CN110487211B - Aspheric element surface shape detection method, device and equipment and readable storage medium - Google Patents
Aspheric element surface shape detection method, device and equipment and readable storage medium Download PDFInfo
- Publication number
- CN110487211B CN110487211B CN201910931375.2A CN201910931375A CN110487211B CN 110487211 B CN110487211 B CN 110487211B CN 201910931375 A CN201910931375 A CN 201910931375A CN 110487211 B CN110487211 B CN 110487211B
- Authority
- CN
- China
- Prior art keywords
- surface shape
- layer
- aperture
- sub
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
- G01B11/2441—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures using interferometry
Abstract
The embodiment of the invention discloses a method, a device and equipment for detecting the surface shape of an aspheric element and a computer readable storage medium. Inputting sub-aperture detection data of the aspheric element to be detected into a surface shape distribution calculation model obtained by training a deep learning network model by using a sample data set in advance to obtain the relative position and the overlapping area of each sub-aperture, and then performing sub-aperture splicing by using a sub-aperture splicing algorithm to realize surface shape detection of the aspheric element; the deep learning network model comprises a cascade deep network framework and a position calculation layer; and the cascade depth network framework estimates the initial position and the position deviation value of the annular sub-aperture layer by layer, and the position calculation layer finely adjusts the ring belt position of each annular sub-aperture by utilizing a multi-level regression algorithm based on the initial position and the position deviation value of the annular sub-aperture. The method and the device realize automatic, rapid and precise determination of the surface shape distribution information of each sub-aperture, and effectively reduce the surface shape detection cost of the aspheric element.
Description
Technical Field
The embodiment of the invention relates to the technical field of optical detection, in particular to a method, a device and equipment for detecting the surface shape of an aspheric element and a computer-readable storage medium.
Background
The aspheric optical element has incomparable good imaging quality of a spherical optical device, can be used for correcting various aberrations well in an optical system, improving the imaging quality and increasing the working distance, can replace more spherical parts with one or a small number of aspheric parts, and is favorable for the development of optical instruments towards the trend of miniaturization and lightness. With the rapid development of optical ultra-precision processing technology, aspheric optical elements are widely applied in the fields of aviation, aerospace, national defense and high-tech civil use.
It can be understood that the aspheric surface detection technology is one of the key factors restricting the application of aspheric surface elements, and the traditional aspheric surface detection method is a zero compensation detection method. In null compensation detection, auxiliary optical components, such as compensators, Computer Generated Holograms (CGH), etc., are typically required to convert the incident planar/spherical wavefront into an aspheric wavefront that matches the measured element. Although the detection method is high in precision, different aspheric surfaces need matched auxiliary optical components for detection, the auxiliary optical components are uniquely corresponding to the aspheric surface structure, and the auxiliary optical components are not universal, so that the method is long in design and processing period, high in production cost and not suitable for being used in mass production.
In order to solve the above problems, the related art uses a sub-aperture stitching detection method to realize surface shape detection of an aspheric element. The method divides the aspheric surface to be detected into a plurality of small detection areas, each small detection area is a sub-aperture area, the aspheric surface degree is greatly reduced because a single detection object is a certain sub-aperture area of the aspheric surface, so that the detection can be directly carried out by using a standard interferometer without additional auxiliary compensation optical elements, the detection of all sub-apertures can be completed by changing the relative position of the standard interferometer and the aspheric surface to be detected, and finally, the detection results of all sub-apertures are spliced by adopting a proper algorithm to obtain the complete detection result of the optical element to be detected.
Sub-aperture stitching detection can be divided into circular sub-aperture stitching and circular sub-aperture stitching. The us QED corporation developed a commercialized sub-aperture Stitching Interferometer workstation (SSI) in 2003, which is suitable for planar, spherical and moderately aspheric surfaces with an aperture of 200mm or less. ygo developed an aspherical surface shape detection interferometer-VeriFire Asphere based on circular subaperture stitching. The sub-aperture splicing method can improve the transverse detection resolution and the vertical detection range of the interferometer, can effectively solve the contradiction between a large view field and high resolution, but requires the realization of six-dimensional precise adjustment between a detected element and the interferometer, so the cost is very high.
In the sub-aperture splicing method, the main factors influencing the detection precision comprise four aspects: 1) the mechanical structure positioning error causes the deviation of the relative position between each sub-aperture and the ideal position; 2) the detection precision of the curvature radius of the best fitting sphere of each sub-aperture is obtained; 3) adjusting errors; 4) systematic error. Wherein, the adjustment error can be calculated and processed by a splicing algorithm; systematic errors can be calibrated through the systematic errors so as to be removed from each sub-aperture; other two factors are mainly realized by an ultra-precise positioning mechanism and an ultra-precise position measuring device at present, so that the aspheric surface shape detection cost is high.
Disclosure of Invention
The embodiment of the disclosure provides a method, a device and equipment for detecting the surface shape of an aspheric element, and a computer-readable storage medium, which realize automatic, rapid and precise determination of surface shape distribution information of each sub-aperture, effectively reduce the surface shape detection cost of the aspheric element on the basis of ensuring the aspheric surface shape detection accuracy, and are very suitable for batch production detection.
