Method and imaging system for determining the gender of the embryo of an egg

Abstract

The invention relates to a method for determining the gender of the embryo of an egg (O), the method including steps of: - placing the egg (O) in a reclining position on a support (4), the longitudinal axis (X) of the egg being parallel to the support on which it rests; - using a light source to emit a flux (F) of visible light towards the egg in a direction transverse to the egg (O); - acquiring images of the egg, over at least three discrete spectral bands, referred to as observation spectral bands, comprised between 500 nm and 630 nm, and over a so-called normalisation spectral band comprised between 630 nm and 700 nm; - processing the acquired images to generate a normalised image; and - processing the normalised image using a predictive model (MOD) to determine the gender of the embryo of the egg.

Description

[0001] Description

[0002] Title of the invention: Method and imaging system for determining the gender of an egg embryo

[0003] Technical field of the invention

[0004] The present invention relates to the field of poultry production and in particular to the non-invasive determination of the gender of avian egg embryos during incubation.

[0005] State of the art

[0006] The search for an early and non-invasive sexing technique for poultry eggs (chicken, duck) has been a subject of interest to the poultry industry for several years. Invasive techniques have been proposed in the prior art, notably in patent applications WO98 / 14781, DE102007013102, W02010 / 1031 11, US2011 / 0144473A1, WO2017 / 174337.

[0007] Non-invasive optical techniques have also been proposed. Some of these techniques relate to egg fertility detection rather than sexing. In this regard, US patent 9,435,732 and US patent 6,029,080 are particularly noteworthy.

[0008] Other non-invasive optical techniques have also been proposed, these being based on the exploitation of spectral and / or spatial characteristics, obtained by mono, multi or hyperspectral imaging, or by spectrometry, in wavelength ranges extending from the visible to the near infrared.

[0009] The referenced publication "COR ION MATTHIAS et AL: In ovo sexing of eggs from brown breeds with a gender-specific color using visible-near-infrared spectroscopy: effect of incubation day and measurement configuration", POULTRY SCIENCE, col. 101, no. 5; 1 May 2022 page 101782, XP055962758, Oxford, describes for example a technique for sexing chicken eggs by spectroscopic analysis.

[0010] This earlier document describes a technique for sexing brown hens by estimating the eumelanin content found in high quantities in the feathers of females. The solution is based on spectral data analysis. This approach focuses on identifying the most significant spectral signatures for sex prediction. However, this sex prediction remains delayed, as it is only possible at 14 èmeThe incubation day coincides with the beginning of feather development. The referenced publication, "Nan Jia et al.: Exploratory Study of Sex Identification for Chicken Embryos Based on Blood Vessel Images and Deep Learning," Agriculture 2023, 13(8), 1480, MDPI, presents a technique for sexing white eggs using conventional imaging (RGB camera) with only the green channel for observing the egg's cardiovascular system. The analysis is limited to the spatial information available in this band and cannot utilize any spectral information due to the lack of finer spectral selection. Furthermore, the technique requires drastic egg selection based on size, shape, cleanliness, and shell quality. These sorting conditions are incompatible with the industrial application of the technique.

[0011] Besides the choice of biological markers used for sexing, there are also several configurations for imaging embryos. Candling for sexing generally uses a visible light-emitting diode (LED) that illuminates the egg from the air sac side and a camera positioned at 90° to the egg's axis of symmetry to capture the light scattering image. This configuration is not ideal because the illumination is poorly homogeneous within the egg. Since the light is scattered in all directions, the longer the path of the light through the egg, the greater the light absorption. This results in a light intensity gradient from the air sac to the tip, which is further accentuated as the light must pass through the yolk, which is more absorbent than the white.The strong absorption of yellow at this stage of incubation comes from the development of the cardiovascular system which contains hemoglobin, exhibiting a very strong absorption capacity in the wavelength range of the light-emitting diode illumination between 400nm and 620nm.

