ES2470065A1 - System and procedure to automatically determine the number of flowers of an inflorescence (Machine-translation by Google Translate, not legally binding) - Google Patents

Abstract

System and procedure for automatically determining the number of flowers in an inflorescence. The method comprises: - capturing an original image (100) of the inflorescence on a background (104) that contrasts with the inflorescence; - convert the original image (100) to the CIELAB color space; - segmenting said converted image by applying a segmentation threshold (200) on the values of the complementary color axes a* or b*, to obtain the pixels representing the inflorescence (202); - filtering the chained pixel groups representing the inflorescence (202) based on their location as intensity values representing local maxima; - grouping in different bright areas (302) the filtered pixels that are interconnected; - filtering on the bright areas (302) to eliminate those that do not correspond to flowers, including filtering by size of the bright area (302) and filtering based on distance from other bright areas (302).

Description

Field of the Invention The present invention relates to a system that provides an estimate of the number of vine flowers through artificial vision. This evaluation allows you to predict robustly, Reliable and very early (four months before the harvest) the grape production of the vineyard.

Background of the invention The determination of the number of flowers of an inflorescence of a plant species in Full field conditions have scientific, biological, agronomic and economic interest. Applied to the vineyard, the reliable and robust determination of the number of vine flowers would mean take a significant step in estimating grape production early and automated, which would allow the wine industry to optimize the management of the vineyard objective form (Le. regulate grape thinning ...) and make predictions of the harvest of final grape, time and day laborers necessary for the harvest, as well as the final price of the grape.

Currently there is no known system in the market that allows to determine in a way automatic number of flowers of an inflorescence of a plant species in conditions full field. It is important to note that the system is valid for all those plants of fruit that have the flowers grouped in an inflorescence, which are the majority. By For example, a concrete and very useful application is the application to the counting of flowers on the vine. Until the date, only grape production estimates are made in the wine sector manual and not objective. The traditional system for estimating the components of the Vineyard grape production is done by weighing the clusters and counting Berries in the laboratory. This requires that the bunches have to be manually shelled then weigh the berries or place them on a tray and be photographed in controlled light conditions for subsequent image processing. Thus, This method is manual and destructive, requiring the harvest of the cluster to be sampled and its Transfer to the laboratory where the images are taken. In addition, for their industriousness, these manual methods require high labor and runtime, so Samples are usually small, insufficient and generally not representative.

Other systems allow the field estimation of the number of berries or grape grains (not flowers) by image analysis. These systems are developed to be used after curdling, at times closer to the grape harvest, when the grapes and clusters are fully formed. In addition, these systems require the taking of images at night (total darkness), to avoid the influence of sunlight.

Therefore, the current systems require specifically constructed instruments, the destruction of the samples taken and hours of unusual images for the work in agriculture, also not allowing an early estimate of the grape production of the vineyard.

There is also quite a lot of scientific literature regarding the importance of determining the number of vine flowers. Flowering and fruit set rate are the most important factors in vineyard production [1]. These two physiological processes define the number of berries per cluster, which together with the weight of the berry determine the architecture and compactness of the cluster (loose or compact clusters), considered as indicators of grape and wine quality [2]. Reproductive efficiency, which determines the setting rate, is dependent on the variety and the selected vine clone ([1], [3], [4]) and influenced by physiological, environmental and pathological factors [5].

The setting rate has been estimated in several works that evaluated the effects of cultivation practices in viticulture since the 19th century ([1], [6]), including pruning in green [7], blunt ([8] , [9]) early leaflessness [10], banding [11] and application of growth regulators ([9], [12]) and nutrients ([13], [14]). In most of these studies, the number of berries per cluster was used to estimate fruit set under the assumption that the number of flowers per initial inflorescence was constant. However, the number of flowers per inflorescence shows a strong variation between strains and between inflorescences of the same plant [1]. Therefore, the determination or counting of the number of flowers per inflorescence is essential for the precise evaluation of the setting rate.

Many methods have been developed to determine the number of flowers per inflorescence. May [15] and Keller et al. [16] proposed to wrap the inflorescence with a dense mesh bag tied to the peduncle of the same from the antecedent to the complete setting, and then count the floral calyces collected and with them estimate the total number of flowers per cluster. This method, although effective, requires a lot of time and effort for each of the inflorescences. In Poni et al. [10] a set of 20 inflorescences were photographed individually with a digital camera perpendicular to it and using a black background. Subsequently, the number of flowers in the photograph was manually counted in an impression of it in order to determine the direct or linear relationship between the number of flowers counted and the total present in the inflorescence.