In order to solve the above technical problems, embodiments of the present invention provide the following technical solutions:
the embodiment of the invention provides a method for detecting the surface shape of an aspheric element on one hand, which comprises the following steps:
training a deep learning network model by utilizing a sample data set in advance to obtain a surface shape distribution calculation model;
inputting sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model to obtain the relative position and the overlapping area of each sub-aperture;
based on the relative position and the overlapping area of each sub-aperture, sub-aperture splicing is carried out by utilizing a sub-aperture splicing algorithm so as to realize surface shape detection of the aspheric element to be detected;
the deep learning network model comprises a cascade depth network framework and a position calculation layer, wherein the sample data set comprises a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result; the cascade depth network framework is used for estimating the initial position and the position deviation value of the annular sub-aperture layer by layer, the position calculation layer is used for fine-tuning the ring belt position of each annular sub-aperture based on the initial position and the position deviation value of the annular sub-aperture by utilizing a multi-level regression algorithm.
Optionally, the cascaded deep network framework includes a global network layer and a plurality of local network layers;
the global network layer is used for directly mapping the surface shape error value to the annulus position parameter by learning a nonlinear mapping function, and calculating to obtain an initial annulus position estimation value and a position deviation value of each annular sub-aperture;
each local network layer is used for updating the zone position value of each annular sub-aperture by gradually iterating the deviation value between the zone position of the current annular sub-aperture and the real zone position.
Optionally, the global network layer includes a first convolutional neural network, a second convolutional neural network, and a third convolutional neural network;
the first convolutional neural network, the second convolutional neural network and the third convolutional neural network sequentially comprise an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and an output layer; the input data of the input layer of the first convolutional neural network is a whole surface-shaped error image, the input data of the input layer of the second convolutional neural network is a first preset area and a second preset area of the surface-shaped error image, and the input data of the input layer of the third convolutional neural network is a third preset area and a fourth preset area of the surface-shaped error image.
Optionally, the structures of the local network layers of the cascaded deep network frame are the same, and the convolutional neural network in the first network of each local network layer is used for learning a mapping function from a surface-shaped error image space to an annular position deviation parameter space of the annular sub-aperture.
Optionally, each convolutional neural network of each local network layer of the cascaded deep network framework is optimized according to an optimization relation:
Δpk＝Δpk(x)＝P(x)-Pk-1(x)；
in the formula (I), the compound is shown in the specification,
is a non-linear function mean square error minimization function, Δ pkIs the average ring zone position deviation of the k local network layer, P (x) is the true position value corresponding to the ring zone image, Pk-1(x) For the position estimation value corresponding to the k-1 th local network layer ring zone image, lkAs a mapping function for the kth convolutional neural network, LkA mapping function for the first network of the kth local network layer.
Optionally, the position calculation layer calculates the annulus position estimation value of the current annular sub-aperture at each layer by using an annulus position estimation relational expression; the annulus position estimation relation is:
in the formula, pnIs an estimate of the location of the annulus,
the position of the ring band of the ith annular sub-aperture of the ith network layer, n is the total number of layers of the convolutional neural network, i is the serial number of the current network layer, and n is the serial number of the current network layeriIn order to be the neural network of the i-th layer,
is the deviation value between the zone position of the ith annular sub-aperture and the real zone position.
Optionally, the training process of the surface shape distribution calculation model includes:
expressing the aspheric surface shape errors corresponding to the surface shape error images in the sample data set by using a predefined relational expression, wherein the predefined relational expression is as follows:
carrying out rotation average processing on the aspheric surface shape error of each surface shape error image so as to train the deep learning network model by using the circular subaperture after rotation average to obtain the surface shape distribution calculation model;
wherein V (x, y) is a surface shape error, anmIs a coefficient, XnmTo describe the orthogonal polynomials of the surface shape, m and n are positive integers.
Another aspect of the embodiments of the present invention provides an aspheric surface element surface shape detection apparatus, including:
the model training module is used for training the deep learning network model by utilizing the sample data set to obtain a surface shape distribution calculation model; the sample data set is a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result, and the deep learning network model comprises a cascade deep network frame and a position calculation layer; the cascade depth network framework is used for estimating initial positions and position deviation values of the annular sub-apertures layer by layer, the position calculation layer is used for fine-tuning the ring belt positions of the annular sub-apertures based on the initial positions and the position deviation values of the annular sub-apertures by utilizing a multi-level regression algorithm;
the surface shape distribution information calculation module is used for inputting the sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model to obtain the relative position and the overlapping area of each sub-aperture;
and the surface shape detection module is used for splicing the sub-apertures by using a sub-aperture splicing algorithm based on the relative position and the overlapping area of each sub-aperture so as to realize the surface shape detection of the aspheric element to be detected.
An embodiment of the present invention further provides an aspheric element surface shape detecting device, which includes a processor, and the processor is configured to implement the steps of the aspheric element surface shape detecting method according to any one of the foregoing embodiments when executing the computer program stored in the memory.
