Sensors and methodologies design for image processing in precision agricultureCrop water status and fruit ripening monitoring

Abstract

This doctoral dissertation has been presented in the form of thesis by publication. Water is an essential resource for human life and its scarcity has a decisive impact on society and economy. Agriculture is the main consumer of fresh water and thus is particularly affected. In arid and semi-arid regions, such as southeastern Spain, the availability of irrigation water conditions the type of crop and the production volume. In addition, the current global context of climate change and population growth stresses the situation with a greater water deficit and demand for agricultural products, which constitute the basis of our diet, respectively. Therefore, we face the unavoidable challenge of being more efficient, to produce more with less. The response of science has been the development of regulated deficit irrigation (RDI) techniques, which aim to significantly reduce the consumption of water resources precisely at certain periods of the season, minimizing the impact on the volume and quality of the harvest. However, the successful implementation of these irrigation strategies requires continuous control of the plantation's status. It is crucial to prevent water stress from exceeding extreme levels, which would threaten production or ultimately lead to an unrecoverable crop condition, and to ensure that water needs are satisfied during critical periods. For this reason, different indicators of water status are employed, being the stem water potential (SWP), measured at midday with a pressure chamber, the one considered as a reference. The disadvantage of this method lies in that it does not allow continuous monitoring, so it is not feasible for automatic irrigation management. With the objective of estimating crop status autonomously, other related indicators have been proposed for indirect measurement. One of these is crop canopy temperature, which has been extensively studied. Plants regulate their temperature by means of evapotranspiration, which they control through stomatal aperture. When they experience water deficit, they cannot afford to lose wáter in an unlimited way, so they restrict stomatal aperture. Hence, their thermal regulation mechanism is limited and, consequently, the increase in leaf temperature is irrepressible under high ambient temperature conditions. This effect can be remotely captured through infrared sensors, such as infrared radiometers (IR), which are the most widely spread in the field. Nevertheless, they lack the capability to discriminate the crop canopy in their temperature measurement, which is critical to ensure accuracy, making them totally dependent on correct orientation. The use of thermography as an alternative has become more common in recent years, favored by cheaper technology. Thermal cameras provide a temperature map on which it is possible to filter out those values that do not correspond to regions of interest. To automate this process, visible image processing is used. Automatic image segmentation is a classic problem for computer vision. Recently, the Artificial Intelligence (AI) revolution has had a decisive impact on the performance of image processing applied to complex natural environments.