Computer vision systems are used for automated inspection in agriculture and food processing ([17], [18], [19]). In addition to the analysis of color or surface damage, the shape, size and texture are characteristics that can be objectively assessed by an image analysis system, which makes it an accurate tool for quality control, automatic recognition or estimation of production ([20], [21], [22]). However, the efficiency of a computer vision system based solely on color analysis is highly influenced by the lighting conditions, angle of capture and composition of the object. In viticulture some work has been done in RGB image analysis, with the aim of estimating the number of berries per bunch at harvest time using only color segmentation [23]. Grossetete et al. [24] developed an application capable of counting the number of berries (in pea size), of RGB images taken at night with a mobile phone with a camera, using a system that identifies a bright spot in the center of the berries created by the Reflection of camera flash light. However, this method is not technically valid for the detection of flowers, because their shape does not generate the bright spot mentioned above.

Therefore, there is currently no system that allows automatic estimation of the number of vine flowers per inflorescence in full field conditions without controlled lighting.

BIBLIOGRAPHIC REFERENCES

one.

May P, Flowering and fruitset in grapevines. Lythrum Press, Adelaide (2004).

2.

Matthews MA and Nuzzo V, Berry size and yield paradigms on grapes and wines quality. Acta Hortic 754: 423-436 (2007).

3.

Dry PR, Longbottom ML, McLoughlin S, Johnson TE and eollins e, elassification of reproductive performance of ten winegrape varieties. Austr J Grape Wine Res 16: 47-55 (2010).

Four.

Galet P, Precis De Viticulture. Oehan, Montpellier (1983).

5.

Carbonneau A, Oeloire A and B. J, La Vigne: Physiologie, Terroir, Culture. �ditions Ounod Paris (2007).

6.

Coombe BG, The effect of removing leaves, flowers, and shoot tips on fruit-set. in Vitis vinifera L. J Hort Sci 37 :: 1-1 5 (1962).

7.

Friend AP and Trought MCT, Oelayed winter spur-pruning in New Zealand can alter yield components of Merlot grapevines. Austr J Grape Wine Res 13: 157-164 (2007).

8.

Skene KGM, A comparison of the effects of "cycocel" and tipping on fruit set in vitis vinifera 1. Austr J Biol Sci 22: 1305-1311 (1969).

9.

Collins C and Ory PR, Response of fruitset and other yield components to shoot topping and 2-chlorethyltrimethyl-ammonium chloride application. Austr J Grape Wine Res 15: 256-267 (2009).

10.

Pony S, Casalini L, Bernizzoni F, Civardi S and Intrieri C, Effects of early defoliation on shoot photosynthesis, yield components, and grape composition. Am J Enol Vitic 57: 397407 (2006).

eleven.

Brown K, Jackson DI and Steans GF, Effects of Chlormequat, girdling, and tipping on berry set in Vitis vinifera L. Am J Enol Vitic 39: 91-94 (1988).

12.

Coombe BG, Fruit set in grape vines: the mechanism of the CCC effect. J Hort Sci 45 (1970).

13.

Christensen P, Timing of Zinc Foliar Sprays. 1. Effects of application intervals preceding and during the bloom and fruit-set stages. 11. Effects of day vs. night application Am J Enol Vitic 31: 53-59 (1980).

14.

Longbottom ML, Ory PR and Sedgley M, A research note on the occurrence of 'star' flowers in grapevines: Observations during the 2003-2004 growing season. Austr J Grape Wine Res 10: 199-202 (2004).

fifteen.

May P, From bud to berry, with special reference to inflorescence and bunch morphology in Vitis vinifera L. Austrian J Grape Wine Res 6: 82-98 (2000).

16.

Keller M, Kummer M and Vasconcelos MC, Reproductive growth of grapevines in response to nitrogen supply and rootstock. Austr J Grape Wine Res 7: 12-18 (2001).

17.

Cubero S, Aleixos N, Molt� E, Gómez-Sanchis J and Blasco J, Advances in machine vision applications for automatic inspection and quality evaluation of fruits and vegetables. Food Bioprocess Tech 4: 487-504 (2011).

18.

Ji W, Zhao O, Cheng F, Xu B, Zhang and and Wang J, Automatic recognition vision system guided for apple harvesting robot. Comput Electr Eng 38: 1186-1195 (2012).

19.

Lorente D, Aleixos N, Gómez-Sanchis J, Cubero S, García-Navarrete Ol and Blasco J, Recent advances and applications of hyperspectral imaging for fruit and vegetable quality assessment. Food Bioprocess Tech 5: 1121-1142 (2012).

twenty.