Finally, an embodiment of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a surface shape detection program of an aspheric element, and the surface shape detection program of the aspheric element, when executed by a processor, implements the steps of the aspheric element surface shape detection method according to any one of the previous items.
The technical scheme provided by the application has the advantages that the deep neural network model is trained by using sample data, the mapping relation between the subaperture surface shape distribution and the subaperture position is established, the subaperture surface shape distribution information can be automatically output after the current subaperture data is input into the trained model, the subaperture surface shape distribution information is automatically, quickly and precisely positioned, finally, the surface shape detection is realized by using a subaperture splicing algorithm according to the relative position and the overlapped area of the subaperture, the precision requirement on the mechanical structure for realizing the subaperture splicing is reduced in the whole surface shape detection process, the aspheric surface shape detection cost can be effectively reduced on the basis of ensuring the aspheric surface shape detection accuracy, and the method is very suitable for batch production detection.
In addition, the embodiment of the invention also provides a corresponding implementation device, equipment and a computer readable storage medium for the aspheric element surface shape detection method, so that the method has higher practicability, and the device, the equipment and the computer readable storage medium have corresponding advantages.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the related art, the drawings required to be used in the description of the embodiments or the related art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a schematic flow chart of a method for detecting a surface shape of an aspheric element according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of measurable sub-aperture areas corresponding to different element locations provided by an embodiment of the present invention;
fig. 3 is a schematic structural diagram of each convolutional neural network in the global network layer according to an embodiment of the present invention;
fig. 4 is a structural diagram of an embodiment of an aspheric surface element surface shape detection apparatus according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the disclosure, the invention will be described in further detail with reference to the accompanying drawings and specific embodiments. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The terms "first," "second," "third," "fourth," and the like in the description and claims of this application and in the above-described drawings are used for distinguishing between different objects and not for describing a particular order. Furthermore, the terms "comprising" and "having," as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, article, or apparatus that comprises a list of steps or elements is not limited to only those steps or elements but may include other steps or elements not expressly listed.
Having described the technical solutions of the embodiments of the present invention, various non-limiting embodiments of the present application are described in detail below.
Referring to fig. 1, fig. 1 is a schematic flow chart of a method for detecting a surface shape of an aspheric element according to an embodiment of the present invention, where the embodiment of the present invention includes the following:
s101: and training the deep learning network model by using the sample data set in advance to obtain a surface shape distribution calculation model.
It can be understood that the sub-aperture position and the surface shape detection result of the aspheric element have a complex mapping relationship, as shown in fig. 2, when the aspheric element to be measured is at different positions relative to the interferometer, the interference fringes in different areas of the aspheric element to be measured are sparse, and the interferometric measurement can be realized. Mathematical description of this complex mapping relationship of subaperture locations and aspheric surface profile detection results can be achieved using any regression modeling method. Conventional regression models can fit relatively simple nonlinear functions, and present significant challenges when used to build complex mapping relationships. Deep neural networks, a recently emerging technology, have the ability to automatically learn complex mappings from large amounts of data. In consideration of the fact that the sub-aperture has the eccentric error within a certain space range in the actual detection process, the method can adopt the deep neural network to establish the regression model aiming at the uncertainty of the annular sub-aperture position so as to accurately depict the complex mapping relation between the annular sub-aperture position and the surface shape error, and the influence of the annular sub-aperture eccentric error on the final detection precision is effectively reduced.
The sample data set used for training the deep learning network model can be a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result. The method comprises the steps of establishing an aspheric interference detection model in optical software such as CodeV, randomly changing the position of an aspheric element, obtaining the corresponding sub-aperture position and sub-aperture profile distribution through model simulation, namely profile detection result data, and repeating for multiple times until the data sampling requirement of a deep learning network model is met. And selecting a series of annular sub-aperture positions and corresponding ideal aspheric surface detection results, inputting the data into a deep learning model, and establishing a deep learning network model between the annular sub-aperture positions and the ideal aspheric surface shape measurement results. The deep learning network model can comprise a cascade deep network framework and a position calculation layer; the cascade depth network framework is used for estimating the initial position and the position deviation value of the annular sub-aperture layer by layer, the position calculation layer is used for fine-tuning the ring zone position of each annular sub-aperture from rough to fine based on the initial position and the position deviation value of the annular sub-aperture by utilizing a multi-layer regression algorithm. The method comprises the steps of inputting a face shape error image in the process of training a face shape distribution calculation model, outputting a corresponding ring belt sub-aperture position, and gradually realizing accurate estimation of the ring belt sub-aperture position from rough to fine by adopting a cascade frame and a multi-level regression algorithm based on a deep neural network.
S102: and inputting the sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model to obtain the relative position and the overlapping area of each sub-aperture.
The sub-aperture detection data of the aspheric element to be detected can be acquired by any interference detection device, and the realization of the application is not affected.
S103: and based on the relative position and the overlapping area of each sub-aperture, sub-aperture splicing is carried out by using a sub-aperture splicing algorithm so as to realize the surface shape detection of the aspheric element to be detected.