[0012] To have a reliable and robust sexing method, several points must be taken into account:

[0013] - The biological variability of eggs. For this, it is important to improve the detection of cardiovascular systems.

[0014] The geometric variability of eggs, the varying dimensions of the eggs, alters the optical absorption properties.

[0015] It is necessary to provide the maximum amount of relevant information to the predictive model used for sexing. During imaging, it is not uncommon for part of the cardiovascular system to be invisible due to incomplete migration during the preparation stage. Rather than discarding these eggs for sexing, a solution should be proposed that allows them to be sexed like the others. Patent application CN115187514A describes a method for determining the sex of an egg embryo. This method mainly consists of:

[0016] Perform a spectral measurement on the egg to be analyzed using a halogen light source in the 400-900nm wavelength band and a fiber optic spectrometer;

[0017] Acquire a color image of the same egg using a 30W white LED source and a color camera. The color temperature of the LED was determined by performing performance tests for detecting the texture of blood vessels between 6000 and 7000K;

[0018] In general, this earlier CN115187514A document presents a method for determining gender that is not deployable in a hatchery because it is too selective in the choice of eggs, requires cleaning of each egg, is complex to implement (requiring a camera, a spectrometer, an LED and a halogen lamp), is very slow since the same egg must be both measured with the spectrometer and pass under the camera, increasing the duration of gender determination (more than 2s per egg).

[0019] Therefore, there is a need for a reliable method in which all steps contribute to determining the sex of the embryo in the egg. The method must be simple and sufficiently rapid to be industrialized, and capable of sexing all types of eggs with varying biological and geometric configurations.

[0020] The process makes it possible, in particular, to characterize an embryo at the earliest possible stage.

[0021] Description of the invention

[0022] This goal is achieved by a method for determining the sex of an egg embryo, said method comprising the following steps:

[0023] - The egg is placed horizontally on a support, with the longitudinal axis of the egg parallel to the support on which it rests.

[0024] Emission, using a light source, of a luminous flux in the visible spectrum, directed towards the egg in a direction transverse to the egg.

[0025] - Acquisition of images of the egg, on at least three distinct spectral bands, called observation spectral bands, between 500nm and 630nm, and on a spectral band called normalization between 630nm and 700nm, Processing of the acquired images to generate a normalized image, Processing of the normalized image using a predictive model to determine the gender of the embryo in the egg.

[0026] In the invention, the choice of the minimum number (three bands) of spectral observation bands and the wavelengths of these spectral observation bands, as well as the choice of the normalization spectral band, are intended to improve the level of prediction. These are not simply arbitrary choices without technical effect.

[0027] However, in the invention, the image normalizations aim to reveal the cardiovascular system with the greatest possible accuracy in the images, which constitutes a real technical effect and allows real advantages to be obtained compared to the solution proposed in the prior document CN115187514A cited above.

[0028] Furthermore, the spatio-spectral information obtained through the implementation of the invention is contained within a single image and not through a combination of spectral and spatial data (as in document CN115187514A). This approach represents a significant departure from the state of the art, as it allows for the determination of the egg's gender in a single measurement (compared to two measurements in the prior document), which is highly advantageous for industrial applications. The method of the invention necessitated the development of a new, specific multispectral imaging tool to collect the relevant information and enables considerably greater information processing efficiency.

[0029] According to a particular feature, image acquisition on a first spectral observation band is implemented on a spectral band centered on 580nm.

[0030] According to another peculiarity, image acquisition on the spectral observation bands is implemented on spectral bands centered on wavelengths equal to 530nm or 550nm, 580nm and 600nm.

[0031] According to another peculiarity, image acquisition on the normalization spectral band is implemented on a spectral band centered on 642nm.

[0032] It should be noted that optimal normalization is indeed obtained at 642nm because it has the double advantage of eliminating the spots on brown eggs related to protoporphyrin and of allowing normalization with a wavelength where all information related to blood has disappeared, revealing the cardiovascular system as best as possible.