Blasco J, Aleixos N, Cubero S, Gómez-Sanch�s J and Molt� E, Automatic sorting of satsuma (Citrus unshiu) segments using computer vision and morphological features. Comput Electron Agr 66: 1-8 (2009).

López-García F, Andreu-García G, Blasco J, Aleixos N and Valiente J-M, Automatic detection of skin defects in citrus fruits using a multivariate image analysis approach. Comput Electron Agr 71: 189-197 (2010).

22

Cubero S, Diago MP, Blasco J, Tardaguila J, Millan B and Aleixos N, A new method for pedicel / peduncle detection and size assessment of grapevine berries and other fruits by image analysis. Biosyst Eng (2013)

2. 3.

Dunn GM and Martin SR, Yield prediction from digital image analysis: A technique with potential for vineyard assessments prior to harvest. Aust J Grape Wine Res 10: 196-198 (2004).

24.

Grossetete M, Berthoumieu Y, Da-Costa JP and Germain C, Early estimation of vineyard yield: site specific counting of berries by using a smartphone, in International Conference of Agricultural Enginweering (C / GR-AgEng2012), Ed, Valencia, Spain ( 2012).

25.

lorenz DH, Eichhorn KW, Bleiholder H, Klose R, Meier U and Weber E, Phenological growth stages of the grapevine (Vitis vinifera L. ssp. vinifera) -Codes and descriptions according to the extended BBCH scale. Aust J Grape Wine Res 1: 100-11 O (1995).

26.

CIE Colorimetry, Second Edition. Publication C.I.E. no. 15.2 (1986).

27.

Soille P, Morphological image analysis: PrincipIes and applications, Springer-Verlag, Berlin (1999).

28.

Witten IH, Frank E and Holmes G, Data mining: Practical machine learning too / s and techniques. Publisher Morgan Kaufmann, Burlington, pp 650 (2011)

29.

Nuske S, Achar S, Bates T, Narasimhan S and Singh S, Yield estimation in vineyards by visual grape detection, in Intelligent Robots and Systems (IROS), 2011 IEEE / RSJ International Conference on, Ed. IEEE, pp 2352-2358 ( 2011).

Description of the invention the present invention allows solving the above-mentioned problems, by means of a simple, economical, fast, precise and robust solution based on artificial vision, by which color photographs taken during the day in field conditions are analyzed. does not lead to the destruction of the cluster, and the number of flowers is estimated automatically by inflorescence.

A first aspect of the present invention relates to a method for automatically determining the number of flowers of an inflorescence, preferably of a vine. The procedure includes:

-

capture an original image of the inflorescence on a homogeneous background that contrasts with the color of the inflorescence;

-

convert the original image to the CIELAB color space;

-

segment the image converted to the CIELAB color space by applying a segmentation threshold in the values of the so-called complementary color axes a * or b *, to obtain the pixels representing the inflorescence; -filtrate the groups of chained pixels that represent the inflorescence based on their situation as intensity values that represent local maximums; -grouping the filtered pixels that are interconnected in different bright areas; - make a filtering on the bright areas to eliminate those that do not correspond to flowers, where said filtering includes a filtering by size of the bright area

(302) and filtering based on distance with other bright areas.

The original inflorescence image is captured preferably in field conditions with uncontrolled lighting.

The segmentation threshold can be automatically selected for each image based on the groupings of values present in the histogram of the coordinate a * or b * interchangeably. The segmentation threshold is preferably the local minimum value after the global maximum in the histogram of the a * or b * coordinate.

In a preferred embodiment the pixels representing the inflorescence are those whose intensity level in the b * coordinate exceeds the segmentation threshold. In another embodiment, the pixels representing the inflorescence are those whose intensity level in the coordinate at * is less than the segmentation threshold.

Filtering performed on the bright areas preferably includes filtering based on the shape of the bright area. The filtering by shape of the bright areas may include the calculation of the relationship between the major axis and the minor axis of each bright area.

The filtering of the interconnected groups of pixels representing the inflorescence can be carried out based on their consideration as local maximums in their luminosity values.

A second aspect of the present invention relates to a system for automatically determining the number of flowers of an inflorescence. The system is preferably implemented in a mobile device and comprises:

-

image capture means configured to capture an original image of the inflorescence on a homogeneous background that contrasts with the color of the inflorescence; -data processing means configured for:

• convert the original image to the CIELAB color space;

•

segment the image converted to the CIELAB color space by applying a segmentation threshold in the coordinate a * or b *, to obtain pixels representing the inflorescence;

•

filter the pixels representing the inflorescence based on its brightness values;

•

group the filtered pixels that are interconnected into different bright areas;

•

perform a filtering on the bright areas to eliminate those that do not correspond to flowers, where said filtering includes a filtering by size of the bright area and a filtering based on the distance with other bright areas.