The sub-aperture splicing can be realized by adopting any sub-aperture splicing algorithm in the related technology based on the relative position and the overlapping area of each sub-aperture, and after the splicing result is obtained, the surface shape detection of the aspheric element to be detected can be realized according to the splicing result.
In the technical scheme provided by the embodiment of the invention, the deep neural network model is trained by using sample data, the mapping relation between the subaperture surface shape distribution and the subaperture position is established, the subaperture surface shape distribution information can be automatically output after the current subaperture data is input into the trained model, the automatic, quick and precise positioning of the subaperture is realized, finally, the surface shape detection is realized by using a subaperture splicing algorithm according to the relative position and the overlapping region of the subaperture, the precision requirement on a mechanical structure for realizing subaperture splicing is reduced in the whole surface shape detection process, the aspheric surface shape detection cost can be effectively reduced on the basis of ensuring the aspheric surface shape detection accuracy, and the method is very suitable for batch production detection.
As a preferred embodiment, the cascaded deep network framework of the surface shape distribution calculation model of the present application may include a global network layer and a plurality of local network layers. The global network layer is used for directly mapping the surface shape error value to the annular zone position parameter by learning a nonlinear mapping function, and calculating to obtain an initial annular zone position estimation value and a position deviation value of each annular sub-aperture. The global network layer may include a first convolutional neural network, a second convolutional neural network, and a third convolutional neural network. The structure of each convolutional neural network is the same, and the convolutional neural networks can comprise an input layer, two convolutional layers, two pooling layers and an output layer, the only difference is that the input data types of the input layers are different, as shown in fig. 3, the first convolutional neural network, the second convolutional neural network and the third convolutional neural network sequentially comprise the input layer, the first convolutional layer, the first pooling layer, the second convolutional layer, the second pooling layer and the output layer; the input data of the input layer of the first convolutional neural network is a whole surface-shaped error image, the input data of the input layer of the second convolutional neural network is a first preset area and a second preset area of the surface-shaped error image, and the input data of the input layer of the third convolutional neural network is a third preset area and a fourth preset area of the surface-shaped error image. For example, different area type error image areas may be input to the input layers of the respective convolutional neural networks, the input area type area of the input layer of the first convolutional layer may be the entire area type error image, the input area type area of the input layer of the second convolutional layer may be the middle area and the upper area of the area type error image, and the input area type area of the input layer of the third convolutional layer may be the middle area and the lower area of the area type error image. These differentiated input images can cover more possibilities, further improving the accuracy of the annulus subaperture position estimation.
If x ∈ RdFor a profile error image of d pixels, P (x) ∈ R3And representing the corresponding position parameters of the surface shape error image. From the above, the problem of the annular subaperture annular position estimation is to learn the nonlinear mapping relationship F from the surface shape error image space to the subaperture annular position space, i.e. F: x → P (x). To better model this complex nonlinear function F, the circular subaperture annulus location estimation problem can be transformed into a minimization of the mean square error function as follows:
qi(ai-1)＝σ(Wiai-1+bi)＝ai，i＝1，2，…，l–1；
ql(al-1)＝Wlal-1+bl；
wherein F ═ Fk((fk-1(…f1)))，fkA mapping function representing a kth convolutional neural network. For a convolutional neural network of l layers, fi＝(ql-1(…q1(x)))。aiRepresents an activation value, (W)iai-1+bi) To output value, σ is the activation function. The use of an activation function, σ, which may be a hyperbolic tangent function having a range of [ -1, 1 ] in convolutional neural network convolutional layers]. In the global network layer, the goal of each convolutional neural network is to predict the value f by minimizing the zone subaperture positioni(x) And the mean square error function between the real value P (x) of the annular sub-aperture position realizes the estimation of the annular sub-aperture position. Wherein f isi(x) The mapping function representing the global network, i.e. the goal of the global network layer, is to estimate the annulus subaperture position parameters.
Based on the above embodiment, the initial zone sub-aperture position estimation value P is obtained through the global network layer0Then, the deviation between the position value of the sub-aperture of the ring zone and the real value can be obtained at the same time. Due to the initial value P0The structure of each local network layer of the cascaded depth network framework may be the same, and the convolutional neural network in the first network of each local network layer is used to learn the mapping function L between the annular position deviation parameter space Δ P (x) from the surface shape error image space to the annular sub-aperture1Can be represented by the following formulaShown in the figure:
in the formula,. DELTA.P1＝ΔP1(x)＝P(x)-P0(x)，L1A mapping function for the first network of the first local network layer, L1The structure of (2) can be similar to the global network layer function F, and is also compounded by a series of functions. According to the predicted value of the position of the first layer sub-aperture zone of the local network layer, the average zone sub-aperture position deviation of the layer can be obtained
Can pass through
Updating the initial value P of the sub-aperture zone position0And obtaining a current estimated value P.
