FR3091382A1 - HYPERSPECTRAL ACQUISITION DETECTION DEVICE - Google Patents

Abstract

The invention relates to a device for capturing (2) a hyperspectral scene and for direct detection (1), from the captured compressed image (11), of particularities in said scene comprising: - means acquisition (2) of a compressed image (11) of a hyperspectral focal plane; and - direct detection means (1) from the compressed image (11) comprising a neural network (12) structured to calculate an image (13) whose light intensity at the coordinates x and y represents the probability of presence of the peculiarity at the same x and y coordinate of the hyperspectral focal plane.

Description

Description

Title of the invention: Hyperspectral acquisition detection device

Technical area

The present invention relates to a device for detecting objects or features in the focal plane of a scene based on a measurement using a method of compressing the hyperspectral scene in three dimensions into a non-homogeneous image in two dimensions, and a processing of the image obtained making it possible to detect the particularities sought in the scene.

The invention finds a particularly advantageous application for on-board systems intended to detect objects or features in a scene from their shape, their texture and their light reflectance.

The invention can be applied to a large number of technical fields in which hyperspectral detection is sought. In a non-exhaustive manner, the invention can be used, for example, in the medical and dental field, to aid in diagnosis. In the plant and mycological field, the invention can also be used to carry out phenotyping, to detect symptoms of stress or disease or to differentiate between species. In the field of chemical analysis, the invention can also be used to measure concentrations. In the fight against counterfeiting, the invention can be used to discern counterfeiting.

Prior art

Within the meaning of the invention, a hyperspectral acquisition detection corresponds to the detection of features in the focal plane of a scene directly from an acquired two-dimensional image containing a representation of the spatial and spectral information of the focal plane of the scene.

Various methods of compressing the focal plane of a hyperspectral scene are described in the literature. The object of these methods is to acquire the focal plane of the hyperspectral scene in a single acquisition without the need to scan the focal plane of the scene in spatial or spectral dimensions.

For example, the Non-scanning imaging spectrometry thesis, Descour, Michael Robert, 1994, The university of Arizona, proposes a way to acquire a single two-dimensional image of the observed scene containing all the information on the influence of different wavelengths. This method, called CTIS (for Computed-Tomography Imaging Spectrometer), proposes to capture a diffracted image of the focal plane of the scene observed by means of a diffraction grating arranged upstream of a digital sensor. This diffracted image acquired by the digital sensor takes the form of multiple projections for which each projection makes it possible to represent the focal plane of the scene observed and contains all of the spectral information of the focal plane.

Another method, called CASSI (for Coded Aperture Snapshot Spectral Imaging), described in the thesis Compressive spectral imaging, D. Kittle, 2010, proposes a way to acquire a single two-dimensional encoded image containing all the spatial information. and spectral. This method proposes to capture a diffracted image, by means of a diffraction prism, and encoded, by means of an encoding mask, of the focal plane of the observed scene.

These methods, although satisfactory for solving the problem of instantaneous acquisition of the focal plane of the hyperspectral scene, require complex algorithms and expensive in computing resources in order to estimate the uncompressed hyperspectral scene. The review of snapshot spectral imaging technologies, Nathan Hagen, Michael W. Kudenov, Optical Engineering 52 (9), September 2013, presents a comparison of hyperspectral acquisition methods and the algorithmic complexities associated with each.

[0009] Indeed, the CTIS method requires an estimation process based on a two-dimensional matrix representing the transfer function of the diffraction optics. This matrix must be inverted to reconstruct the hyperspectral image. The matrix of the transfer function not being completely defined, iterative and costly matrix inversion methods of computing resources allow to approach the result step by step.

The CASSI method and its derivatives also require matrix calculations that are not completely defined, and use iterative calculation methods that are costly in computing resources in order to approach the result.

In addition, the three-dimensional hyperspectral image reconstructed by these calculation methods does not contain additional spatial or spectral information compared to the two-dimensional compressed image obtained by these acquisition methods. The estimation by the calculation of the hyperspectral image is therefore not necessary for a direct detection of the particularities sought in the focal plane of the scene, on the condition of using a method of direct detection of the particularities from the compressed image and not homogeneous in two dimensions.

Image processing methods with the aim of detecting particularities are widely described in the scientific literature. For example, a method based on neural networks is described in auto-association by multilayer perceptrons and singular value decomposition. Biological cybernetics, 59 (4): 291-294, 1988. ISSN 0340-1200, H. Bourlard and Y. Kamp. AT.