The data processing means are preferably configured to select the segmentation threshold automatically for each image based on the groupings of values present in the histogram of the a * or b * coordinate.

The filtering performed by the data processing means on the bright areas preferably includes filtering based on the shape of the bright area, and may include the calculation of the relationship between the major axis and the minor axis of each bright area.

As advantages of the invention, it should be noted that the system used for the detection of flowers is robust and allows the analysis of photographic images obtained in full field conditions. In addition, it is possible to implement it in mobile devices (smartphones, tablets, etc.) that incorporate a processor and a camera for image acquisition.

In summary, the present invention has the following advantages:

•

The photos can be taken in full field conditions, with sunlight, and with the only restriction of using a black cardboard as a background.

•

A fixed distance from the imaging device to the inflorescence is not necessary. The system adjusts itself to the distance taken.

•

The destruction of the inflorescence is not necessary, the photograph can be taken without separating it from the vine.

•

Simple and robust application with possible use of low cost devices.

•

Allows implementation on mobile devices, such as smartphones.

BRIEF DESCRIPTION OF THE DRAWINGS A series of drawings that help to better understand the invention and that expressly relate to an embodiment of said invention that is presented as a non-limiting example thereof is described very briefly below.

Figure 1 shows an original RGB image taken in field conditions to be processed by the proposed system, where you can see the existence of areas with variable lighting.

Figures 2A, 2B and 2C show different histograms of the number of pixels in an image of the Tempranilla variety in the RGB color space (upper row), CIELAB color space (intermediate row) and filtered CIELAB color space.

Figure 3 shows the original image of Figure 1 after applying the segmentation process according to the brightness of the pixels.

Figure 4 shows the result of the processing of the original image of Figure 1, with the detected flowers indicated in black dots.

Figure 5 shows the original image of Figure 1 with the flowers marked manually with an "X".

Figures 6A, 6B, 6C and 60 show the relationship between the number of flowers present in the inflorescence and those detected manually and automatically in the photographs for three different varieties of Vitis vinífera L. (Graciano, Mazuelo and Tempranillo).

DETAILED DESCRIPTION OF THE INVENTION The system for evaluating the number of flowers present in a vine inflorescence is based on the computer analysis of photographic images of vine inflorescences taken in full field conditions (without controlled lighting), with unique use of a black cardboard used as a background, being possible to use other materials such as plastic sheets or other colors with sufficient contrast with the inflorescence. It is not necessary to regulate the distance from the camera to the object, being the only conditions that the inflorescence appears intact in the image and that no foreign objects are observed in it.

The developed system allows to determine the number of flowers visible in the image and, from them, the total number of flowers present in the inflorescence, for which a model has been generated from a set of images that has been validated afterwards. another different and independent set. Segmentation of the inflorescence occurs automatically, by detecting thresholds for each image in the CIELAB color space, which offers great immunity against the influence of external lighting.

Image processing comprises three stages:

1. Image pre-processing: The first phase includes the conversion of the image from the RGB color space to CIELAB and the segmentation of the inflorescence with respect to the background. The segmentation threshold is calculated based on the histogram of the a * or b * coordinate of the filtered CIELAB color space. The CIELAB color space [26] is an international standard for color definition. Developed by the "Commission Internationale d" Eclairage "(CIE) in 1976, it is widely used in color analysis, which is represented by 3 coordinates called L *, a * and b *. The L * component defines the luminosity, between zero and one hundred, which is combined with the other two chromatic components or coordinates: a * representing the variation from green to red (position between green and red, where negative values indicate green and positive values indicate red) and b * representing the variation from blue to yellow (position between blue and yellow, where negative values indicate blue and positive values indicate yellow).

Figure 1 shows, by way of example, an original image 100 of the inflorescence 102 of a vine (of the Tempranilla grape variety) taken on a homogeneous background 104, for example a dark gray or black cardboard, in field conditions for be processed by the proposed system. Figures 2A, 2B and 2e show the color histograms used for the segmentation of the first stage (pre-processed). Represents the number of pixels (y axis) with a certain intensity value (x axis) for the original image 100 of the Tempranilla variety:

Figure 2A: _Histograms A), B), and C) represent the red, green and

RGB color space blue respectively.

Figure 2B: Histograms O), E), and F) represent the components L *, a * and b *

of the CIELAB color space respectively.