For the k local network layer, based on the obtained deviation between the predicted value and the true value of the sub-aperture zone position of the previous network layer, i.e. delta Pk＝ΔPk(x)＝P(x)-Pk-1(x) In that respect Each convolutional neural network of each local network layer of the cascaded deep network framework can be optimized according to an optimization relation:
Δpk＝Δpk(x)＝P(x)-Pk-1(x)；
in the formula (I), the compound is shown in the specification,
is a non-linear function mean square error minimization function, Δ pkIs the average ring zone position deviation of the k local network layer, P (x) is the true position value corresponding to the ring zone image, Pk-1(x) For the position estimation value corresponding to the k-1 th local network layer ring zone image, lkAs a mapping function for the kth convolutional neural network, LkIs the k-thMapping function of the first network of the local network layer LkThe structure of (2) can also be similar to the global network layer function F, and is also compounded by a series of functions.
Based on the above embodiments, the structures of the convolutional neural networks in the global network layer are similar, but the areas of the convolutional neural networks input into the layer shape error image are different, and the different input areas of the shape error image can cover as many possibilities as possible. Therefore, the method not only can better help to predict the initial zone position estimation, but also can provide a priori knowledge for the deviation prediction of the next local network layer. The initial zone position value obtained by the global network layer is more accurate but insufficient compared with the zone position value initialized randomly or averagely. Therefore, it is necessary to design the local network layer learning zone position deviation to fine tune the initialized zone position value so that it approaches the real value step by step. Theoretically, the more local network layers are, the more the fine tuning times are, and the initial value of the girdle position will approach the true value of the girdle position infinitely. Then, it can be known from many experiments that, in the actual situation, after the number of local network layers reaches a certain number, the estimation effect does not become better, and instead, the calculation efficiency may be reduced due to the gradual increase of the resolution of the surface shape error image, so that a plurality of local network layers are not needed. Based on the above, after the initial position of the global network layer and the deviation value of each local network layer are obtained, the estimation value of each network layer can be calculated by adopting multi-layer regression. That is, the position calculation layer can calculate the annular zone position estimation value of the current annular sub-aperture at each layer by using the annular zone position estimation relational expression; then, for n in the i-th layer networkiThe calculation formula of the estimated value, the zone position estimation, can be expressed as follows:
in the formula, pnIs an estimate of the location of the annulus,
the position of the ring band of the ith annular sub-aperture of the ith network layer, n is the total number of layers of the convolutional neural network, i is the serial number of the current network layer, and n is the serial number of the current network layeriIn order to be the neural network of the i-th layer,
is the deviation value between the zone position of the ith annular sub-aperture and the real zone position. If n is 3, n1＝n2＝n33. The method can be divided into two parts, wherein the first part is an average item and corresponds to the average value of predicted values output by each convolutional neural network in the global network layer, namely an initialized zone position estimated value. The second term is the sum of average terms, which corresponds to the sum of average values of predicted values output by the convolutional neural network in the subsequent local network layer and is used for fine tuning the deviation value of the girdle position.
It can be understood that for an actual aspheric element, the profile detection situation is more complicated due to inevitable profile errors of the aspheric surface. If the annular subaperture positioning is carried out by using the deep neural network model trained by the ideal aspheric annular subaperture splicing model, a certain positioning error can be generated, so that the splicing result is influenced, and the accuracy of the surface shape error detection result is finally reduced. In view of this, in the process of training the surface shape distribution calculation model, the problem of low accuracy of the surface shape detection result due to the existence of the aspheric surface shape error can be solved by introducing the aspheric surface shape error. The aspheric surface shape error corresponding to each surface shape error image in the sample data set can be expressed by using a predefined relational expression, wherein the predefined relational expression is as follows:
that is, the aspheric surface shape error is developed as a Zernike polynomial, which in a polar coordinate system can be defined as:
wherein V (x, y) is a surface shape error, anmIs a coefficient, XnmTo describe the orthogonal polynomials of the surface shape, m and n are positive integers.
And carrying out rotation average processing on the aspheric surface shape error of each surface shape error image so as to train a deep learning network model by utilizing the circular subaperture after rotation average to obtain a surface shape distribution calculation model. And performing rotary average processing on all the items, wherein the other items are zero except that the size of the rotation symmetry item is equal to that of the original rotation symmetry item. Therefore, the annular sub-aperture after the rotary averaging is used for sub-aperture splicing depth neural network modeling of the aspheric surface shape error, and therefore modeling errors introduced by the aspheric surface shape error can be greatly reduced.
The embodiment of the invention also provides a corresponding implementation device for the aspheric element surface shape detection method, so that the method has higher practicability. The following describes an aspheric surface shape detection device according to an embodiment of the present invention, and the aspheric surface shape detection device described below and the aspheric surface shape detection method described above may be referred to correspondingly.
Referring to fig. 4, fig. 4 is a structural diagram of an aspheric surface element surface shape detecting device according to an embodiment of the present invention, where the device includes:
the model training module 401 is configured to train the deep learning network model by using the sample data set to obtain a surface shape distribution calculation model; the sample data set is a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result, and the deep learning network model comprises a cascade deep network frame and a position calculation layer; the cascade depth network framework is used for estimating the initial position and the position deviation value of the annular sub-aperture layer by layer, the position calculation layer is used for fine-tuning the ring belt position of each annular sub-aperture based on the initial position and the position deviation value of the annular sub-aperture by utilizing a multi-level regression algorithm.