New methods based on deep and convolutional neural networks are also widely used with results showing very low rates of false detections. For example, such a method is described in Stacked Autoencoders Using Low-Power Accelerated Architectures for Object Récognition in Autonomous Systems, Neural Processing Letters, vol. 43, no. 2, pp. 445-458,2016, J. Maria, J. Amaro, G. Falcao, L. A. Alexandre.

These methods are particularly suitable for detecting elements in a color image (generally having 3 channels - Red, Green and Blue) of a scene by taking into account the characteristics of shapes, textures and colors of the particularity to detect. These methods consider the image homogeneous, and process the entire image by convolution by the same process.

The processing of the two-dimensional compressed images obtained by the CTIS and CASSI methods cannot therefore be operated by means of a standard deep and convolutional neural network. Indeed, the image obtained by these methods are not homogeneous, and contain non-linear features in the dimensions either spectral or spatial.

The technical problem of the invention consists in directly detecting the particularities or objects sought after the acquisition of a compressed, non-homogeneous, and non-linear representation in two dimensions containing all the spatial and spectral information of a scene hyperspectral in three dimensions by means of a convolutional neural network adapted to the particularities of the image obtained.

Statement of the invention

The present invention proposes to respond to this technical problem by directly detecting the specific features sought by means of a deep and convolutional formal neural network, the architecture of which is adapted to direct detection, applied to the compressed image. in two dimensions from the hyperspectral focal plane of the scene.

The three-dimensional hyperspectral image does not contain more spatial and spectral information than the compressed image obtained by the CTIS or CASSI acquisition methods since the hyperspectral image is reconstructed from the compressed image. Thus the invention proposes to detect directly in the compressed image the particularities sought in the focal plane of a scene.

To this end, the invention relates to a device for detecting features in the hyperspectral focal plane of a scene, said device comprising: means for acquiring a compressed and non-homogeneous image in two dimensions of a focal plane of a three-dimensional hyperspectral scene; and means for direct detection of features in said hyperspectral scene from the compressed and non-homogeneous image in two dimensions;

The invention is characterized in that the means for direct detection of features in the hyperspectral scene integrates a deep neural network and convolutional architecture to calculate a probability of presence of the feature (s) sought in said hyperspectral scene from the compressed image and generating an image for each particular feature sought, the value of each pixel with coordinates x and y corresponding to the probability of the presence of said feature at the same% and y coordinates of the hyperspectral focal plane of the scene.

The compressed image obtained by the optical system contains the focal plane of the non-diffracted scene in the center, as well as the diffracted projections along the axes of the different diffraction filters. Thus, the neural network uses the following information for the direct detection of the specific features sought: the light intensity in the central and non-diffracted part of the focal plane of the scene at the x and y coordinates; and light intensities in each of the diffractions of said compressed image whose coordinates x 'and y' are dependent on the coordinates% and y of the non-diffracted central part of the focal plane of the scene.

The invention thus makes it possible to correlate the information contained in the different diffractions of the diffracted image with information contained in the non-diffracted central part of the image obtained.

The different diffractions of the diffracted image containing important spectral information but each pixel of which contains a sum of the diffractions at different wavelengths, the invention can also use the central part of the image, not diffracted, and allows you to search the complete image for spatial (shape, texture, etc.) and spectral (reflectance) characteristics.

In practice, unlike the state of the conventional art of the CTIS method, the invention makes it possible to detect features in a hyperspectral scene in real time between two acquisitions of the hyperspectral focal plane of the scene observed. In doing so, it is no longer necessary to postpone the processing of the compressed images and it is no longer necessary to store these compressed images after detection. Also, it is no longer necessary to reconstruct the hyperspectral image before applying the detection method.

Alternatively, the compressed image obtained by the optical system contains the diffracted focal plane and encoded according to the coding scheme of a mask introduced into the optical path before diffraction of the scene. Thus, the neural network uses the following information for the direct detection of the specific features sought: the diagram of the encoding mask used to encode the diffractions of the focal plane of the scene; and light intensities in the compressed and diffracted image whose x 'and y' coordinates are dependent on the% and y coordinates of the focal plane of the observed scene.

