Method and apparatus for selecting, detecting, counting and identifying pollens and / or molds initially suspended in atmospheric air.

Abstract

Method and apparatus for selecting, detecting, counting and identifying pollen and / or mold initially suspended in atmospheric air. The invention essentially consists of an apparatus which allows the suction of air, the concentration of pollen and / or mold particles suspended in the air, the detection of the passage of each pollen and / or mold particle, the triggering of the taking of an image of the pollen and / or mold with a lensless microscope, a counting of the number of pollen and / or molds and an identification of the pollen and / or molds thanks to the analysis of the dynamic images acquired. . Figure for abstract: Fig. 1

Description

Title of the invention: Method and apparatus for selecting, detecting, counting and identifying pollens and / or molds initially suspended in atmospheric air. Technical field

[0001] The present invention relates to the field of detection and identification of pollens and / or molds likely to be present in suspension in atmospheric air.

[0002] By "pollen" is meant here and within the scope of the invention, any grain or particle of microscopic size produced by the stamens and which represents an element of flowering plants. By "mold" is meant here and within the scope of the invention, any spore from a biological mold consisting of a fungus of microscopic size which grows in a form of multicellular or unicellular filaments.

[0003] The term “grain” or “particle” is used hereinafter interchangeably to designate pollen or a mold spore.

[0004] The invention aims firstly to propose an apparatus which makes it possible to detect, count and identify pollens and / or molds present in the atmospheric air, whatever their type and in real time, that is to say an identification without waiting, immediate.

[0005] Although described with reference to pollens, the invention can just as easily be implemented for the selection, detection, counting and identification of molds. Prior art

[0006] Since the 1970s, awareness of the environmental and health effects caused by aerosols has led to new technological developments to better assess the associated risks.

[0007] Wind-borne particles of biological origin, such as pollen and mold, are present in the atmospheric air. Depending on their allergenic potential, these particles can cause allergy symptoms in sensitive individuals. In Europe, it is estimated that 20% of the population suffers from pollen and / or mold allergies. Seasonal pollen allergies are therefore a public health problem.

[0008] By having reliable, real-time information on the pollens present in the atmospheric air, it would be possible to better adapt the treatments of people suffering from related allergies.

[0009] The measurement of pollen particles in atmospheric air is a very specific measurement. particular in their sizes (diameter generally of the order of 20 to 40 μm), their morphologies (very varied shapes) and their low concentrations in the air relative to finer particles, in particular particles called PM2.5, whose aerodynamic diameter is less than 2.5 μm.

[0010] Typically, on this last point, the concentration of pollens in the atmospheric air can be of the order of 500 pollens / m3 at the height of the pollen season, while that of PM2.5 is 106 / m3.

[0011] There are currently several methods and related systems for identifying pollen grains.

[0012] For many years, the reference method for measuring concentrations of biological particles has been the so-called Hirst method: volumetric sensors continuously draw in a flow of air, the drawn-in biological particles, including pollen, impacting a strip in the form of a rotating adhesive disc. Collection takes an average of 1 week. The adhesive disc is then recovered by an operator and identification is carried out exclusively by observation under an optical microscope by an operator trained in pollen recognition. There is therefore an average delay of 10 days to obtain the pollen concentration at a given time. This Hirst method therefore suffers from too long a delay and requires the use of an experienced technician, which increases costs and potentially introduces measurement errors.

[0013] Other methods and related systems allowing automated measurements for the purposes of counting and identifying pollen with much shorter response times are currently emerging.

[0014] The existing solutions for counting and identifying pollen can be classified into three main categories.

[0015] The first solution essentially consists of sucking in the ambient air, collecting pollen particles by impact on an adhesive disc and then analyzing images of this disc, obtained by optical microscope and recognition by automatic image processing based on shape recognition.

[0016] For example, we can cite the pollen detector marketed under the name “BAA500” by the company Hund, which includes a camera with a CCD sensor (acronym for “Charge Coupled Device”) and an image processing unit which extracts the morphological parameters of each particle collected. This detector can analyze up to one disc per hour.

