VISIBLE SPECTRUM IMAGE ANALYSIS FOR ESTIMATION OF PHENOLOGICAL STAGES IN IRRIGATED BEAN CROPPING

Objective: Analyze the spectro-temporal behavior based on vegetation indices based on the visible portion of the electromagnetic spectrum, using images acquired by a drone in comparison with satellite images. Theoretical Framework: Beans (Phaseolus vulgaris L.) are one of the most economically important crops in Brazil and applying technologies aimed at precision agriculture have been more accessible and are fundamental tools for crop management and monitoring. Method: Drone and satellite image captures were carried out in seven moments to obtain vegetation indices, the products generated are thematic maps of: GLI. VARI. NGRDI and VEG, which were tested using various statistical tools to ensure reliability and validity. Results and Discussion: In normality tests at a level of statistical significance of 5% for the satellite and drone data sets, both showed the same behavior, in all drone data indicated normality assumptions (p-value = 2.2e-16) and the satellite data followed the same behavior, (p-value < 2.2e-16). Research Implications: These results highlight the great potential of using visible spectrum images from UAVs and Sentinel-2 for harvest management due to the spatial variability of bean maturation. Originality/Value: The use of precision agriculture to estimate phenological stages optimizes the use of water, fertilizers and pesticides, influencing the efficiency of resource use and the profitability of the crop.


INTRODUCTION
In the world agricultural practices are constantly modified, with this, precision agriculture (PA), coupled with technological innovation, has become indispensable to the farmer.Improving shortcomings and highlighting the potential of agricultural production, besides assisting in decision-making and consequently contributing to boost productivity and reduce the environmental impacts generated by agricultural production.
With technological developments, the unmanned aerial vehicle (UAV) is gaining more and more space in agriculture.The collected images are analyzed in software assisting the management of the most diverse aspects of a crop, as they provide quantitative and qualitative data to measure the efficiency of sowing/planting, quantify areas with diverse faults, detect diseases and/or pests, as well as excess/absence of irrigation.
Analysis of the phenological stage of the culture can be acquired by calculating vegetation indices from visible images composed of RGB bands.Vegetation indices can be used to determine growth parameters such as leaf area index, plant biomass, crop productivity, water stress level, and a range of vegetation parameters and characteristics.
Thus, the objective of this study was to examine the relationship between the indicators of vegetation of the bean plant obtained by drone and satellite images, with a view to assessing the spatial variability of the results.This aids in decision making during harvesting, considering the estimation of phenological stages, such as the filling phase of the pods (R8) and the maturation (R9) of the crop of the irrigated beans.

THEORETICAL FRAME
Remote sensing (SR) emerged in the 1960s and refers to the acquisition of data on objects without direct contact, a practice that has been continuously improved (FUSSELL et al., 1986;JENSEN, 2007).This technique is used to obtain information for various purposes.
The process of obtaining images by SR involves the energy of the sun hitting a target, such as a planting field, where part of this energy is reflected and captured by sensors in satellites or drones.This data is then received by a data station and distributed to databases (Figure 1).

Remote Sensing Scheme (SR).
Currently there are numerous satellites in Earth's orbit, and each one exercise a function as a source of data capture from the most diverse areas of knowledge, positively attributing with the understanding of the geophysical characteristics of the use and occupation of the soil, environmental quality, meteorological forecasting and among others.
The use of remote sensing equipment is essential in precision agriculture, allowing the management of soil and crop variability over space and time (COELHO, 2005;LAMPARELLI, 2016).This management system is directly linked to economic and sustainable return (BRAZIL, 2012), contributing to more sustainable production (THOMPSON et al., 2019).
Thus, acceptance by farmers has been positive over the years, with increasing confidence in technology.
As already seen, the main basis for SR is the relationship between the target and electromagnetic radiation (REM).There is an interaction of the physical, chemical and biological properties arising from the energy source with the target and later the data are transformed, able to be analyzed and interpreted (LIU, 2006;NOVO, 2010).
Such data is collected remotely by Unmanned Aerial Vehicles (UAVs), known as drones, a term derived from the English term Unmanned Aerial Vehicles (UAVs).The market offers a variety of models of UAVs with specific characteristics and purposes, being a multidisciplinary technology.According to Alves Júnior (2015), UAVs are classified into five categories: fixed wing, rotary wing, air balloons, flapping-wing and hybrids.
In rotary-wing drones as well as fixed-wing drones, it is possible to plan flights, perform automated missions and manage flight data using different applications for flight planning that allow access to information collected by an RGB camera or a multi-spectral camera.