In practice, unlike the state of the conventional art of the CASSI method, the invention makes it possible to detect features in a hyperspectral scene in real time between two acquisitions of the hyperspectral focal plane of the scene observed. In doing so, it is no longer necessary to postpone the processing of the compressed images and it is no longer necessary to store these compressed images after detection. Also, it is no longer necessary to reconstruct the hyperspectral image before applying the detection method.

According to one embodiment, the capture method is a compact mechanical embodiment and can be integrated into a portable and autonomous device and in which the detection method is executed by said portable and autonomous device.

According to one embodiment, the capture method is a compact mechanical embodiment which can be integrated in front of the lens of the camera of a smartphone and in which the detection method is executed by the smartphone.

According to one embodiment, said compressed image is obtained by an infrared sensor. This embodiment provides information invisible to the human eye.

According to one embodiment, said compressed image is obtained by a sensor whose wavelength is between 0.001 nanometer and 10 nanometers. This embodiment makes it possible to obtain information on the X-rays present on the observed scene.

According to one embodiment, said compressed image is obtained by a sensor whose wavelength is between 10,000 nanometers and 20,000 nanometers. This embodiment makes it possible to obtain information on the temperature of the scene observed.

According to one embodiment, said compressed image is obtained by a sensor comprising: a first converging lens configured to focus the information of a scene on an opening; a collimator configured to capture the rays passing through said opening and to transmit these rays over a diffraction grating; and a second converging lens configured to focus the rays from the diffraction grating on a collection surface.

This embodiment is particularly simple to perform and can be adapted on an existing sensor.

According to one embodiment, said compressed image is obtained by a sensor comprising: a first converging lens configured to focus the information of a scene on a mask; a collimator configured to capture the rays passing through said mask and to transmit these rays over a prism; and a second converging lens configured to focus the rays from the prism on a sensing surface.

This embodiment is particularly simple to perform and can be adapted on an existing sensor.

With regard to the detection of features from said compressed image, the invention uses a deep and convolutional neural network for determining the probabilities of the presence of these features. Learning of said deep and convolutional neural network makes it possible to indicate the probability of the presence of the particularities sought for each% and y coordinates of said hyperspectral scene. For example, learning by back-propagation of the gradient or its derivatives from training data can be used.

The deep and convolutional neural network used for direct detection from the compressed image has an input layer structure suitable for direct detection. The invention has several architectures of the deep layers of said neural network. Among these, an auto-encoder architecture as described in the SegNet document: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation, Vijay Badrinarayanan, Alex Kendall, Roberto Cipolla allows to indicate the probability of the presence of the particularities sought for each % and y coordinates of the hyperspectral scene.

Said input layer of the neural network is adapted to the structure of the compressed image obtained by the acquisition means. Thus, the input layer is a tensor of order three and has two spatial dimensions of size X _MAX and Y _MAX , and a depth dimension of size D _MAX .

The invention uses the non-linear relation f (x ,, y _t , d _t ) -> (x _img , y _img ) defined for x, [0..Χ _ΜΑ χ [, y _t [0 .. Υ _ΜΑ χ [and d _t [0..D _MAX [allowing to calculate the coordinates x _img and y _img of the pixel of said compressed image whose intensity is copied in the tensor of order three of said input layer of neural network at coordinates (x _t , y _t , d _t ) ·

According to one embodiment, the capture method comprises a means of acquiring a compressed image containing the diffractions of said hyperspectral focal plane obtained with diffraction filters. The compressed image obtained contains the focal plane of the non-diffracted scene in the center, as well as the diffracted projections along the axes of the different diffraction filters. The input layer of the network of d _t ] = neuron is a copy of the eight chromatic representations of the hyperspectral scene of the compressed image according to the following non-linear relationship:

^X img ~ ^x + ^X _o ffsetx ( ⁿ sliceX j ^with:

y. = y + y i mg

Π = floor (d _t / d _MAX ); A = d _t mo dd _MAX ; ⁿ between 0 and 7, the number of diffraction of the compressed image; df between 0 and D max ', ^xt between 0 and X max / between 0 and Y max; Λ s / j _c ? γ, the depth constant of the third order tensor of said input layer; λ _s n _ce x ^ l ^a spectral step constant of the pixel in X of said compressed image; λ _s n _Cë y, the spectral step constant of the pixel in Y of said compressed image; X _o ff _s ) corresponding to the offset along the X axis of the diffraction ⁿ ; y (/ 1) coroffsetY '' corresponding to the offset along the Y axis of the diffraction ⁿ