[0017] The second solution also consists of sucking in the ambient air and chemically characterizing the sucked-in particles using Raman spectrometric techniques and Fourier Transform Infrared Spectroscopy or IRTF spectroscopy (an English acronym FTIR for “Fourier Transform InfraRed spectroscopy”).

[0018] For example, we can cite the device marketed under the name "Rapid-E" by the company PLAIR, in which the aspirated particles pass one by one in front of two lasers, one infrared and one ultraviolet. Infrared illumination makes it possible to study elastic diffusion, which can vary between two species of pollen. Ultraviolet allows the excitation of endogenous fluorescent elements. A spectrum is measured as well as a measurement of the fluorescence lifetime.

[0019] The device marketed under the name "KH300" by the company SHINYEI, is similar but exclusively intended for the detection of Japanese cedar pollen: it comprises a laser in front of which each aspirated pollen particle passes. The intensity of the light diffused by it is measured by two photodetectors arranged at two different angles. The ratio of the measured intensities makes it possible to differentiate cedar pollen from other particles.

[0020] Patent application WO 2021 / 136889 A1 discloses a method consisting of sucking in ambient air and then measuring, from different angles, using photodetectors, the intensity of light scattering generated by the passage of a sucked-in pollen particle in a light beam. For a given angle, the scattering is dominated by certain characteristics of the particle such as its size, its refractive index, its morphology, etc. A comparison of the ratios of measured values between them with a database allows the discrimination of different pollen species.

[0021] Finally, pollens can be identified by biomolecular characterization and identification of DNA sequences. However, this approach is more complex than the purely optical analyses mentioned above in that the DNA of the pollens must be extracted and then mixed with reagents, which are expensive and stored at low temperature, in very precise quantities and then heated for at least 10 minutes. These characterizations are therefore rather carried out in the laboratory by qualified personnel.

[0022] All the solutions mentioned require the use of expensive devices. Either the samples are analyzed in the laboratory, reducing the sampling frequency, or the devices are installed outdoors in very large and very expensive boxes.

[0023] There is therefore a need to improve existing methods and related devices for automatically detecting, counting and identifying pollens and / or molds present in atmospheric air, whatever their type, in real time and at lower cost.

[0024] The aim of the invention is to meet at least part of this need. Statement of the invention

[0025] To this end, the invention relates, in one of its aspects, to an apparatus for selecting, detecting, counting and identifying pollens and / or molds initially in suspension in atmospheric air comprising:

[0026] - a conduit for circulating a flow of atmospheric air capable of containing pollen particles from an air inlet to an air outlet;

[0027] - a device for concentrating pollen particles, connected to the air inlet and adapted to concentrate particles present in atmospheric air and allow pollen particles in a selected size range to circulate downstream in the duct;

[0028] - a pollen particle detection device, arranged around the duct and in downstream of the concentration device, comprising:

[0029] a laser illumination source(s), adapted to illuminate the air flow circulating in the duct,

[0030] at least one photodetector adapted to detect by diffusion or by fluorescence the passage of each particle of pollen and / or mold in the air flow illuminated by the illumination source; - a lensless microscope, arranged around the conduit and downstream of the detection device, comprising:

[0031] a diode illumination source, adapted to illuminate the air flow circulating in the duct,

[0032] at least one CMOS type image sensor, arranged opposite the diode source(s) and connected to the photodetector so as to trigger the taking of a number N of holographic images produced by the light diffracted by each illuminated and detected pollen and / or mold particle, the image sensor being adapted to record the N holographic images;

[0033] - an image processing unit, connected to the CMOS image sensor, adapted for respectively digitally reconstruct the N images, select from the N reconstructed images the one on which each detected pollen and / or mold particle is present and then identify, from the selected image, each pollen and / or mold particle according to its size and / or morphology.

[0034] According to an advantageous embodiment, the particle concentration device is a virtual impactor.

[0035] As a reminder, an impactor works by accelerating particles suspended in the air using a calibrated nozzle to project them onto a collection plate placed opposite the air jet. The largest particles, whose inertia exceeds a certain threshold value, cannot follow the air streams and impact the collection plate. The finer particles bypass the plate by following the air flow.