[0033] Another distinctive feature is that the acquisition of images of the egg is carried out from several angles around the longitudinal axis of the egg.

[0034] Another unique feature is that images of the egg from multiple angles are acquired by rotating the egg around its longitudinal axis. Another unique feature is that image acquisition is performed using a monochrome camera.

[0035] Another distinctive feature is that the predictive model is developed through learning, and in that it incorporates spectral images of eggs.

[0036] According to another particularity, the predictive model includes metadata, including temporal data, said temporal data being chosen from one or more of the following data: the date of laying of the egg, the date and time of incubation, the date and time of acquisition of each image.

[0037] The invention also relates to a multi-band imaging system used for determining the sex of an egg embryo, said system comprising:

[0038] A suitable support to hold the egg in a horizontal position.

[0039] A controlled light source to emit, towards the egg, a luminous flux in the visible spectrum, following a direction transverse to the egg,

[0040] Means of acquiring images of the egg, on at least three distinct spectral bands, called observation spectral bands, between 500nm and 630nm, and on a spectral band called normalization between 630nm and 700nm,

[0041] - A unit for processing acquired images, configured to generate a normalized image and to process said normalized image using a predictive model in order to determine the gender of the embryo in the egg.

[0042] According to one particular feature, the acquisition means include a camera (C).

[0043] According to another particularity, the acquisition means include a filtering set, adapted to filter the luminous flux emitted by the light source according to the said spectral observation bands and the said spectral normalization band.

[0044] According to another peculiarity, a first spectral band of observation is centered on 580nm.

[0045] According to another peculiarity, the spectral bands of observation are centered on wavelengths equal to 530nm or 550nm, 580nm and 600nm.

[0046] According to another peculiarity, the normalization spectral band is centered on 642nm.

[0047] Brief description of the figures. Other features and advantages will appear in the detailed description that follows, given in relation to the attached drawings in which:

[0048] Figure 1 schematically represents the structure of an egg;

[0049] Figure 2 schematically represents the imaging system of the invention applied to the sexing of an egg;

[0050] Figure 3 shows a diagram illustrating the spectral absorptions of the major chemical elements that make up an egg;

[0051] Figures 4A and 4B show the relative transmission spectra of a white egg and a brown egg respectively;

[0052] - Figure 5 shows a diagram representing the transmission spectra of the main components of a white duck egg;

[0053] - Figure 6 illustrates the principle of spectral normalization with a brown egg, with a first image acquired on a reference spectral band centered on 580nm and normalized images of this reference spectral band by different wavelengths greater than 580nm;

[0054] - Figure 7 illustrates the principle of normalization with a brown egg and its interest, the top images being those acquired at three distinct wavelengths and the bottom images being those obtained after normalization at 596nm and 642nm;

[0055] - Figure 8 shows images obtained after normalization to three wavelengths for white eggs;

[0056] - Figure 9 shows an example of the implementation of a support allowing multi-angle acquisition;

[0057] Detailed description of at least one embodiment

[0058] In the following description and as illustrated by Figure 1, the egg O is defined by its axis of revolution symmetry (axis (X)) in the longitudinal direction and by its ovoid surface. Its equator corresponds to the imaginary line drawn on its surface along its entire circumference, at its widest cross-section.

[0059] The system of the invention is used to determine the sex (male or female) of the embryo in the egg.

[0060] The method and system of the invention are based, in particular, on imaging analysis, and not on spectroscopic analysis. As is known, with reference to Figure 1, the egg O comprises a shell 20, in which is present the embryo 21, to which the eye 22 belongs, and the cardiovascular system 23, containing, in particular, hemoglobin. The egg also comprises an air chamber or sac 24, which is generally located on the opposite side from the tip of the egg. In Figure 1, the egg O is shown horizontally, with the air sac 24 positioned to the side.