5
Cameras operating in the visible range are known as RGB sensors, referring to the red (red), green (green) and blue (blue) bands (MARQUES FILHO and VIEIRA NETO, 1999).
The RGB images are colored by the fusion of these three bands (PARANHOS FILHO et al., 2008).According to Furlanetto et al. (2017), multispectral sensors are the most commonly used in agriculture.
The newest, the multispectral, combining high resolution cameras in the visible with individual multispectral sensors, providing high spatial and spectral resolution and precise spectral measurements (SZABÓ et al., 2018).They typically feature bands such as Red-Edge (725nm), Near Infrared (850nm), Red (660nm), Green (550nm), and Blue (450nm), and include GPS for geo-referencing of images.
The thermal cameras have a radiometric thermal sensor with resolution ranging from 160x120 to 620 x 540 and temperature range from -10°C to +400°C allow combined thermal and RGB images in thermal visualization (temporal synchronization of thermal and RGB images).
After data collection, these can be analyzed by several spectra, with vegetation indices being of great agricultural importance.Vegetation indices are arithmetic operations applied to the spectral bands of SR images, with the aim of highlighting the presence or vigor of vegetation, verifying the interaction of the plant with different values of wavelengths (LIU, 2006;PONZONI, 2001;BOHRER et al., 2009 andSHIMABUKURO, 2010).
The leaf is the main element of the plant that often has interaction with electromagnetic energy, in which the spectral reflectance will result in different pigments, fundamental for studies of vegetation indices (PONZONI, 2001;AMRI et al., 2011).Important indexes include: • The Green Leaf Index (GLI) aims to determine the degradation of vegetation through aerial images.It is based on the normalized difference of the spectral reflectance of red, green and blue (LOUHAICHI et al., 2001).
• Normalized Green Red Difference Index (NGRDI): It is widely used to estimate the fraction of vegetation, green biomass and as an indicator of plant phenologies.It uses the difference of green and red and was an index presented by Tucker (1979).
• Visible Atmospherically Resistant Index (VARI): It is known for its low sensitivity to atmospheric effects and for providing accurate vegetation results.Proposed by Gitelson et al. (2002) and its formula uses the spectral bands of red, green and blue, with the subtraction of the blue band in the denominator to minimize atmospheric effects.
Vegetative (VEG): Aims at identifying the existence of vegetation or exposed soil.
Proposed by Marchant and Onyango (2002), it uses the spectral bands of red, green and blue.

METHODOLOGY
The work was carried out at the Buenos Aires Farm, located in the municipality of Luziânia, in the state of Goiás, in the Center-West region of Brazil.Located under the geographical coordinates 47°52'55"W ;16°19'00"S and 930 meters and distance 70 km from Brasília-DF.According to the Köppen climate classification (1972), the region is of the CWa type, characterized by two well-defined seasons: a dry season, which begins in late April and runs until September.
The area of interest was being cultivated with beans (Phaseolus vulgaris) and sowing was carried out on 26/06/2021.Table 1 presents the collection dates, catch times and Julian Day (DJ) of the year 2021, as well as the processing data of the collected images.The total area of the pivot is 50 ha.The Unmanned Aerial Vehicles (UAVs) used in the study was an Anafi Thermal Parrot The flight plan was carried out by the Pix4D application, with the following configurations: height with 60 meters above ground level, speed of 1.5 m/s and the overlap of the images were set to 90% (Table 1).