According to one embodiment, the capture method comprises a means of acquiring a compressed image containing a two-dimensional encoded representation of said hyperspectral focal plane obtained with a mask and a prism. The compressed image obtained contains the focal plane of the diffracted scene and encoded in the center. The input layer of the neuron network is a copy of the image compressed according to the following non-linear relationship:

i ^{X + X} offset ' ^SiX ~ ¹ ' Havecjf

I mask (x _t , y _t ), if η ο nj (y + y, siy> 1 / r ⁷ offset ' ⁷ I] mask (x _f , y _t ), otherwise I between 0 and D _Μ χ _x ; ^x t between 0 and Χ ^ χχ; Λ _s n _ce including between 0 and Y x / χ x ', ^ - _o ff _se ty corresponding to the shift along the X axis of the diffraction ⁿ Y _o ff _se ^ - corresponding to the offset along the Y axis of the diffraction ⁿ

These non-linear relationships make it possible to quickly find the intensity of the pixels of interest in each diffraction. Indeed, some pixels can be neglected if the wavelength of the diffracted image is not significant.

The architecture of the deep and convolutional neural network is composed of an encoder making it possible to search for the elementary characteristics specific to the desired detection, followed by a decoder making it possible to generate an image of probabilities of presence of the characteristics to be detected in said compressed image of the hyperspectral focal plane. The encoder / decoder structure makes it possible to search for the elementary characteristics specific to the main characteristic sought in said compressed image.

According to one embodiment, the encoder is composed of a succession of convolutional neural layers alternating with pooling layers (operator of decimation of the previous layer) making it possible to reduce the spatial dimension.

According to one embodiment, the decoder is composed of a succession of layers of deconvolution neurons alternating with layers of unpooling (operation of interpolation of the previous layer) allowing an increase in the spatial dimension.

For example, such an encoder / decoder structure is described in the SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation, Vijay Badrinarayanan, Alex Kendall, Roberto Cipolla.

According to one embodiment, a set of fully connected neural layers can be positioned between the encoder and the decoder.

SUMMARY DESCRIPTION OF FIGURES

The manner of carrying out the invention as well as the advantages which ensue therefrom, will emerge clearly from the embodiment which follows, given for information but not limitation, in support of the appended figures in which Figures 1 to 5 represent :

[fig.l]: a schematic front view of the elements of a capture and detection device in a hyperspectral scene according to an embodiment of the invention;

[fig.2]: a schematic structural representation of the elements of the device of FIG. 1;

[fig.3]: an alternative structural schematic representation of the elements of the device of FIG. 1;

[fig.4]: a schematic representation of the diffractions obtained by the acquisition device of FIG. 2;

[fig.5]: a schematic representation of the architecture of the neuron network of FIG. 2.

WAY TO DESCRIBE THE INVENTION

[0049] FIG. 1 illustrates a device 2 for capturing a hyperspectral scene 3 comprising a sensor making it possible to obtain a two-dimensional compressed image 11 of a focal plane 3 of an observed scene.

As illustrated in FIG. 2, the capture device 1 comprises a first converging lens 21 which focuses the focal plane 3 on an opening 22. A collimator 23 captures the rays passing through the opening 22 and transmits these rays to a diffraction grating 24. A second converging lens 25 focuses these rays from the diffraction grating 24 on a collection surface 26.

The structure of this optical assembly is relatively similar to that described in the scientific publication Computed-tomography imaging spectrometer: experimental calibration and reconstruction results, published in APPLIED OPTICS, volume 34 (1995) number 22.

This optical structure makes it possible to obtain a diffracted image 11, illustrated in FIG. 3, having several diffractions R0-R7 of the focal plane 3 arranged around a non-diffracted image of small size C. In the example of FIG. 4, the diffracted image presents eight distinct R0-R7 diffractions obtained with two diffraction axes of the diffraction grating 24.

Alternatively, three diffraction axes can be used on the diffraction grating 24 so as to obtain a diffracted image 11 with sixteen diffractions.

The capture surfaces 26 or 46 may correspond to a CCD sensor (for charge-coupled device in the English literature, that is to say a charge transfer device), to a CMOS sensor (for complementary metaloxide-semiconductor in Anglo-Saxon literature, a technology for manufacturing electronic components), or any other known sensor. For example, the scientific publication Practical Spectral Photography, published in Eurographics, volume 31 (2012) number 2, proposes to associate this optical structure with a standard digital camera for capturing the diffracted image.