[0061] The invention applies more particularly to duck or chicken eggs but is relevant for sexing all types of eggs.

[0062] The method of the invention allows for sexing using an imaging system. According to the invention, sexing is performed non-invasively. Figure 2 illustrates this principle for a single O egg, but it can be replicated for a set of several eggs.

[0063] Referring to Figure 2, the imaging system primarily comprises a light source E capable of generating a luminous flux F and an image sensor, for example, a camera C. The system also includes a processing unit coupled to the camera C to acquire and process the images. The processing unit UC uses a predictive model (referenced MOD) to perform sexing.

[0064] The imaging system is configured to implement:

[0065] A multi-band acquisition with specific spectral filtering (via a filtering set 5) to retain only photons of wavelengths suitable for observing the cardiovascular system and which provide relevant information for determining gender;

[0066] A spectral normalization principle to attenuate, in the image obtained, the elements / artifacts of the egg O disturbing the observation of the cardiovascular system (such as calcium stains on the shell, dirt, feathers, etc.) while standardizing the spectral response (with respect to variable biological / geometric phenomena such as the size of an egg,...);

[0067] Within the scope of the invention, the system can also provide for multi-angle image acquisition for repeating measurements and / or detecting hidden information to improve the accuracy of the sexing model. To this end, means 3 can be provided to act on the support to rotate the egg O around its axis. It would also be possible to move the sensor C. The principles of the invention described below are suitable both for enriching a predictive sexing model and for the sexing operation itself, that is, for determining the sex of the embryo in an egg.

[0068] It should be noted that for the application to in-ovo sexing in ducks and chickens, observation of the cardiovascular system has been estimated to be optimal between 72h (beginning of observation of the development of the cardiovascular system) and up to 120h after the start of incubation (beyond this, the cardiovascular system becomes more complex).

[0069] Multi-band acquisition with spectral filters specific to determining the gender of the egg embryo

[0070] The aim is to identify the wavelengths suitable for determining the gender of the embryo in the O egg, that is to say ultimately the spectral bands of observation relevant for sexing.

[0071] In conjunction with Figure 3, we observe that the strong absorption of the egg yolk at an incubation stage exceeding 3 days results from the development of the cardiovascular system, which contains oxygen-carrying oxyhemoglobin (HbO2) and deoxyhemoglobin (HbR), which no longer carries oxygen. These two compounds exhibit very strong absorption in the visible wavelength range between 300 nm and 630 nm (as illustrated in Figure 3), with particularly strong peaks between 540 and 580 nm.

[0072] Beyond 700nm, water, the main component in an egg, quickly absorbs all the light, making any observation impossible.

[0073] From Figure 4A and Figure 4B, we can also see that light transmission through an egg with its shell only starts beyond 500nm, all wavelengths below 500nm being absorbed by the shell, whether in white or brown eggs.

[0074] The invention aims to use observation wavelengths of the egg within a spectral range between 500 nm and 630 nm (in the diagram in Figure 3, two successive peaks can indeed be observed in this wavelength range). This filtering allows for the specific selection of wavelengths of interest for effective observation of the cardiovascular system through an egg. According to the invention, an acquisition is therefore performed using several filtered wavelengths within the 500 nm to 630 nm wavelength range. With richer spectral information, a signature / fingerprint of the concentration of the different chemical elements constituting the observed egg can thus be obtained.

[0075] This is not an exhaustive list; image acquisition is performed on at least three distinct spectral bands. Generally, the aim is to find a compromise between the complexity and cost of the acquisition system, the accuracy of the model, and the computational speed.

[0076] Even though sexing on a single spectral band would be possible (but less efficient), this joint use of extended spatial and spectral information allows the development of a more accurate and robust predictive model, particularly for in-ovo sexing in ducks and chickens.

