Automated canopy temperature estimation via infrared thermography: A first step towards automated plant water stress monitoring
Introduction
Plant canopy temperature acts as a good indicator of the plant water status. When plants are undergoing water stress they increase their temperature. The conductance of water through stomata on leaves decreases when water supply becomes limited to the plant roots. Canopy temperature, relative to ambient temperature, changes as a result of stomatal conductance control of plant transpiration. Thus, if plant water stress increases, transpiration decreases and plant temperature may exceed air temperature. On the other hand, non-stressed plants will have canopy temperatures less than air temperature, particularly when vapour pressure deficit (VPD) is not greater than 4 kPa (Olivo et al., 2009). The crop water stress index (CWSI) relates canopy–air temperature difference to net radiation, wind speed and vapour pressure deficit (Jackson et al., 1981). However, a surrogate measure is calculable from the temperatures of the canopy and reference leaf surfaces corresponding to fully transpiring and non-transpiring canopies (Jones, 1999a, Jones et al., 2002, Moller et al., 2007). Thus, by monitoring plant canopy temperature and the temperatures of wet and dry leaves, it is possible to estimate the underlying plant water stress status and therefore, intelligently control the related irrigation process. Fig. 1 illustrates a typical plant irrigation strategy, where the plant water status information acquisition (inside the red dotted box) plays a critical role in the optimization of plant productivity and water usage.
The Melbourne School of Land and Environment at the University of Melbourne is currently conducting a research program to estimate spatial and temporal variation in water status of grapevines using data collected from remotely sensed infrared (IR) and optical digital images (Wheaton et al., 2007). The ultimate research goal is to design an automatically controlled irrigation system using CWSI via non-destructive IR thermography sensing and automated measurement processing (Jones, 1999b, Jones, 2004).
Typically, measurement data of the IR thermography sensing system consists of a reference optical image and an IR image. The optical image, taken at the same location as the IR image using a digital camera, provides a visible light view of the IR image scene. In the process of plant water stress analysis, the value of CWSI is calculated based on the knowledge of the underlying plant canopy temperature, and the dry and wet reference leaf temperatures. These temperatures can be estimated once the temperature distribution relative to leaf area is obtained from the measurement data. The acquisition of the temperature distribution of plant leaf area is a major component in the measurement data processing. With the reference optical image, the area of interest (e.g., the plant canopy, particularly the transpiring leaves other than ground or sky) may be identified in the IR image. Subsequently, temperature data associated with the area of interest is extracted and statistical analysis can then be performed. The effectiveness and robustness of the analysis method have been reported on several occasions in the literature (Moller et al., 2007, Jones, 2004, Jones and Leinonen, 2003). However, automating this process for automatic controlled irrigation program poses major challenges.
In previous work (Yang et al., 2009), an automated cross-correlation (ACC) based image registration method was developed and described. The ACC algorithm is able to effectively determine the overlapping area between the IR image and reference optical image without human intervention and without the assistance of artifacts placed in the scene to facilitate registration. The registered optical image can then be used to identify plant canopy area and extract the associated temperature distribution from measurement data.
In this paper, our focus is to estimate the plant temperature distribution from the IR image via the registered optical image. Given a perfect match between the optical and IR image, a simple colour identification technique, based on a single Gaussian model in colour space, can be utilized to extract leaf area from the optical image. The temperature distribution of the plant leaf area can then be found from the temperature data associated with the IR image. While this simple method is effective and robust, it relies on the assumption that both optical–IR image alignment and leaf area extraction are perfect. In practice, when such an assumption invariably fails to hold, the estimate can differ considerably from results obtained manually. In particular, the simple colour identification technique fails when the temperatures of the reference leaves, which are embedded in the scene and usually deemed to be the highest and lowest temperatures in the leaf area, are required to be estimated. Here we address this issue and propose a novel algorithm to solve the problem. In this approach, the underlying plant canopy temperature distribution is considered to be the fusion of two temperature densities separable via colour identification and Gaussian mixture distribution extraction techniques. In addition to estimating leaf temperature via colour identification, the canopy temperature distribution, also assumed to be a single Gaussian, is estimated directly from the associated temperature data. The final leaf temperature distribution is obtained by fusing the two estimates. The ideas behind, and the operation of, the proposed algorithm are illustrated in Fig. 2. Furthermore, a N-average method is tested to estimate reference leaf temperatures from the estimated leaf temperature distribution. Results from the analysis of 10 image pairs demonstrate the effectiveness and consistency of the proposed algorithm.
