital cameras are, however, the natural light illuminance and the necessity of data postprocessing. In the framework of this study, a novel technological solution was developed to address these issues. A Raspberry Pi single-board computer together with a Pi Camera and a simple LED incorporated in a 3D print case created a prototype called Rasp2SPAD, which was programmed to acquire and analyse a colour image. The prototype and its setup were further tested on the experimental plant material of the winter rapeseed. A set of 22 chlorophyll-related parameters across various colour representation models were generated, from which an SPAD value was modelled using i) a simple linear model, ii) a generalized linear model, and iii) an artificial neural network. The blue (Cb) and red (Cr) chroma components of the YUV colour space were found to be most suitable for SPAD value modelling. Calibration equations were determined, and the results reached relatively high accuracy (mean absolute deviance 1.85 and R-squared 0.81 for simple linear model) while keeping the costs significantly low compared to the most commonly used commercial sensor. In this way, a simple and cheap methodology was introduced to bring the results of research closer to practice, which should help first spread the precision agriculture concept to a wider audience and second allow them to utilize with it.