[0036] A virtual impactor implements the same principle but without a solid collection plate. This solid plate is generally replaced by a calibrated tube in which the flow rate is lower than the incoming flow rate. According to a variant, air can be injected at counter-current to the incoming flow.

[0037] Preferably, the virtual impactor is adapted to reduce the air flow in the duct between the air inlet and the air outlet by a factor of at least 5, preferably of the order of 10.

[0038] More preferably, the virtual impactor is adapted to allow particles larger than 5 μm to circulate downstream in the conduit.

[0039] According to an advantageous characteristic, the laser(s) of the illumination source of the detection device is (are) emitted in the visible range, for example in the red range.

[0040] According to an advantageous embodiment variant, the photodetector(s) of the detection device is(are) one or more photodiodes, opposite which (each of which) is arranged a parabolic mirror.

[0041] According to another advantageous embodiment variant, the diode(s) of the illumination source of the lensless microscope is(are) one or more blue-emitting light-emitting diode(s).

[0042] Advantageously, the CMOS image sensor(s) of the lensless microscope is(are) integrated into a high-speed camera or a camera adapted to carry out multi-acquisition of images.

[0043] According to an advantageous embodiment, the processing unit is further adapted to merge the data of the N holographic images with the diffusion intensity signals of each particle provided by the photodetector(s).

[0044] According to another advantageous embodiment, the apparatus comprises a portable housing housing at least the conduit, the device for concentrating pollen and / or mold particles, the device for detecting pollen and / or mold particles, and the lensless microscope.

[0045] The invention also relates to a method for selecting, detecting, counting and identifying pollens and / or molds initially suspended in atmospheric air, in particular implemented by the apparatus which has just been described, comprising the following steps:

[0046] a / suction in a conduit, preferably arranged vertically, from an inlet to an outlet of a flow of atmospheric air likely to contain pollen and / or mold;

[0047] b / selection of a given range of pollen particles from among the particles sucked into the air flow;

[0048] c / illumination of the air flow and detection by diffusion or fluorescence of the passage of each selected pollen and / or mold particle;

[0049] d / recording of a number N of holographic images produced by the light diffracted by each illuminated and detected pollen and / or mold particle;

[0050] e / digital reconstruction of the N images, selection from the N reconstructed images of the one on which each detected pollen and / or mold particle is present, then identification, from the selected image, of each pollen and / or mold particle according to its size and / or morphology.

[0051] Preferably, the detection according to step c / is carried out at 90° to the light source illuminating the air flow.

[0052] More preferably, the detection according to step c / makes it possible to trigger the acquisition of the N holographic images.

[0053] According to an advantageous variant, step e / comprises the merging of the data from the N holographic images with the diffusion intensity signals of each particle provided by the photodetector(s).

[0054] According to an advantageous embodiment, the identification of the image selected according to step e / is carried out by a deep learning device, preferably a convolutional neural network.

[0055] According to this mode and an advantageous variant, step e / comprises the following steps:

[0056] el / creation of a learning base comprising images of pollens and / or molds, or “known images”, each known image being that of a known pollen and / or mold;

[0057] e2 / training the deep learning device, using the base learning;

[0058] e3 / submission of the selected image to said at least one learning device deep so that it determines at least one probability relative to the selected image;

[0059] e4 / determination, as a function of said probability, of the belonging of the pollen and / or from the mold in the selected image to known pollen and / or mold.

[0060] Thus, the invention essentially consists of an apparatus which allows the suction of air, the concentration of pollen and / or mold particles suspended in the air, the detection of the passage of each pollen and / or mold particle, the triggering of the taking of an image of the pollen and / or mold with a lensless microscope, a counting of the number of pollens and / or molds and an identification of the pollens and / or molds thanks to the analysis of the dynamic images acquired.

[0061] Ultimately, the invention provides numerous advantages, among which we can cite:

[0062] - due to the combination of its technological components (virtual impactor, photodiode detection, lensless microscope), the device according to the invention overcomes the various limitations (response time, automation, cost, size) of the systems according to the state of the art, making possible the continuous measurement of the concentration of the different families of pollen and / or molds (counting and identification);

[0063] - thereby, responding to the major public health problem that is seasonal allergies- pollen and / or mold infestations.