Figure 2C: Histograms G), H), and 1) represent the filtered components or

smoothed L *, a * and b * of the CIELAB color space respectively,

by applying a filter (e.g. Savitzky-Golay filter) that softens the

signal shape reducing peaks inherent in sensor noise

photographic, facilitating the subsequent automatic selection of a threshold of

200 segmentation.

Two clusters are observed in the coordinates a * (histogram E and H) and b * (histogram Fe 1) of the CIELAB color space, indicating that it is possible to separate the background 104 with respect to the inflorescence 102. In the histogram of coordinate b * (histogram 1), the local minimum after the global maximum corresponds to the value of the segmentation threshold 200 selected for the segmentation of the image. The distribution of values corresponding to fund 204 is greater than that of inflorescence 202 due to the greater number of pixels that comprise it. In the histograms A, B, and C corresponding to the RGB color space, differentiated clusters are not appreciated, so segmentation based on the values of this color space is not possible. Although the system uses component b * for segmentation, it can be seen that the histogram of component a * (histogram H) also allows separation into two groups.

Therefore, these histograms serve for the adaptive segmentation of the inflorescence 102 with respect to the background 104. For this purpose, channel b * of the CIELAB color space is preferably used where two clusters corresponding respectively to the background and the inflorescence are found. Once the clusters have been detected, a segmentation threshold (200) is established, which allows separation of the inflorescence in the image for later analysis automatically.

The threshold for image segmentation (see Histogram 1) is automatically selected for each image based on the groupings of values present in the histogram of the b * coordinate (the histogram of the coordinate a * can also be used). The pixels corresponding to the background 204 have a similar color and therefore similar values in the b * coordinate. The same is true for the pixels corresponding to inflorescence 202.

2. Flower count: Once the preprocessing has been applied in the original image 100, only those pixels whose intensity level exceeds the segmentation threshold 200, which are the pixels 202 that represent the inflorescence. Flowers have a greater capacity to reflect light than other areas of the image. The objective of this stage is to count the areas of greatest brightness, which generally correspond to flowers. Specifically, the following steps are developed:

to.

Calculation of the "extended-maximum" transform which is the maximum of the region pixel clusters with a constant intensity value surrounded by pixels with lower intensity values of the "H-transform" [27] ( elimination of the maximum values below a threshold), in order to find and identify the brightest points of the brightness of the image (coordinate L *). In this step, the original image 1 00 is converted into a binary image 300, as shown in Figure 3, where the pixels that do not exceed a given brightness threshold are represented in black according to the values of the neighboring and blank pixels those pixels (bright areas 302) that exceed said brightness threshold. The brightness threshold can be selected manually based on the analysis performed on a set of images.

b.

Search of the components connected in the binary image 300 to label them as flowers; To do this, the pixels identified in the step are grouped

earlier found in the vicinity of bright areas 302, as they are considered to belong to the same flower.

C. Measurement of the size, shape and identification of the centroid of the candidate regions to be identified as flowers (bright areas 302).

3. Image post-processing: The third stage focuses on the elimination of bright areas 302 that do not correspond to flowers. Filtering is carried out as a sequential process of three consecutive steps:

to.

Filtering by region size: In this step the bright areas 302 are filtered based on their size. The removal (filtering) of those areas that are larger than 2.5 times the median size of the bright areas 302 is removed.

b.

Distance filtering: This process requires the calculation (and ordering) of the distances between the centroids with which a histogram is subsequently constructed. As the flowers in the inflorescence are presented in clusters, among them they will be presented with a similar distance in the distance histogram, which allows the filtering of those areas that have values far from the majority. In this way it is possible to filter the false positives without the need to set a distance for the taking of images, since the value from which the filtering is performed is calculated for each of the images.

C.

Filtering by form: This filtering includes the calculation of the ratio of the major axis between the minor for each bright area 302. The shape of the bright areas 302 corresponding to flowers have similar values between them and different from bright areas of other areas such as the stem. of the inflorescence or reflections in the background, so that the bright areas that correspond to flowers will be presented grouped in the histogram of the major axis / minor axis relationship, being possible to automatically filter for each image those that do not correspond to flowers.

Once the three processing steps have been applied to the original image 100, a processed image 400 can be obtained showing the detected flowers highlighted with black dots that would correspond to the resulting bright areas 302, once filtered.

Figure 4 shows the overprinted detected flowers in the original image 100 after applying the described method. As can be seen, the developed system allows to determine the number of flowers visible in the image.