Flight dates and times (A) and Processing time (B).
In order to obtain the vegetation indices with the data collected by the drone, images collected on the above dates were used, but the dates of revisitation of the satellite in the area of interest occurred with at most one day, more or less, of difference, being the respective dates: 01, 06, 11 and 19/09.
The three bands B2, B3 and B4, which correspond approximately to blue, green and red light, respectively, are typically used to generate "true color" products, directly mapping these band reflections to pixel RGB values.
To illustrate and understand the dynamics of the field and office for obtaining the vegetation indices, below (Figure 2) is a flowchart.For the calculations, in the QGIS 3.10.12, in the raster tab was selected the tool "raster calculator", the bands of each orthosaic were separated from the alpha layer, so we formulated equations 1, 2, 3 and 4, respectively.Table 2 presents each equation for obtaining the vegetation index, with their respective references.Where: GLI -Green Leaf Vegetation Index; NGRDI -Vegetation index of the normalized green and red difference; VARI -Vegetation index with visible atmospheric resistance; B: Reflectance value in the blue spectral band; G: Reflectance value in the green spectral band; R: Reflectance value in the red spectral band; B: Reflectance value in the blue spectral band and a: Value equal to 0.667.

RESULTS AND DISCUSSIONS
Statistical analysis can be observed in Tables 3 and 4, respectively for UAV and satellite data, in which a pattern of behavior of drone images in variations of vegetation indices GLI, NGRDI and VARI is notable when compared with VEG.However, smaller amplitudes are identified in GLI, NGRDI and VARI, while in VEG these variations are higher.In addition to being able to verify that in the UAV data the NGRDI occurs the same pattern of behavior of the minimum, maximum and average values, making it very similar to the VARI.The data from the sentinel-2 satellite, on the other hand, show that the behavior of the vegetation indices is very close in the NGRDI, although the results obtained in the VEG index are very different.In all the flights carried out with Drone, it was possible to generate orthomosaics suitable for calculating the vegetation indices (GLI, NGRDI, VARI and VEG) on the cultivation of beans irrigated by a central pivot.All the aerial surveys obtained data with sufficient quality for mosaicing, without the need for repetitions of flights.
Figure 3 shows the orthosaics and their respective indices of the first flights.In a visual way, we can note that there is a similarity in the representation of the maps coming from the satellite images, in comparison with the images from drones.10 The image is part of the pivot, approximately 10% of the area.The GLI, NGRDI and VARI are characterized by everything that is close to 1 is intense green leaf and all the behaviors that approaches -1 is characteristic of exposed soil, so the edge of the map is of exposed soil, are roads that interconnect other areas of cultivation of the property, and the northeast region of the maps, towards the center of the area, which is presented with the most green and intense green color, characterized by reflectance and presence of chlorophyll, being an indication of the amount of nitrogen present in the leaves (WINDER, 2018).
According to Santos et al., (2020), the GLI index has made it possible to clearly visualize the areas of greater vigor and critical areas, the lower values reflect the green index, so very dry pasture areas are highlighted by the orange.With this, it can be stated that with the use of this methodology it is possible to carry out the monitoring of the crop, assisting in the taking of decisions on the harvest, estimating the percentage of area at the point of harvest and making possible an analysis of the spatial variability of the senescence of the crop.
The first evaluation of the crop was carried out at 68 and 69 days after sowing (DAS), when the plant was in the reproductive stage R8, represented by the yellow coloring, because in this phase there is the appearance and beginning of the filling of the pods (FANCELLI et al., 2007;DIEL, 2010).Characterized by the physiological ripening point of the plants, where senescence of the older leaves occurs, resulting from the degradation of chloroplasts (MEYER et al., 2013).This stage was observed with greater evidence in the drone images.
In the second data collection, present in the UAV map, we noticed a growing area of yellow tones in the data, expression of class R9, distinguishing from the representation of the map of the sentinel-2, which presents the class R8 in its border, however, as it approaches the center there is predominance of green.
Resulting in: the VEG of the UAV and sentinel-2 data showed greater similarity among the indices, in which case, sentinel-2 fell into R9.The NGRDI and VARI values remain standard for UAV images, and the VEG percentage has the same behavior as the first flight.
The VARI index correlated with GLI presents good experimental accuracy (OLIVEIRA et al., 2019).In the research of Silva et al. (2022), the results were quite similar in all indices (GLI-TGI-VEG), but the most consistent with the RGB image came from VEG.
The applicability of studies in the monitoring of plant coverage may be responses to reduced photosynthetic rate or changes in the structure of the plant canopy (MENESES, 2001;ZANZARINI et al., 2013), and NGRI has correlation of the value of the index with green biomass (WAN et al., 2018).And the VEG index presents sensitivity when the vegetation coloration, thus managing to diagnose where there were weak areas and areas with very healthy vegetation (SILVA et al., 2022).
Figure 4 shows the maps and their corresponding indexes of the last flights.The third repetition is predominant phase R8 (yellow).At the northern end of the area we can observe that at 78 days of sowing the vegetation is in transition from R8 to R9 phase, so the characteristic that brings the coloration expressed in yellow to orange is characteristic of the start yellowing of the leaves due to the death of our annual culture.Therefore, the R9 stage is where the ripening of the vargens takes place, because of this occurs the desiccation of the leaves and even of the vargens, where it brings a reflective response due to the presence of greater exposure of the soil and the color that the plant in a state of senescence.
In the fourth repetition, sentinel-2 and Vant, to the east of the area we noticed in highlight that it was in the harvesting phase.As it is a high slope area, mechanized harvesting is not feasible, and the technique used is manual grubbing in installments according to the A B 12 availability of workers, but harvesting takes place east to west, accompanying in turn the lines that have ripening point R9.
Embrapa in 2018 already pointed out the applicability of the use of UAVs in the imaging and applicability in the GLI and VARI vegetation indices, with enormous potential for monitoring and obtaining agronomic parameters throughout the cycle of a corn plantation Advantage of applying the methodology is to fly over the area of interest for about 20 to 30 minutes, after which it will take 40 minutes to 2 hours to perform the orthosaic of the area.
This technique in turn brings the producer an immediate answer as to the real state of the crop in specific areas of his property.Bezerra et al. (2020) remote sensing is a strong ally of environmental studies, mainly in areas where there is insufficient surface data.
In view of the above, we can use the tool as an indicator of maturity point due to the responses of the crop spectrum.With this the aid of monitoring the final stage of cultivation can be favorable for making decisions of the choice of the ideal moment to start the harvest, optimizing in the management of time and estimation of the man-hour workforce, as in the case of the property under study.