[0064] Other advantages and characteristics of the invention will become more apparent upon reading the detailed description of examples of implementation of the invention given by way of illustration and not limitation with reference to the following figures. Brief description of the drawings

[0065] [Fig-1] [Fig.l] is a schematic view of a selection, detection, counting and identification of pollens and / or molds initially suspended in atmospheric air, according to the invention.

[0066] [Fig.2] [Fig.2] is a schematic view showing in more detail the part of detection, counting and identification of the device according to [Fig.l].

[0067] [Fig.3] [Fig.3] illustrates the scattering signal detected by a photodetector of the detection part of the device, with peaks corresponding to the passage of a grain of pollen.

[0068] [Fig.4] [Fig.4] illustrates the conversion of the analog peaks of the detected raw signal from [Fig.3] into logic signals.

[0069] [Fig.5] [Fig.5] illustrates an example of a raw holographic image of a grain of pollen data recorded by the device's camera.

[0070] [Fig.6] [Fig.6] illustrates an example of an image reconstructed by the processing unit of device images from the raw holographic image in [Fig.5]. Detailed description

[0071] In the description which follows as well as throughout the application, the terms "inlet", "outlet", "upstream", "downstream", are used with reference to the direction of circulation of the flow of atmospheric air containing the pollens and / or molds in the apparatus according to the present invention.

[0072] Figures 1 and 2 show an apparatus 1 according to the invention with the pollen detection that it carries out.

[0073] It firstly comprises a conduit 10 for circulating a flow of atmospheric air likely to contain pollen particles from an air inlet 11 to an air outlet 12. This conduit 10 must be transparent at least to the wavelengths of the illumination sources of the apparatus as described below. The material of the conduit 10 may be quartz, glass, or any optical quality plastic.

[0074] By way of example, the conduit 10 is a straight tube, of rectangular section, of wall thickness of the order of 1 mm and of internal section equal to 7.2x2 mm.

[0075] A virtual impactor 2 is arranged downstream of the air inlet 11. It makes it possible both to reduce the air flow rate by a factor of 10 and to reduce the concentration in the detection system of undesired particles p of a size less than 5 pm. The virtual impactor preferentially allows the particles P of size greater than 5pm among which are the particles to be analyzed. For example, the virtual impactor 2 can be the device marketed under the name “PCVI 8100” by the company Brechtel.

[0076] A pollen particle detection device 3 is arranged around the duct 10 and downstream of the virtual imager 2. This device 3 comprises a red laser 30 (670nm) which is adapted to illuminate the air flow Q circulating in the duct 10, and a photodiode 31 adapted to detect by diffusion or by fluorescence the passage of each pollen particle in the air flow illuminated by the illumination source. In order to maximize the quantity of light detected by the photodiode 31, a parabolic mirror is advantageously arranged on the other side of the duct 10 symmetrically to the photodiode 31. Preferably, the photodiode 31 used has a large surface area, typically of the order of 1cm2.

[0077] A lensless microscope 4 is arranged around the duct 10 and downstream of the detection device 3. This lensless microscope 4 comprises a diode 40 adapted to illuminate the air flow Q circulating in the duct 10, and a CMOS type image sensor 41, arranged opposite the diode 40 and connected to the photodiode 31 so as to trigger the taking of a number N of holographic images produced by the light diffracted by each illuminated and detected pollen particle, the image sensor being adapted to record the N holographic images. The CMOS image sensor 41 is advantageously integrated into a camera.

[0078] Given the anticipated speed of the pollen in the duct, typically of the order of m / s, a high-speed camera is advantageous in order to avoid blurring of the images produced, which would be linked to the movement of each particle. It is also possible to use a camera in multi-acquisition mode. The camera may be a CMOS camera marketed under the name “UL3250CP Monochrome 1600x 1200 pixels” by the company IDS. Such a multi-acquisition mode makes it possible to have several views of a moving object on a single image, without blurring associated with the movement.