To validate the present automatic flower counting procedure, the results obtained by said automatic procedure are then compared with a manually performed count. Figure 5 shows, by way of example, the original image 100 of Figure 1 with the flowers manually marked with an "X". In order to validate the present invention, 90 images of inflorescences of Vitis vinifera L. variety Graciano, Mazuelo and Tempranillo were taken in a vineyard with a compact digital camera before flowering and used to validate the counting system of flowers.

The main objective of this comparison is to determine the linear relationship between experimental measurements, manual counting, and automatic counting with the system described. To validate the system, the number of flowers present in the inflorescence was counted manually, shelling the flowers after the capture of the photograph and also on the printed image to obtain the number of flowers per image.

The efficiency of the presented system was mainly evaluated with the Pearson correlation coefficient (r), and the determination coefficient (R2). The efficiency of the system was also assessed manually in five images of each variety, from which three measurements were extracted: Counting result (CR), which represents the number of flowers counted by the described system, true positive (VP), identified flowers by the system that correspond to real flowers, false positive (FP), which represents the number of flowers identified by the system that do not correspond to real flowers and false negative (FN) that corresponds to the flowers present in the image that were not detected . Sensitivity was calculated, which is defined as the percentage of real flowers detected [28]:

VP

Sensitivity = (1)

VP + FN and precision, which represents the percentage of flowers detected correctly, and is defined as:

VP

Precision = - (2)

VP + FP

Statistical analyzes were performed using R-project v.3.0.1. (RCore Team: 'R: a language and environment for statistical computing'; 2012, Vienna, R Foundation for Statistical Computing).

The results of the image analysis are consistent for all the images used, with slight variations in the value of the segmentation threshold 200 used to separate the inflorescence 102 from the background 104, due to fluctuations in the illumination during the taking of the photographs.

The effectiveness of the presented system is validated against the number of flowers obtained by re-grazing and manually counting each inflorescence, and also manually counting the number of flowers present in each image on an impression of the same. The developed method was able to distinguish the flowers against other parts of the inflorescence or the background.

Figures 6A, 68, 6C and 60 show the relationship between the total number of flowers present in the inflorescence, those counted manually (or, dashed line) and automatically (e, continuous line) from the photograph of the inflorescence:

Figure 6A: Graciano variety (manual: y = 1.75x + 0.93, R2 = 0.81 at p <0.001;

automatic: y = 2.44x-59.5, R2 = 0.83 at p <0.001).

Figure 68: Mazuelo (manual: y = 2.46x-89.1, R2 = 0.89 at p <0.001; automatic:

y = 2.96x-81.2, R2 = 0.84 at p <0.001).

Figure 6C: Tempranillo (manual: y = 1.56x + 17.1, R2 = 0.87 at p <0.001; automatic:

y = 1.82x + 18.1, R2 = 0.83 at p <0.001).

Figure 60: All varieties (manual: y = 1.68x + 16.2, R2 = 0.81 at p <0.001;

automatic: y = 2.08x + 5.4, R2 = 0.76 at p <0.001).

The relationship between the number of flowers counted manually and automatically from the images, and those counted when shelling the inflorescence, are observed for Graciano (Figure 6A), Mazuelo (Figure 68), Tempranillo (Figure 6C) and the three varieties together (Figure 60). For these three varieties, the coefficient of determination (R2) of the relationship between the estimated number of flowers, both manually and automatically by the method developed, is shown for each of them independently, with the total flowers counted when the inflorescence, being in any case greater than 0.8 (p <0.001). I dont know

R2

they found significant differences in the value of the manual count and the automatic count, although the value of R2 fell to 0.76 when the three varieties were evaluated together with the automatic method.

In contrast to these results, Grossetete et al. [24] obtained a strong non-linear relationship (R2 = 0.92 with a polynomial model) for the estimation of the number of berries per cluster from images taken between the curd (phenological state BBCH 71: the fruits begin to thicken and there are still remains of flowers) and envero (phenological state BBCH

79: fully developed berries). Unlike the invention presented, Grossetete et al. [24] they captured the photographs at night, using a flash built into the camera itself. Its algorithm is based on the reflection of light on the surface of the berry, which has a maximum or specular reflection in the center of it, a unique and distinctive peak, whose value decreases rapidly as it moves away from the center. This methodology cannot be applied to pre-flowering flowers, because the shape of the flower buds is not spherical. The curd of the flowers of the vine does not usually exceed 50% [1], which means that only a proportion of the flowers will eventually become a berry, so the possibility of occlusion of the berries in an image is different from that of flowers. This would explain the best fit to a nonlinear model [24] instead of the linear one presented in this invention, and the differences in R2 values.