CONCLUSION
The study allowed the conclusion that the UAV has the advantage of images with greater accuracy with the coloring symbolized on the maps, to overcome the deficiencies of repetitivity and spatial limitations.
With regard to the indices, the GLI (satellite) was considered the index that showed the best results, while the VARI had inverse values, representing higher values for exposed soil and lower values for intense green leaf.While the VEG (UAV) showed a lot of sensitivity to the typical behaviors of each phenological stage, thus presenting great potential for application for plant monitoring.
multirotor model.It has two FLIR Lepton 3.5 cameras with infrared sensor, able to identify temperatures between -10 ° and + 400 °C, being possible to define the absolute temperature of each pixel.And another visible image sensor (RGB): 1/2.4''21MP CMOS.Combines 4K HD video recording capabilities Thermal imaging and RGB in thermal visualization.Where the controller was connected to a cellular device enabling the creation of the field flight plan.

Figure 2
Figure 2 Flowchart of the field and office steps to obtain the values of the GLI indices.VARI.NGRDI and VEG.

Figure 3
Figure 3 Maps of the bean vegetation indices (Phaseolus vulgaris) at 68 and 69 (A) and 73 and 74 (B) after sowing.

Figure 4
Figure 4 Maps of the bean vegetation indices (Phaseolus vulgaris) at 78 (A) and 84 and 86 (B) days after sowing.

Table 2
Description of vegetation indices used.

Table 3
Description of vegetation indices used from UAV data.Descriptive statistics of the bean vegetation indices (Phaseolus vulgaris).

Table 4
Description of vegetation indices used from satellite data.Descriptive statistics of the bean vegetation indices (Phaseolus vulgaris).