[0079] For example, the diode 40 is a fiber LED with a wavelength of 470nm. The diode 40 is arranged at a distance of between 4 and 12 cm from the conduit 10. The camera integrating the CMOS sensor 41 is glued against the conduit 10. Thus, due to the wall thickness of the latter, the pollen grains which circulate in the conduit 10 are 1 to 3 mm away from the sensor 4L.

[0080] Finally, the system comprises an image processing unit 5 connected to the CMOS image sensor 4L

[0081] According to the invention, as detailed below, this unit 5 is adapted to respectively digitally reconstruct the N holographic images, select from the N reconstructed images the one on which each pollen particle is present detected and then identify, from the selected image, each pollen particle according to its size and / or morphology. For the digital reconstruction of holographic images, reference may be made to algorithms described in patent applications WO2017 / 162985 (steps 100 to 170) or WO2016 / 189257 (steps 100 to 500).

[0082] All the components 10, 2, 3, 4 or even the unit 5 which are compact can be housed in the same portable case 100. It is possible to envisage a remote connection by wireless communication protocol, of the type between the components 10, 2, 3, 4 and the unit 5. It is also possible to envisage integrating a microprocessor into the case 1000 in place of a separate unit 5.

[0083] An example of operation of the device is now described.

[0084] First of all, it is specified that the longitudinal axis X of the conduit is vertical. This makes it possible to reduce the phenomena of particle sedimentation and a continuous flow.

[0085] Step a / : a suction is carried out in the conduit 10 of the particles present in the atmosphere at a flow rate of the order of 100 / min.

[0086] Step b / : when passing through the virtual impactor 2, the air flow is reduced by a factor of 10 to reach IL / min in the channel 10 leading to the detector and the particles p of a size less than 5 pm are evacuated in the air flow of 9 L / min. The pollens P of a size greater than approximately 5 pm then circulate downstream in the conduit 10.

[0087] Step c / : the air flow is illuminated by the laser 30 according to a beam Fl and the photodiode 31 detects by lateral diffusion or by fluorescence the passage of each selected pollen particle.

[0088] More precisely, the light scattered by the pollen grains at 90° relative to the illumination axis is measured by the photodiode 31. The analog signal provided by this photodiode 31 indicates the variation of light within the conduit in which the pollen passes. The passage of a pollen induces a strong increase in the scattering signal and therefore a peak in the signal detected by the photodiode. An example of the raw signal obtained S is shown in [Fig.3].

[0089] A digital algorithm analyzes in real time the raw signal provided by the photodiode to detect peaks indicating the passage of particles in the conduit 10.

[0090] This may be a Z-score type algorithm measuring the deviation of a point from the mean in terms of standard deviation, according to the following formula:

[0091] A - (y

[0092] where X is the value of the point studied, p is the mean of the signal, o is the standard deviation of the group and Z is the score obtained. If Z exceeds a certain chosen threshold, we consider that the point corresponds to the passage of a particle in the conduit. For example, the moving average from the peaks in [Fig.3] is calculated over 32 points and the threshold is 2 (au).

[0093] A logic signal is then generated for each detected peak, after a certain delay corresponding to a chosen value less than the average flight time of the particles in the conduit 10 between the detection device 3 and the lensless microscope 4. This delay is determined experimentally. An example of conversion of analog peaks into logic signals is shown in [Fig.4] in which the top signal is the raw signal from the photodiode 31 and the bottom signal is the logic signal emitted for each peak.

[0094] The analog signal is then sent to the camera integrating the CMOS sensor 41 in order to trigger the taking of a sequence of N images to ensure that the images are taken at the moment the particle passes in front of the camera. This triggering makes it possible to reduce the quantity of images to be taken and processed compared to continuous image taking. In other words, the detection according to step c / makes it possible to trigger the acquisition of the N holographic images downstream in the conduit 10.

[0095] Step d / : When the camera is triggered, an image acquisition is made with the lensless microscope 4. Thus the sensor 41 then records a number N of holographic images produced by the light diffracted by each pollen particle illuminated by the beam F2 of the diode 40.