The sub-estimation of the flowers present in the inflorescence occurs for all varieties, regardless of the method used (manual or automatic) to count them from the image, but occurs with greater intensity with automatic counting. As the number of flowers is estimated from a single 2D image, the sub-estimation was predictable, since the flowers that are located on the opposite side of the inflorescence will not be visible, neither for the automatic algorithm, nor for the human eye, so they cannot be counted. The sub-estimation of the automatic method is greater, due to its conservative design, since one of the priorities in the development of the algorithm was to minimize the number of false positives, avoiding the counting of what was not a flower. That is why the application of the three consecutive filtrates - size, distance and shape - to the bright areas 302 that represent flowers is very demanding, which eliminates the number of false positives increasing that of true positives.

In contrast to previous implementations [24], artificial light is not necessary for image capture, despite the high variability in the position of the brightness on the black background due to natural light. Another improvement of the developed algorithm is that it is not necessary to fix the size of the flowers, so the distance between the camera and the inflorescence is not predefined, which facilitates the taking of images.

Pre-calibration of the system is not necessary, and the parameters set are valid for any range of distances between the camera and the inflorescence. This is a clear advantage over other systems, in which the threshold values of red, green and blue must be adjusted manually for each image [23].

The low cost, simplicity and ease of use of imaging in the field, and subsequent processing by the described method, for the estimation of the number of flowers per inflorescence in vineyards (for example by the implementation of this method in devices such as "tablets "or" smartphones ") can be of great help to the wine industry and estimate production based on data on the number of flowers per inflorescence.

As a summary, the results of the tests show that the algorithm developed to analyze images taken with a digital camera in uncontrolled field conditions is able to automatically estimate the number of flowers per vine inflorescence in early flowering phases. . This can help those responsible for the vineyards to make decisions with information about the setting rate or production. The developed algorithm can be implemented in a mobile device such as a "smartphone" or even use cloud computing, to show data on the number of georeferenced flowers for each vineyard or even through maps.

The algorithm developed demonstrates that the analysis of images captured with compact cameras, under uncontrolled field conditions, is capable of generating an estimate of the number of flowers per inflorescence of vines in the early stages of flowering.

Claims (19)

1. Procedure for automatically determining the number of flowers of an inflorescence, characterized in that it comprises: -capturing an original image (100) of the inflorescence on a background (104) homogeneous that contrasts with the color of the inflorescence;

-

convert the original image (100) to the CIELAB color space;

-

segment the image converted to CIELAB color space through the application

of a segmentation threshold (200) in the values of the so-called complementary color axes a * or b *, to obtain the pixels representing the inflorescence (202); -filtrate the groups of chained pixels representing the inflorescence (202) based on their situation as intensity values representing local maximums; - group the filtered pixels that are interconnected in different bright areas (302); - make a filtering on the bright areas (302) to eliminate those that do not correspond to flowers, where said filtering includes a filtering by size of the bright area (302) and filtering based on distance with other bright areas (302).

2.

Method according to claim 1, characterized in that the inflorescence is of a vine.

3.

Method according to any of the preceding claims, characterized in that the capture of the original image (100) of the inflorescence is carried out in field conditions with uncontrolled lighting.

Four.

Method according to any of the preceding claims, characterized in that the segmentation threshold (200) is automatically selected for each image according to the groupings of values present in the histogram of the coordinate a * or b * interchangeably.

5.

Method according to claim 4, characterized in that the segmentation threshold (200) is the local minimum value after the global maximum in the histogram of the coordinate a * or b *.

6.

Method according to any of the preceding claims, characterized in that the pixels representing the inflorescence (202) are those whose intensity level in the b * coordinate exceeds the segmentation threshold (200).

7.

Method according to any one of claims 1 to 5, characterized in that the pixels representing the inflorescence (202) are those whose intensity level in the coordinate a * is less than the segmentation threshold (200).

8.

Method according to any of the preceding claims, characterized in that the filtering performed on the bright areas (302) includes a filtering depending on the shape of the bright area (302).

9.

Method according to claim 8, characterized in that the filtering by shape of the bright areas (302) includes the calculation of the relationship between the major axis and the minor axis of each bright area (302).

10.