And the surface shape distribution information calculation module 402 is configured to input sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model, so as to obtain a relative position and an overlapping area of each sub-aperture.
And a surface shape detection module 403, configured to perform sub-aperture stitching by using a sub-aperture stitching algorithm based on the relative position and the overlapping area of each sub-aperture, so as to implement surface shape detection on the aspheric element to be detected.
Optionally, in some embodiments of this embodiment, the cascaded deep network framework may include a global network layer and a plurality of local network layers;
the global network layer can be used for directly mapping the surface shape error value to the annulus position parameter by learning a nonlinear mapping function, and calculating to obtain an initial annulus position estimation value and a position deviation value of each annular sub-aperture; each local network layer may be used to update the annulus position value for each annular sub-aperture by gradually iterating the deviation value between the annulus position of the current annular sub-aperture and the true annulus position.
In some embodiments of this embodiment, the global network layer may further include, for example, a first convolutional neural network, a second convolutional neural network, and a third convolutional neural network;
the first convolutional neural network, the second convolutional neural network and the third convolutional neural network sequentially comprise an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and an output layer; the input data of the input layer of the first convolutional neural network is a whole surface-shaped error image, the input data of the input layer of the second convolutional neural network is a first preset area and a second preset area of the surface-shaped error image, and the input data of the input layer of the third convolutional neural network is a third preset area and a fourth preset area of the surface-shaped error image.
In other embodiments of this embodiment, the structure of each local network layer of the cascaded deep network framework is the same, and the convolutional neural network in the first network of each local network layer is used to learn the mapping function from the surface shape error image space to the annular sub-aperture annulus position deviation parameter space.
Each convolutional neural network of each local network layer of the cascaded deep network framework can be optimized according to an optimization relation:
Δpk＝Δpk(x)＝P(x)-Pk-1(x)；
in the formula (I), the compound is shown in the specification,
is a non-linear function mean square error minimization function, Δ pkIs the average ring zone position deviation of the k local network layer, P (x) is the true position value corresponding to the ring zone image, Pk-1(x) For the position estimation value corresponding to the k-1 th local network layer ring zone image, lkAs a mapping function for the kth convolutional neural network, LkA mapping function for the first network of the kth local network layer.
As a preferred embodiment, the position calculation layer calculates the zone position estimation value of the current annular sub-aperture at each layer by using the zone position estimation relation; the annulus position estimation relationship is:
in the formula, pnIs an estimate of the location of the annulus,
the position of the ring band of the ith annular sub-aperture of the ith network layer, n is the total number of layers of the convolutional neural network, i is the serial number of the current network layer, and n is the serial number of the current network layeriIn order to be the neural network of the i-th layer,
is the deviation value between the zone position of the ith annular sub-aperture and the real zone position.
Optionally, in another embodiment, the model training module 401 may include:
the aspheric surface shape error representation submodule is used for representing the aspheric surface shape errors corresponding to the surface shape error images in the sample data set by using a predefined relational expression, and the predefined relational expression is as follows:
the training module is used for carrying out rotary average processing on the aspheric surface shape errors of each surface shape error image so as to train a deep learning network model by utilizing the circular subaperture after rotary average to obtain a surface shape distribution calculation model;
wherein V (x, y) is a surface shape error, anmIs a coefficient, XnmTo describe the orthogonal polynomials of the surface shape, m and n are positive integers.
The functions of the functional modules of the aspheric element surface shape detection device according to the embodiments of the present invention may be specifically implemented according to the method in the foregoing method embodiments, and the specific implementation process may refer to the related description of the foregoing method embodiments, which is not described herein again.
Therefore, the embodiment of the invention realizes automatic, rapid and precise determination of the sub-aperture surface shape distribution information, effectively reduces the aspheric surface shape detection cost on the basis of ensuring the aspheric surface shape detection accuracy, and is very suitable for batch production detection.
The embodiment of the present invention further provides an aspheric surface element surface shape detection device, which specifically includes:
a memory for storing a computer program;
a processor for executing a computer program to implement the steps of the aspheric element surface shape detection method according to any one of the above embodiments.
The functions of each functional module of the aspheric element surface shape detection device according to the embodiment of the present invention may be specifically implemented according to the method in the foregoing method embodiment, and the specific implementation process may refer to the related description of the foregoing method embodiment, which is not described herein again.
Therefore, the embodiment of the invention realizes automatic, rapid and precise determination of the sub-aperture surface shape distribution information, effectively reduces the aspheric surface shape detection cost on the basis of ensuring the aspheric surface shape detection accuracy, and is very suitable for batch production detection.
The embodiment of the present invention further provides a computer-readable storage medium, in which a surface shape detection program of an aspheric element is stored, and the steps of the aspheric element surface shape detection method according to any one of the above embodiments are performed by a processor.
The functions of the functional modules of the computer-readable storage medium according to the embodiment of the present invention may be specifically implemented according to the method in the foregoing method embodiment, and the specific implementation process may refer to the related description of the foregoing method embodiment, which is not described herein again.