[0096] An example of a holographic image is given in [Fig.5] with a zone of interest ZI of the recorded pollen. In this example the pollen was seen 5 times by the CMOS imager 41 in multi-acquisition mode.

[0097] Step e / : the processing unit 5 then proceeds to the digital reconstruction of the N images, then to a selection from among the N reconstructed images of the one on which each detected pollen particle is present and finally to an identification, from the selected image, of each pollen particle according to its size and / or its morphology.

[0098] The reconstruction and pre-processing of images can be carried out in the following manner.

[0099] From the sequence of N holographic images obtained, an average image is calculated. This image, called “background”, represents the entire background which does not correspond to moving objects.

[0100] A numerical algorithm, for example that known under the name “L1 Sobel” or that described in the aforementioned patent applications WO2017 / 162985 and WO2016 / 189257, digitally reconstructs the N holographic images as well as the image. More precisely, each of the N images is reconstructed at different steps, typically in steps of approximately 50 / 100 pm. Indeed, due to the thickness of the wall of the conduit 10, the particles are not all at the same distance Z. We therefore obtain for each image of the time series, Nz images reconstructed at different Z distances.

[0101] These Nz images are then combined to form a single one where each particle in the image is as sharp as possible. Thus, each pixel of the combined image will be obtained by taking the minimum value taken by this pixel on the Nz reconstructed images. Indeed, the sharper the particle, the lower the value of the pixels associated with the particle. This comes from the fact that light is absorbed more by each particle than by the air surrounding it. This step is carried out for each of the Nz images and the background image. At the end of this step, we then obtain N reconstructed images with the background image, on which the particles are sharp.

[0102] Unit 5 then performs a subtraction of the background image from each of the N images of the time series.

[0103] Unit 5 then selects, from among the N reconstructed images, the one on which the pollen particle is present using a detection algorithm.

[0104] From the images obtained, unit 5 can characterize the pollens by obtaining information on their size or morphology.

[0105] This identification can also be carried out by a deep learning device, preferably a convolutional neural network.

[0106] Thus, the identification can comprise the following steps:

[0107] Step el / : creation of a learning base comprising pollen images, or “known images”, each known image being that of a known pollen;

[0108] Step e2 / : training the deep learning device, using the learning base;

[0109] Step e3 / : submission of the selected image to said at least one deep learning device so that it determines at least one probability relating to the selected image;

[0110] Step e4 / : determination, as a function of said probability, of the belonging of the pollen of the selected image to the known pollen.

[0111] To improve the identification of pollens, step e / can also comprise the fusion of the data from the N holographic images with the diffusion intensity signals of each particle provided by the photodiode 31. For example, the diffusion signal that is detected can also be discriminating and therefore, after deep learning, a probability of belonging to a class can be provided. Thus, it is possible to combine the probabilities from the diffusion signal and the image from the image sensor to have a more precise identification of each pollen particle.

[0112] The invention is not limited to the examples which have just been described; it is possible in particular to combine characteristics of the examples illustrated within non-illustrated variants.

[0113] Other variants and improvements can be envisaged without departing from the scope of the invention.

[0114] In the example illustrated, the device and related operating method are concerned with pollens. It is entirely possible to implement such a device for the selection, detection, counting and identification of molds. In fact, molds, most often smaller than 10 μm, are generally smaller than pollens most often larger than 20 μm. Also, for molds, the lensless microscope 4 would require a higher spatial resolution.