Alternatively, as illustrated in FIG. 3, the capture device 1 may include a first converging lens 41 which focuses the focal plane 3 on a mask 42. A collimator 43 captures the rays passing through the mask 42 and transmits these rays to a prism 44. A second converging lens 45 focuses these rays coming from the prism 44 on a collection surface 46.

The structure of this optical assembly is relatively similar to that described in the scientific publication Compressive Coded Aperture Spectral hnaging, Gonzalo R. Arce, David J. Brady, Lawrence Carin, Henry Arguello, and David S. Kittle.

Alternatively, the capture surfaces 26 or 46 may correspond to the photographic acquisition device of a smartphone or any other portable device including a photographic acquisition arrangement, by adding the device for capturing the hyperspectral scene 3 in front the photographic acquisition device.

Preferably, each pixel of the diffracted image 11 is coded in three colors red, green and blue and on 8 bits thus making it possible to represent 256 levels on each color.

Alternatively, the capture surfaces 26 or 46 can be a device whose sensed wavelengths are not in the visible part. For example, the device 2 can integrate a sensor whose wavelength is between 0.001 nanometer and 10 nanometers or a sensor whose wavelength is between 10,000 nanometer and 20,000 nanometers.

When the image 11 of the hyperspectral focal plane observed is obtained, the detection means implements a neuron network 12 to detect a particular feature in the scene observed from the information of the compressed image 11.

This neural network 12 aims to determine the probability of the presence of the specific feature sought for each pixel located at the% and y coordinates of the hyperspectral scene observed.

To do this, as illustrated in FIG. 3, the neural network 12 comprises an input layer 30, capable of extracting the information from the image 11, and an output layer 31, capable of processing this information so as to generate an image whose intensity of each pixel at the x and y coordinate, corresponds to the probability of the presence of the particularity at the% and y coordinates of the hyperspectral scene 3.

The architecture of said neural network 12 is composed of a set of convolutional layers such as layer 31 assembled linearly and alternately with decimation (pooling) or interpolation (unpooling) layers.

A convolutional layer of depth d, denoted CONV (d), is defined by d convolution nuclei, each of these convolution nuclei being applied to the volume of the input tensor of order three and of size X j _{n ni} , _r , Y. d,. The convolutive input 'mput layer thus generates an output volume, tensor of order three, having a depth d. An ACT activation function is applied to the calculated values of the output volume. The parameters of each convolution kernel of a convolutional layer are specified by the learning procedure of the neuron network.

Different ACT activation function can be used. For example, this function can be a ReLu function, defined by the following equation:

R C Lu (X) = FM <2 X (0, X) Alternating with the convolutional layers, decimation layers (pooling) or interpolation layer (unpooling) are inserted.

A decimation layer makes it possible to reduce the width and the height of the tensor of order three at the input for each depth of said tensor of order three. For example, a MaxPool decimation layer (2,2) selects the maximum value of a sliding tile on the surface of 2x2 values. This operation is applied to all the depths of the input tensor and generates an output tensor having the same depth and a width divided by two, as well as a height divided by two.

A neuron network architecture allowing the direct detection of features in the hyperspectral scene can be as follows:

Input

CONV (64) MaxPool (2,2)

CONV (64) MaxPool (2,2)

CONV (64) MaxPool (2,2)

CONV (64) CONV (64)

MaxUnpool (2,2) CONV (64)

MaxUnpool (2,2) CONV (64)

MaxUnpool (2,2) CONV (l)

Output

The weights of said neural network are calculated by means of learning. For example, backward propagation of the gradient or its derivatives from training data can be used to calculate these weights.

Alternatively, the neuron network 12 can determine the probability of the presence of several distinct features within the same scene observed. In this case, the last convolutional layer will have a depth corresponding to the number of distinct features to be detected. Thus the convolutional layer CONV (l) is replaced by a convolutional layer CONV (x), where x corresponds to the number of distinct features to be detected.