[0077] To select the relevant spectral bands for observation, one can take into account the absorption level of the main components of the O egg. From Figure 5, it can be seen that absorption increases with blood concentration, which explains a lower transmission level at the level of the embryo (heart) compared to the transmission level of the blood vessels.

[0078] By examining each component of the egg independently, we observe that greater accuracy in sex determination is obtained using the transmission spectra of the embryo after 3 days of incubation, compared to the yolk and blood vessels. This demonstrates that sex information, at this stage, is primarily located in the blood, which has the highest concentration in the embryo. The spectra of the yolk and blood vessels also carry sex information, but it is less pronounced than that found in the embryo.

[0079] If we take, nevertheless, the spectra of the embryo with those relating to the yolk between 480 and 680 nm, we obtain a high sexing accuracy which varies little with the dataset.

[0080] The table below shows that sexing accuracy, using a simple spectral analysis-based algorithm, achieves the highest values ​​when using transmission data between the 576 nm and 587 nm spectral bands. This wavelength band is particularly relevant for sexing eggs. Below 530 nm or above 620 nm, spectral information is more limited for accurately sexing eggs.

[0081] Acquisitions using several combinations were also performed, each combination based on a core wavelength of 587 nm and other complementary wavelengths. For example, by choosing one acquisition at a wavelength of 587 nm and another at a different wavelength such as 533 nm or 555 nm, a precision level of 80% is obtained, very close to that obtained by performing acquisitions at twenty distinct wavelengths. In other words, it is possible to limit the selection to a few relevant wavelengths to obtain a sufficient level of precision.

[0082] For example, we can limit the acquisition to three spectral observation bands containing the wavelengths 533nm, 555nm and 587nm with a bandwidth of 10nm.

[0083] Spectral normalization

[0084] In addition to image acquisitions on the spectral observation bands, according to the principles described above, the invention also consists of carrying out at least one image acquisition on at least one other spectral band, called the normalization spectral band.

[0085] This spectral band is used to normalize the bands containing the cardiovascular system. Normalization reduces the presence of artifacts (such as shell stains, dirt, etc.) and also standardizes measurements between eggs of different sizes (and therefore with different absorption levels). Indeed, the spectral absorption of an element within the egg is proportional to the volume of light passing through it. However, the volume ratios (and therefore the absorption ratios) of the egg's constituent elements remain independent of size. For example, if the overall size of the egg is twice that of the egg, the yolk and white are also twice that of the egg and absorb twice as much light. Nevertheless, the volume ratios, and therefore the absorption ratios, remain unchanged.Normalization is thus implemented to normalize the images acquired on the selected spectral observation bands. Once this normalization process is completed, sexing is performed based on the normalized image, from which the defects described above have been eliminated, and the MOD predictive model.

[0086] Figure 6 shows examples of possible spectral normalizations for a brown egg, with an observation spectral band at 580 nm and various normalization bands (from 593 nm to 808 nm). It can be seen that in brown eggs, not all spectral normalization bands are equally relevant, both in terms of contrast between the different components and in terms of their ability to correct shell defects (speckles). Two spectral normalization bands are particularly relevant: the one between 593 and 604 nm and the one between 639 and 647 nm.

[0087] The second spectral normalization band around 642 nm is the most relevant, as it provides both excellent contrast and very good image resolution. At the 642 nm wavelength, there is no longer any signal related to blood absorption, unlike around 596 nm. Figure 7 clearly illustrates this phenomenon, where the embryo's heart is still visible at 596 nm. We also confirm that the normalization quality is better at 642 nm than at 596 nm, with improved suppression of protoporphyrin speckles.

[0088] The most relevant spectral normalization bands are those corresponding to the absorption of protoporphyrin pigments present in the shell, namely 539, 589, and 642 nm. The 642 nm normalization band proves to be the most effective, as it is where protoporphyrin absorption is strongest. Figure 4B clearly illustrates a reduction in transmittance around 642 nm due to protoporphyrin absorption in brown eggs.