Claims

1. Claims Apparatus (1) for selecting, detecting, counting and identifying pollens and / or molds initially suspended in atmospheric air comprising: - a conduit (10) for circulating a flow of atmospheric air likely to contain pollen and / or mold particles from an air inlet (11) to an air outlet (12); - a device (2) for concentrating pollen and / or mold particles, connected to the air inlet and adapted to concentrate particles present in the atmospheric air and allow pollen and / or mold particles in a selected size range to circulate downstream in the duct; - a device (3) for detecting pollen and / or mold particles, arranged around the conduit and downstream of the concentration device, comprising: a laser illumination source(s) (30), adapted to illuminate the air flow circulating in the duct, at least one photodetector (31) adapted to detect by diffusion or by fluorescence the passage of each pollen and / or mold particle in the air flow illuminated by the illumination source; - a lensless microscope (4), arranged around the conduit and downstream of the detection device, comprising: a diode illumination source (40), adapted to illuminate the air flow circulating in the duct, at least one CMOS type image sensor (41), arranged opposite the diode source(s) and connected to the photodetector so as to trigger the taking of a number N of holographic images produced by the light diffracted by each illuminated and detected pollen and / or mold particle, the image sensor being adapted to record the N holographic images; - an image processing unit (5), connected to the CMOS image sensor, adapted to respectively digitally reconstruct the N images, select from the N reconstructed images the one on which each particle is present pollen and / or mold detected and then identify, from the selected image, each particle of pollen and / or mold according to its size and / or morphology.

2. Apparatus according to claim 1, the concentrating device being a virtual impactor.

3. Apparatus according to claim 2, the virtual impactor being adapted to reduce the air flow in the duct between the air inlet and the air outlet by a factor of at least 5, preferably of the order of 10.

4. Apparatus according to claim 2 or 3, the virtual impactor being adapted to allow particles of a size greater than 5 pm to circulate downstream in the conduit.

5. Apparatus according to one of the preceding claims, the laser(s) of the illumination source of the detection device being emitted in the visible range, for example in the red range.

6. Apparatus according to one of the preceding claims, the photodetector(s) of the detection device being one or more photodiodes, in front of which (each of which) is arranged a parabolic mirror.

7. Apparatus according to one of the preceding claims, the diode(s) of the illumination source of the lensless microscope being a blue-emitting light-emitting diode(s).

8. Apparatus according to one of the preceding claims, the CMOS image sensor(s) of the lensless microscope being integrated into a high-speed camera or a camera adapted to carry out multi-acquisition of images.

9. Apparatus according to one of the preceding claims, the processing unit being further adapted to merge the data of the N holographic images with the scattering intensity signals of each particle provided by the photodetector(s).

10. Apparatus according to one of the preceding claims, comprising a portable housing (100) housing at least the conduit, the device for concentrating pollen and / or mold particles, the device for detecting pollen and / or mold particles, and the lensless microscope.

11. Method for selecting, detecting, counting and identifying pollens and / or molds initially suspended in atmospheric air, in particular implemented by the apparatus according to one of the claims preceding, comprising the following steps: a / suction in a duct, preferably arranged vertically, from an inlet to an outlet of a flow of atmospheric air likely to contain pollen and / or mold; b / selection of a given range of pollen and / or mold particles from among the particles sucked into the air flow; c / illumination of the air flow and detection by diffusion or by fluorescence of the passage of each selected pollen and / or mold particle; d / recording of a number N of holographic images produced by the light diffracted by each illuminated and detected pollen and / or mold particle;e / digital reconstruction of the N images, selection from the N reconstructed images of the one on which each detected pollen and / or mold particle is present, then identification, from the selected image, of each pollen and / or mold particle according to its size and / or morphology.;

12. Method according to claim 11, the detection according to step c / being carried out at 90° to the light source illuminating the air flow.

13. Method according to claim 11 or 12, the detection according to step c / making it possible to trigger the acquisition of the N holographic images.

14. Method according to one of claims 11 to 13, step e / comprising the fusion of the data of the N holographic images with the diffusion intensity signals of each particle provided by the photo-detector(s).

15. Method according to one of claims 11 to 14, the identification of the image selected according to step e / being carried out by a deep learning device, preferably a convolutional neural network.

16. Method according to claim 15, step e / comprising the following steps: el / creating a learning base comprising images of pollen and / or mold, or “known images”, each known image being that of a known pollen and / or mold; e2 / training the deep learning device, by means of the learning base; e3 / submitting the selected image to said at least one deep learning device so that it determines at least one pro- ability related to the selected image; e4 / determination, based on said probability, of the belonging of the pollen and / or mold of the selected image to the known pollen and / or mold.

Images