This paper is organized as follows: Section 2 describes the problem of plant water stress status monitoring via IR thermography and the issues of automating the process; Section 3 presents a solution for the estimation of plant leaf temperature distribution; Section 4 demonstrates the estimation of the embedded reference leaf temperatures via a N-average method from leaf temperature distribution; Section 5 presents the results of image analysis and algorithm performance and Section 6 presents the conclusions.
Section snippets
Problem of canopy temperature estimation
A key procedure for the evaluation of crop water stress from plant canopy temperature is to calculate CWSI based on the data collected from IR thermography systems. The CWSI described by (Jones et al., 2002) is of the following generic formwhere and represent the reference temperatures for dry (non-transpiring) and wet (fully transpiring) leaf surfaces respectively. is the temperature of the transpiring surface, i.e., the actual measured temperature of all sunlit leaves
Canopy temperature estimation
A straightforward estimator for uses leaf temperature extraction via colour identification and is described in Section 3.1. However, if the optical and IR images are misaligned, this simple method is often prone to errors, and the CWSI is very sensitive to these kinds of errors. Therefore, an alternative method for leaf temperature extraction via temperature data is considered in Section 3.2. The ultimate leaf temperature distribution is then obtained by fusing these two temperature
Computing temperature of reference leaves
Once the plant leaf temperature distribution is computed, the leaf canopy temperature is estimated by . When reference leaves are embedded in the scene of image, the estimation of these reference temperatures ( and ) are required in order to compute the plant CWSI. Intuitively, both temperatures are extreme values in and can be simply found as and , respectively. However, the and are typically represented by multiple pixels and the CWSI is
Experiment results and discussions
The procedure described in Sections 3 Canopy temperature estimation, 4 Computing temperature of reference leaves with the aim of
- 1.
Automation of all processes (see Fig. 2).
- 2.
Checking consistency with results obtained using an alternative manual image processing method.
- 3.
Examining the robustness of the algorithm. For example, we check CWSI performance when leaf colour is calculated from different thresholds.
Test computations were performed using measurement data from 10 optical and IR image pairs and
Conclusions
Following the work initiated in (Yang et al., 2009), we have developed an automatic canopy temperature estimation method. It attempts to automate the sensing and information processing for plant water status monitoring with the aim of providing the sensing for an automatic controlled irrigation system. In the proposed algorithm, the underlying plant leaf temperature distribution is fused from two temperature densities obtained via colour identification and Gaussian mixture distribution
References (13)
- et al.
Stem water potential is a sensitive indicator of grapevine water status
Annals of Botany
(2001) - et al.
One the relationships between stomatal resistance and leaf temperatures in thermography
Journal of Agricultural and Forest Meteorology
(2008) Use of infrared thermometry for estimation of stomatal conductance as a possible aid to irrigation scheduling
Journal of Agricultural and Forest Meteorology
(1999)- et al.
Maximum likelihood from incomplete data via the EM algorithm
Journal of Royal Statistical Society
(1997) - et al.
Canopy temperature as a crop water-stress indicator
Journal of Water Resources Research
(1981) Use of thermography for quantitative studies of spatial and temporal variation of stomatal conductance over leaf surfaces
Journal of Plant Cell and Environment
(1999)
Cited by (72)
Identifying sunlit leaves using Convolutional Neural Networks: An expert system for measuring the crop water stress index of pistachio trees
2022, Expert Systems with ApplicationsDeep learning for identification of water deficits in sugarcane based on thermal images
2022, Agricultural Water ManagementRevisiting crop water stress index based on potato field experiments in Northern Germany
2022, Agricultural Water ManagementApplication of infrared thermography for irrigation scheduling of horticulture plants
2021, Smart Agricultural Technology