[0089] The principle of normalization is therefore particularly relevant in the characterization of brown eggs, due to their color and the presence of speckles linked to the presence of protoporphyrin. In the case of white eggs, normalization can be performed by choosing a spectral band that distinguishes only the presence of speckles. Numerous normalization bands are thus possible with the same quality (see Figure 8). To obtain a single normalization band for both white and brown eggs, the 642 nm spectral band is preferentially chosen.

[0090] Artifact removal and measurement standardization through spectral normalization improve the visualization of the cardiovascular system in the egg, thereby increasing the accuracy of the predictive model. Several wavelengths can be used for this normalization, but a wavelength that no longer carries information about the cardiovascular system is preferred; therefore, a wavelength greater than 630 nm, and ideally 642 nm, which is compatible with both white and brown eggs.

[0091] In summary, the removal of artifacts and the standardization of measurement by spectral normalization improves the visualization of the cardiovascular system in the egg and thus increases the accuracy of the models.

[0092] Multi-angle acquisition

[0093] Optionally, but still advantageously, it is possible to acquire images of the egg from several distinct angles. Ideally, these acquisitions are performed at multiple angles around the egg's longitudinal axis, either by rotating the egg around its axis (the simplest solution) and / or by rotating the acquisition system (source + camera). This multi-angle acquisition allows for multiple measurements on the same egg, thus improving prediction using the predictive model. It also integrates additional information on the development of the cardiovascular system by making visible portions of the system that might otherwise have remained hidden. When rotating the egg, a low-acceleration movement is preferable to maintain the cardiovascular system's position on the top of the egg when it is placed flat.

[0094] The rotation of the egg around its axis can be implemented using a diabolo 40 type support (see figure 9).

[0095] Functional imaging system

[0096] The general principle of the multi-band imaging system for visualizing the cardiovascular system for in-ovo sexing in ducks and chickens is shown in Figure 2 and has already been mentioned above.

[0097] As mentioned above, the multi-band imaging system consists of... Error!

[0098] Source of the reference not found, primarily from:

[0099] A light source E emitting a luminous flux F, A support for the egg 4,

[0100] A multi-band imaging device, A processing unit CU.

[0101] light source

[0102] The light source E, for example, consists of one or more light-emitting diodes (LEDs) positioned to illuminate the egg O. Advantageously, the light source E is positioned to illuminate the egg O from below, and the image acquisition device C is advantageously placed above it, along the same axis, to perform transmission acquisition. In this case, it should be noted that the support 4 of the egg O must be chosen to allow the passage of the luminous flux F emitted by the light source. It can thus be chosen to be transparent, allowing, for example, at least 90% of the luminous flux F to pass through, with minimal diffusion, and / or, for example, with an opening through which the luminous flux is emitted, and / or have at least one grid-shaped section designed to allow the luminous flux to pass through.

[0103] The light source is chosen to emit in the visible spectrum with a fairly high power in the wavelength range of 500nm to 700nm.

[0104] The light source, while not limited to standard applications, consists of a light-emitting diode (LED) with a color temperature between 2700K and 3000K. It has been observed that maximizing emission around the 600nm spectral band is more effective in achieving maximum transmission of light from eggs. This approach, which is advantageous, uses only a single LED and therefore a single type of lighting (rather than a combination of several methods), resulting in a particularly simple and efficient solution.

[0105] Egg stand

[0106] A support 4 is chosen to place the egg O, advantageously in a lying position.

[0107] It should be noted that placing the O egg in a horizontal position optimizes lighting and provides the best possible positioning of the embryo and its cardiovascular system relative to the camera, as the cardiovascular system is generally located near the highest point of the egg. In this configuration, the light captured by the camera positioned above is very homogeneous because all the rays that reach it have passed through a roughly equivalent portion of the egg (adjusting for the egg's geometry). All the rays also pass through the same components of the egg (albumen, yolk, shell).

