With a mean production of 450–550 ×106 kg/year of olive oil, olive tree cultivation is undoubtedly one of the main sources of agricultural revenue for Italy. In particular, the Apulia region in the south, with over 360 kha (kilohectares) covered with 21 different olive cultivars with a prevalence of Ogliarola and Coratina cultivars, is the region with the highest percentage of production (>35% of the total yearly Italian production) [1]. In the past decade, this production has been greatly impacted by many threats, primarily Xylella fastidiosa (Xf), a pathogen that has been known around the world for decades, but which since 2013 has put the survival of Apulian olive cultivation at great risk. It is a bacterium that can attack olive trees, vines, oleander, and some species of citrus fruits, causing them to rapidly dry out. This phenomenon, when observed on the olive trees, is known as olive quick decline syndrome (OQDS) [2,3,4,5,6,7,8,9,10,11].

Xf is endemic to the American continent and, until recently, it did not exist in Europe [12]; indeed, its arrival in Europe was tracked back to the import of some infected ornamental plants from Costa Rica (Central America) to Gallipoli (province of Lecce in southern Italy) in 2013 [13]. From there, the bacterium spread to the northwest provinces of Brindisi and Taranto, and some infected trees have also been recently reported in the province of Bari (northeast). A large number of publications about the impact of Xf in Puglia are available; a small selection is included in the references [14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31].

The main problems concerning the detection of this disease in olive trees are the possible lack of symptoms over an incubation period ranging from 6 to 18 months from infection and the nonuniform distribution of the bacterium on the infected plants, making it somewhat difficult to identify until it is too late.

Characterization of infection spread may potentially be achieved as was done previously for citrus using geostatistical analysis and kriging estimation, which might also be combined with Kalman filter prediction; however, the availability of data from extended monitoring is fundamental [32,33].

To date, the most accurate method to detect the presence of the Xf bacterium is by means of laboratory genetic analyses using the PCR technique (polymerase chain reaction [34,35]), a sophisticated and complex technique used to reproduce small segments of DNA many times in order to be able to process them in successive tests. The PCR technique is more sensitive than serological analyses of the ELISA type (enzyme-linked immunosorbent assay [36,37]), which are sensitive to antibodies or antigens of a given pathogen. Due to their inherent lower sensitivity, ELISA-type tests can produce a greater number of false negatives. However, both these techniques require medium to long waiting times (some days) to produce results and are applicable only in a laboratory using high-cost analytical instruments, so they are not applicable in real time in the field. Indeed, they require intensive in situ inspection though interesting methods useful to estimate water content and thermal characterization have been proposed [38,39,40,41]. Recently introduced alternative techniques involve proximity or remote sensing, i.e., the use of electromagnetic radiation and its interaction with objects and living beings. The advancement of satellite and aerial detection techniques, telecommunications systems and optical sensors has led to the application of these analysis techniques to images acquired by satellites (remote sensing by satellite), small manned planes or helicopters, or aerial platforms with a remote pilot (unmanned aerial vehicles; UAVs), commonly called “drones” in a wide range of fields [41,42,43,44,45]. UAV platforms can be further distinguished as fixed-wing types, which allow monitoring of large areas from medium–high altitudes, and rotary-wing types, which allow observation of less extensive areas from medium–low altitudes. In recent years, the application of UAVs in precision agriculture as well as in many other fields is becoming more and more common, requiring test systems able to guarantee and certify both electrical and mechanical performance aspects of their propulsion subsystems [46,47,48].

Typically, the radiation reflected by vegetation is concentrated in the visible (VIS), near-infrared (NIR), and medium infrared (SWIR; short-wave infrared) spectral regions, while the emitted radiation is concentrated in the thermal infrared (TIR) spectral region.

Spectral analysis finds frequent and extensive use in many areas of the physical sciences; this technique is one of the statistical methods used to characterize and analyze sequenced data in one-, two-, and three-dimensional space. In this area, many studies have been devoted to reducing bias and variance of the estimates [49,50].

The spectral signature of vegetation, which is the relative intensity of the re-radiated radiation as a function of the wavelength of the incident light, contains a range of information. Indeed, the shape of this curve

The use of terrestrial or aerial drones, both manned and unmanned, equipped with multi- or hyperspectral image cameras to study the health status of plantations of various kinds is not a novelty in precision agriculture or for forestry monitoring [51,52,53,54,55]. There are also pioneering applications of drones for the detection of Xf-infected plants [56,57,58,59].

Xf represents such a serious threat to the future survival of the olive orchards in Italy that the Italian Ministry of Economic Development (MiSE) recently committed €3.5 million to funding another important research project named REDoX (Remote Early Detection of Xylella), which is focused on the detection and monitoring of Xf using multispectral, hyperspectral, and thermal imagery obtained by aircrafts, UAVs, and satellites. The REDoX project is being coordinated by the Apulian Aerospace Technological District (DTA) [60].

In this paper, it is shown that multispectral imagery shot using a midsized rotary-wing UAV can be successfully used to evaluate the health of olive trees in nearly real time with respect to olive quick decline syndrome due to Xf. For this purpose, a tree segmentation algorithm was developed and linear discriminant analysis (LDA) was applied to multispectral stacks. In Section 2, after a brief introduction to remote sensing in agriculture, the equipment used in this research and standard vegetation indexes are described. The proposed algorithm is also presented: image preprocessing is described Section 2.1; 3D reconstruction of the scene is described in Section 2.2; tree segmentation is detailed in Section 2.3; and health status classification is described in Section 2.4. Experimental results and performance evaluation are provided in Section 3, followed by discussion in Section 4 and conclusions in Section 5.