Method according to any of the preceding claims, characterized in that the filtering of the interconnected groups of pixels representing the inflorescence

(202) is carried out based on its consideration as local maximums in its luminosity values. 11. System to automatically determine the number of flowers of an inflorescence, characterized in that it comprises: - image capture means configured to capture an original image (100) of the inflorescence on a homogeneous background (104) that contrasts with the color of the inflorescence; -data processing means configured for: • convert the original image (100) to the CIELAB color space;

•

segment the image converted to the CIELAB color space by applying a segmentation threshold (200) in the coordinate a * or b *, to obtain pixels representing the inflorescence (202);

•

filter the pixels representing the inflorescence (202) based on their brightness values;

• group the filtered pixels in different bright areas (302) interconnected; 19 • filtering on the bright areas (302) to eliminate those that do not correspond to flowers, where said filtering includes a filtering by size of the bright area (302) and filtering based on the distance with other bright areas (302).

12.

System according to claim 11, characterized in that the data processing means are configured to select the segmentation threshold (200) automatically for each image based on the groupings of values present in the histogram of the coordinate a * or * .

13.

System according to any of claims 11 to 12, characterized in that it is implemented in a mobile device.

14.

System according to any of claims 11 to 13, characterized in that the filtering performed by the data processing means on the bright areas (302) includes filtering according to the shape of the bright area (302).

fifteen.

System according to claim 14, characterized in that the filtering by shape of the bright areas (302) includes the calculation of the relationship between the major axis and the minor axis of each bright area (302).

16.

System according to any of claims 11 to 15, characterized in that the data processing means are configured to filter the interconnected groups of pixels representing the inflorescence (202) based on their consideration as local maximums in its brightness values.

 Fig. 1   ....-- .. ('i ") or T "" " ......... in OR) OR) 0 .. O) "O or "- OR) AND ':: J Z ....-- .. ('i ") or T "" " ......... in O) O) , ~ or ex:> or (OR ..q or N or 0 ex:> or (OR or or .. "<: t O) "O or "- OR) AND ':: J z or N or ....-- .. ('i ") or T "" " ......... in o O) ex:>

-

or

O) (O x 'or. or "<: t OR) "OR or N or " or OR) AND ':: J Z Red Channel TO O 50 100 200 Pixel Intensity L * D O 50 100 200 Pixel intensity L * Filtered G O 50 100 200 Pixel intensity ....-- .. ('i ") or T "" "or ......... 0 in ~ 0) 0 O) ex:> 'o.xo (O O) or "O" <: t or or N "- O) or AND ':: J z ....-- .. ('i ") or T "" "or ......... or in (O O) or

-

or

O) "<: t X 0..0 O) "O or "- OR) AND ':: J z ......... in OR) OR) x o .. O) "OR or "- OR) AND ':: J z or N or Green Channel B O 50 100 200 Pixel Intensity Fig. 2A to* AND O 50 100 200 Pixel Intensity Fig. 28 to * Filtered 80 100 120 140 Pixel Intensity
Fig. 2C ....-- .. ('i ") ~ o in o ~ O) ex:> ~ o

,

(OR

0 ..

0

OR)

..q

"00 or N "-or OR) AND ':: J z ....-- .. ('i ") or T "" "" <: t ......... 0 eno O) ('i ") O) or X or , -N 0 .. 0 O) or ~ o o "- OR) AND ':: J z ....-- .. ('i ") or T "" " ......... in OR) OR) x 'or. O) "O or "- OR) AND ':: J z Blue Channel C O 50 100 200 Pixel intensity b * F O 50 100 200 Intensity of pixel b * Filtered 200 204 202 100 140 180 220
Pixel intensity AC . (,) s :: CI) (,) fn ~ e C "" 'I or �� :: or s :: 0_ ('\ P' \ fn AC 0 "'C tn ca N ..E ~ ~ fn or e \ 1 "') or ....... �� :: CI) or "'C ...... or • .... CI) AND GOAL 009 OO ~ oot OO (OO ~ .-AC (,) s :: CI) (,) .., ~, ,, 8 e I, f) or or 'or • •, �� :: or s ::  or "'' ''" ~~ "'C AC 8 ~ tf "\ fn CI) �e fn '1. "1 �� :: or CI)
"'C 8 e ..... CI) AND . ::: s z nJ . (.) and G) • (.) or •. ~ tn or G) lI "i ... OR � ;: and or ~ tn nJ "OR 8 E nJ t'f '\.-... tn G) tn ~ or G) or ... N OR � ;: G) or "O or .... or ... G) AND 008 009 OOt ool '::: s Z e!:> ua:> saJOllU! �saleaJ saJ011 ap OJawl) N nJ (.) e G) (.) ~ tn and or � ;: e o or tn

-

~ nJ "OR nJ AND  or.-... ~ tn or G) N tn and or
� ;: G) "O ..... o ... G)

-

AND

009 OO ~ C) (tt OO (OOZ '::: s z e!:> ua:> saJOllU! �saleaJ saJ011 ap oJawl) N