[0108] If necessary, the support 4 can integrate means 3 for moving (rotating, translating) the egg for repeating the measurement or extending the measurement area.

[0109] When acquiring images from multiple angles (see above), the support can be chosen to allow the egg to rotate around its longitudinal axis. This multi-angle acquisition compensates for certain variations in the positioning of the cardiovascular system, thus increasing the number of treatable eggs. It would also be possible to incorporate a translational movement of the egg along its longitudinal axis.

[0110] To rotate the egg around its axis, a support in the form of two 40 diabolo skewers (figure 9) can be used, allowing the rotation to be carried out while keeping a space to ensure lighting from below.

[0111] Multi-band imaging device

[0112] In the system, a multi-band imaging device allows image acquisition over one or more spectral bands within the wavelength range of 500 nm to 700 nm (500 nm to 630 nm for the spectral bands used to observe the cardiovascular system and 630 to 700 nm for the normalization spectral band). The device can advantageously consist of a conventional monochrome camera C (sensitive in the 400 to 1000 nm range) and a filter wheel forming the filtering assembly 5 described above. The number of filters depends on the number of spectral bands used for observation and acquisition. The filter wheel is rotated so that one of its filters is positioned between the light source E and the camera C to filter the luminous flux F emitted by the light source E onto the desired spectral band.The multispectral imaging approach using a monochrome camera coupled with specific filters makes it possible to eliminate useless information and capture only the information of interest.

[0113] It should be noted that it can be beneficial to use other parameters during image acquisition to enrich the predictive model and thus improve the accuracy of subsequent sexing. In particular, it is important to consider the egg's temporal data: laying date, incubation date and time, and the date and time of each image acquisition (and potentially tracking the egg with its history / origin and other metadata).

[0114] This temporal metadata allows for better control over image acquisition at the same developmental stage across different eggs, reducing observation variability and improving the accuracy of the predictive model. Furthermore, in addition to image data, this metadata can complement the predictive model to increase its accuracy. For example, in ducks and chickens, metadata (laying date, incubation date and time, acquisition date and time) resulted in an almost 5% improvement in accuracy.

[0115] According to an advantageous configuration, the different aspects of the system and the process are as follows:

[0116] Filtering around the wavelength of 580nm is optimal for observing the cardiovascular system by combining maximum absorption of the cardiovascular system with good transmission of light flux through the egg;

[0117] - Other spectral observation bands are chosen in the 500-630nm range to obtain spectral signatures of the different egg components (blood, yolk, or white). For sufficient spectral richness in the 500-630nm range, we recommend selecting one or two filters before 580nm, one filter at 580nm, and one filter after 580nm.

[0118] Spectral normalization of images optimizes the detectability of the cardiovascular system. The best normalization bands produce very high-quality images in both ducks and chickens. These optimal spectral normalization bands are the protoporphyrin absorption bands, specifically 598 nm and 642 nm. The 642 nm spectral normalization band is preferred because it contains no blood absorption information and corresponds to the maximum protoporphyrin absorption, thus best correcting any defects related to this molecule.

[0119] - An advantageous combination is formed of 10nm bandwidth filters: Spectral observation bands centered on 530nm, 550nm, 580nm and 600nm and normalization spectral band centered on 642nm.

[0120] - Optional multi-angle acquisition. - Temporal tracking using metadata such as laying date, incubation date and time, and acquisition date and time. This allows for acquisition of images from eggs at the same stage of cardiovascular system development and the creation of more effective multimodal models (integrating images and metadata from each egg).

[0121] Operating principle

[0122] The principle of the invention is thus:

[0123] Place the egg O on its support 4 advantageously in a horizontal position to allow the migration of the cardiovascular system towards the upper part of the egg.