Therefore, the embodiment of the invention realizes automatic, rapid and precise determination of the sub-aperture surface shape distribution information, effectively reduces the aspheric surface shape detection cost on the basis of ensuring the aspheric surface shape detection accuracy, and is very suitable for batch production detection.
The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts among the embodiments are referred to each other. The device disclosed by the embodiment corresponds to the method disclosed by the embodiment, so that the description is simple, and the relevant points can be referred to the method part for description.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The steps of a method or algorithm described in connection with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module may reside in Random Access Memory (RAM), memory, Read Only Memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.
The present invention provides a method, an apparatus, a device and a computer readable storage medium for detecting aspheric surface shape. The principles and embodiments of the present invention are explained herein using specific examples, which are presented only to assist in understanding the method and its core concepts. It should be noted that, for those skilled in the art, it is possible to make various improvements and modifications to the present invention without departing from the principle of the present invention, and those improvements and modifications also fall within the scope of the claims of the present invention.
Claims (10)
1. A method for detecting the surface shape of an aspheric element is characterized by comprising the following steps:
training a deep learning network model by utilizing a sample data set in advance to obtain a surface shape distribution calculation model;
inputting sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model to obtain the relative position and the overlapping area of each sub-aperture;
based on the relative position and the overlapping area of each sub-aperture, sub-aperture splicing is carried out by utilizing a sub-aperture splicing algorithm so as to realize surface shape detection of the aspheric element to be detected;
the deep learning network model comprises a cascade depth network framework and a position calculation layer, wherein the sample data set comprises a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result; the cascade depth network framework is used for estimating the initial position and the position deviation value of the annular sub-aperture layer by layer, the position calculation layer is used for fine-tuning the ring belt position of each annular sub-aperture based on the initial position and the position deviation value of the annular sub-aperture by utilizing a multi-level regression algorithm.
2. The aspheric element surface shape detection method according to claim 1, wherein the cascaded deep network framework comprises a global network layer and a plurality of local network layers;
the global network layer is used for directly mapping the surface shape error value to the annulus position parameter by learning a nonlinear mapping function, and calculating to obtain an initial annulus position estimation value and a position deviation value of each annular sub-aperture;
each local network layer is used for updating the zone position value of each annular sub-aperture by gradually iterating the deviation value between the zone position of the current annular sub-aperture and the real zone position.
3. The aspheric surface element surface shape detection method according to claim 2, characterized in that the global network layer comprises a first convolutional neural network, a second convolutional neural network and a third convolutional neural network;
the first convolutional neural network, the second convolutional neural network and the third convolutional neural network sequentially comprise an input layer, a first convolutional layer, a first pooling layer, a second convolutional layer, a second pooling layer and an output layer; the input data of the input layer of the first convolutional neural network is a whole surface-shaped error image, the input data of the input layer of the second convolutional neural network is a first preset area and a second preset area of the surface-shaped error image, and the input data of the input layer of the third convolutional neural network is a third preset area and a fourth preset area of the surface-shaped error image.
4. The aspheric surface element surface shape detection method as claimed in claim 2, characterized in that the structure of each local network layer of the cascaded depth network framework is the same, and the convolutional neural network in the first network of each local network layer is used for learning the mapping function from the surface shape error image space to the annular position deviation parameter space of the annular sub-aperture.
5. The aspheric element surface shape detection method as claimed in claim 4, characterized in that each convolutional neural network of each local network layer of the cascaded deep network framework is optimized according to an optimization relation:
Δpk＝Δpk(x)＝P(x)-Pk-1(x)；
in the formula (I), the compound is shown in the specification,
is a non-linear function mean square error minimization function, Δ pkIs the average ring zone position deviation of the k local network layer, P (x) is the true position value corresponding to the ring zone image, Pk-1(x) For the position estimation value corresponding to the k-1 th local network layer ring zone image, lKMapping function for Kth convolutional neural network, LkA mapping function for the first network of the kth local network layer.
6. The aspheric surface element surface shape detection method as claimed in any one of claims 1 to 5, wherein the position calculation layer calculates the zone position estimation value of the current annular subaperture at each layer by using the zone position estimation relation; the annulus position estimation relation is:
in the formula, pnIs an estimate of the location of the annulus,
the position of the ring band of the ith annular sub-aperture of the ith network layer, n is the total number of layers of the convolutional neural network, i is the serial number of the current network layer, and n is the serial number of the current network layeriIn order to be the neural network of the i-th layer,
is the deviation value between the zone position of the ith annular sub-aperture and the real zone position.
7. The aspheric surface element surface shape detection method according to any one of claims 1 to 5, wherein the training process of the surface shape distribution calculation model comprises:
expressing the aspheric surface shape errors corresponding to the surface shape error images in the sample data set by using a predefined relational expression, wherein the predefined relational expression is as follows:
carrying out rotation average processing on the aspheric surface shape error of each surface shape error image so as to train the deep learning network model by using the circular subaperture after rotation average to obtain the surface shape distribution calculation model;
wherein V (x, y) is a surface shape error, anmIs a coefficient, XnmTo describe the orthogonal polynomials of the surface shape, m and n are positive integers.