Claims (1)

Claims [Claim 1] Device for capturing (2) an image (11) of the focal plane of a hyperspectral scene and for direct detection (1) of features in this scene (3), said device comprising: a means of acquisition (2) of a compressed and non-homogeneous two-dimensional image (11) of a focal plane of a three-dimensional hyperspectral scene; and means for direct detection (1) of features or object in said hyperspectral scene (3) from the compressed and non-homogeneous image in two dimensions; characterized in that it comprises a direct detection means (1) integrating a deep and convolutional neural network (12) architecture for calculating a probability of presence of the characteristic (s) sought in the hyperspectral scene (3) from the image compressed (11) and generate an image for each particular feature sought whose value of each pixel with coordinates% and y corresponds to the probability of presence of said feature at the same coordinates% and y of the hyperspectral focal plane of the scene (3). [Claim 2] Device according to claim 1, in which the capture method comprises means for acquiring a compressed image (11) containing the diffractions of said hyperspectral focal plane (3) obtained with diffraction filters (24). The compressed image (11) obtained contains the focal plane of the non-diffracted scene (C) in the center, as well as the diffracted projections (R0-R7) along the axes of the different diffraction filters (24). The input layer of the neural network (30) is a copy of the eight chromatic representations (R0-R7) of the hyperspectral scene (3) of the compressed image (11) according to the following non-linear relationship: fiv \ - ~ ^{X + X} offse.tx ( ⁿ ) + ^ · λ sîicex 1 H ^X t> y a _t ) - 'y = y + y (n) + λ .λ γ img ^{z 2} offsetY''s / reeij with: - n = floor (d _t / d _MAX ); Λ = d τη o dd μ. γ γ; - ⁿ between 0 and 7, the number of diffraction of the compressed image; - d _t between 0 and D _{MA x} ; - ^x t between 0 and X ma x; - Λ _s ii _ce between 0 and Y max - λ _s n _ce χ, the depth constant of the third order tensor of said input layer; - λ _s ] χ, the spectral step constant of the pixel in X of said compressed image; - Λ _v γ, the spectral step constant of the pixel in Y of said compressed image; - X _o ff _se t γ ( ^w ) corresponding to the offset along the X axis of the diffraction ⁿ ; - y (n) corresponding to the offset along the Y axis of the fraction ⁿ [Claim 3] Device according to claim 1, in which the capture method comprises a means of acquiring a compressed image containing a representation in two encoded dimensions of said hyperspectral focal plane obtained with a mask and a prism. The compressed image obtained contains the focal plane of the diffracted scene and encoded in the center. The input layer of the neuron network is a copy of the image compressed according to the following non-linear relationship: [ ^{x + x} _O ffse _t ' ^{six 3: 1} | mask (x _t , y _t ), if η ο n (y + y, siy> 1) l ^{z 7} offset ' ⁷ I] mask (x _t , ysinon \ with: - d _t between 0 and D _{MA x} ; - ^x t between 0 and Xmax - λ _s ] i _ce ^ between 0 and Ymax; X offse ^ corresponding to the offset along the X axis of the diffraction n. -y offse Ω, corresponding to the offset along the Y axis of the diffraction [Claim 4] Device according to claims 1 to 3 in which the capture method is a compact mechanical embodiment and can be integrated into a portable and autonomous device and in which the detection method is executed by said portable and autonomous device. [Claim 5] Device according to Claims 1 to 3, in which the capture method is a compact mechanical embodiment which can be integrated in front of the lens of the camera of a smartphone and in which the detection method is executed by the smartphone. [Claim 6] Device according to one of claims 1 to 4, in which the compressed image (11) is obtained by an infrared sensor. [Claim 7] Device according to one of claims 1 to 4, in which the compressed image (11) is obtained by a sensor whose wavelength is between 0.001 nanometer and 10 nanometers. [Claim 8] Device according to one of claims 1 to 4, in which the compressed image (11) is obtained by a sensor whose wavelength is between 10,000 nanometers and 20,000 nanometers. [Claim 9] Device according to one of claims 1 to 8, in which said compressed image (11) is obtained by a sensor (26) comprising: a first converging lens (21) configured to focus the information of a scene on an aperture (22 ); a collimator (23) configured to capture the rays passing through said opening (22) and to transmit these rays over a diffraction grating (24); and a second converging lens (25) configured to focus the rays from the diffraction grating (24) on a collection surface (26). [Claim 10] Device according to one of claims 1 to 8, in which said compressed image (11) is obtained by a sensor (46) comprising: a first converging lens (41) configured to focus the information of a scene on a mask (42 ); a collimator (43) configured to capture the rays passing through said mask (42) and to transmit these rays over a prism (44); and a second converging lens (45) configured to focus the rays from the prism (44) on a sensing surface (46). 1/3