[0124] Perform multiple image acquisitions of egg O using the multi-band imaging system, on at least three distinct spectral bands, called observation spectral bands, between 500nm and 630nm, and on a spectral band called normalization between 630nm and 700nm,

[0125] The observation spectral bands are chosen centered on 530nm, 550nm, 580nm and 600nm and the normalization spectral band is chosen centered on 642nm,

[0126] - Processing of acquired images to generate a normalized image,

[0127] - Processing of the normalized image using the predictive model (MOD) to determine the gender of the egg embryo.

[0128] The MOD predictive model can be trained on images acquired by the multi-band imaging device described above, possibly in connection with the metadata, including temporal metadata, defined above.

[0129] Thanks to the principles of the invention, the quality (of the spatial / spectral information) of the images obtained from the cardiovascular system has enabled the implementation of high-performance models for sexing and in-ovo analysis of duck and chicken eggs, allowing sexing with accuracy levels exceeding 95%.

Claims

DEMANDS 1. A method for determining the sex of the embryo in an egg (O), said method being characterized in that it comprises the steps of: - The egg (O) is placed in a horizontal position on a support (4), with the longitudinal axis (X) of the egg parallel to said support on which it rests. Emission, using a light source, of a luminous flux (F) in the visible range, directed towards the egg, along a direction transverse to the egg (O), - Acquisition of images of the egg, on at least three distinct spectral bands, called observation spectral bands, between 500nm and 630nm, and on a spectral band called normalization between 630nm and 700nm, - Processing of acquired images to generate a normalized image, - Processing of the normalized image using a predictive model (MOD) to determine the gender of the egg embryo.

2. Method according to claim 1, characterized in that the acquisition of images on a first spectral observation band is carried out on a spectral band centered on 580nm.

3. Method according to claim 1 or 2, characterized in that the acquisition of images on the spectral observation bands is carried out on the spectral bands centered on wavelengths equal to 530nm or 550nm, 580nm and 600nm.

4. A method according to any one of claims 1 to 3, characterized in that image acquisition on the normalization spectral band is implemented on a spectral band centered on 642nm.

5. A method according to any one of claims 1 to 4, characterized in that the acquisition of images of the egg is carried out at several angles around the longitudinal axis of the egg.

6. Method according to claim 5, characterized in that the acquisition of images of the egg from several angles is carried out by rotating the egg (O) on itself around its longitudinal axis (X).

7. A method according to any one of claims 1 to 6, characterized in that image acquisition is carried out using a monochrome camera (C).

8. A method according to any one of claims 1 to 7, characterized in that the predictive model (MOD) is developed by learning, and in that it integrates spectral images of eggs.

9. A method according to any one of claims 1 to 8, characterized in that the predictive model includes metadata, including temporal data, said temporal data being chosen from one or more of the following data: the date of laying of the egg, the date and time of incubation, the date and time of acquisition of each image.

10. A multi-band imaging system used for determining the gender of an egg embryo (O), said system comprising: A support (4) adapted to receive the egg (O) in a horizontal position, A light source (E) controlled to emit, towards the egg, a luminous flux (F) in the visible range, following a direction transverse to the egg (O), Means of acquiring images of the egg, on at least three distinct spectral bands, called observation spectral bands, between 500nm and 630nm, and on a spectral band called normalization between 630nm and 700nm, An acquired image processing unit, configured to generate a normalized image and to process said normalized image using a predictive model (MOD) in order to determine the gender of the egg embryo. 1 1. System according to claim 10, characterized in that the acquisition means comprise a camera (C).

12. System according to claim 10 or 11, characterized in that the acquisition means comprise a filtering assembly, adapted to filter the luminous flux emitted by the light source according to said spectral observation bands and said spectral normalization band.

13. System according to any one of claims 10 to 12, characterized in that a first spectral band of observation is centered on 580nm.

14. System according to claim 13, characterized in that the spectral observation bands are centered on wavelengths equal to 530nm or 550nm, 580nm and 600nm.

15. System according to any one of claims 10 to 14, characterized in that the normalization spectral band is centered on 642nm.

Images