8. An aspheric element surface shape detection device, comprising:
the model training module is used for training the deep learning network model by utilizing the sample data set to obtain a surface shape distribution calculation model; the sample data set is a plurality of surface shape error images representing the mapping relation between the annular sub-aperture position and the aspheric surface shape detection result, and the deep learning network model comprises a cascade deep network frame and a position calculation layer; the cascade depth network framework is used for estimating initial positions and position deviation values of the annular sub-apertures layer by layer, the position calculation layer is used for fine-tuning the ring belt positions of the annular sub-apertures based on the initial positions and the position deviation values of the annular sub-apertures by utilizing a multi-level regression algorithm;
the surface shape distribution information calculation module is used for inputting the sub-aperture detection data of the aspheric element to be detected into the surface shape distribution calculation model to obtain the relative position and the overlapping area of each sub-aperture;
and the surface shape detection module is used for splicing the sub-apertures by using a sub-aperture splicing algorithm based on the relative position and the overlapping area of each sub-aperture so as to realize the surface shape detection of the aspheric element to be detected.
9. An aspherical-element surface shape detection apparatus characterized by comprising a processor for implementing the steps of the aspherical-element surface shape detection method according to any one of claims 1 to 7 when executing a computer program stored in a memory.
10. A computer-readable storage medium, wherein the computer-readable storage medium stores thereon an aspheric element profile detection program, which when executed by a processor implements the steps of the aspheric element profile detection method according to any one of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910931375.2A CN110487211B (en) | 2019-09-29 | 2019-09-29 | Aspheric element surface shape detection method, device and equipment and readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910931375.2A CN110487211B (en) | 2019-09-29 | 2019-09-29 | Aspheric element surface shape detection method, device and equipment and readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110487211A CN110487211A (en) | 2019-11-22 |
| CN110487211B true CN110487211B (en) | 2020-07-24 |
Family
ID=68544570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910931375.2A Active CN110487211B (en) | 2019-09-29 | 2019-09-29 | Aspheric element surface shape detection method, device and equipment and readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110487211B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111290118A (en) * | 2020-03-10 | 2020-06-16 | 北京空间机电研究所 | Decoupling control method and device for deformable mirror |
| CN112985308B (en) * | 2021-05-12 | 2021-11-19 | 中国人民解放军国防科技大学 | Intelligent compensation interference detection system and method for local large surface shape error |
| CN113689495A (en) * | 2021-08-03 | 2021-11-23 | 湖南视比特机器人有限公司 | Hole center detection method based on deep learning and hole center detection device thereof |
| CN114593693B (en) * | 2022-02-16 | 2022-11-01 | 苏州英示测量科技有限公司 | Optical deflection detection method, electronic device and optical deflection detection system |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5071251A (en) * | 1989-06-12 | 1991-12-10 | California Institute Of Technology | Wavelength independent interferometer |
| EP2525188B1 (en) * | 2011-05-20 | 2014-08-20 | Baden-Württemberg Stiftung gGmbH | Adjustment of an optical surface to be tested in a test device |
| CN105241396A (en) * | 2015-10-20 | 2016-01-13 | 北京航空航天大学 | Digital hologram-based high-precision spherical surface sub-aperture splicing fusion method |
| CN106802136A (en) * | 2017-01-16 | 2017-06-06 | 暨南大学 | One kind is based on Legendre's Fourier polynomial cylinder interference joining method and system |
| CA2973074A1 (en) * | 2016-07-13 | 2018-01-13 | Sightline Innovation Inc. | System and method for surface inspection |
| CN107808122A (en) * | 2017-09-30 | 2018-03-16 | 中国科学院长春光学精密机械与物理研究所 | Method for tracking target and device |
| EP3432215A1 (en) * | 2017-07-17 | 2019-01-23 | Siemens Healthcare GmbH | Automated measurement based on deep learning |
| CN109291657A (en) * | 2018-09-11 | 2019-02-01 | 东华大学 | Laser Jet system is identified based on convolutional neural networks space structure part industry Internet of Things |
| CN109949363A (en) * | 2019-03-26 | 2019-06-28 | 北京遥感设备研究所 | A kind of object recognition and detection system suitable for Terahertz bionic compound eyes imaging system |
-
2019
- 2019-09-29 CN CN201910931375.2A patent/CN110487211B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5071251A (en) * | 1989-06-12 | 1991-12-10 | California Institute Of Technology | Wavelength independent interferometer |
| EP2525188B1 (en) * | 2011-05-20 | 2014-08-20 | Baden-Württemberg Stiftung gGmbH | Adjustment of an optical surface to be tested in a test device |
| CN105241396A (en) * | 2015-10-20 | 2016-01-13 | 北京航空航天大学 | Digital hologram-based high-precision spherical surface sub-aperture splicing fusion method |
| CA2973074A1 (en) * | 2016-07-13 | 2018-01-13 | Sightline Innovation Inc. | System and method for surface inspection |
| CN106802136A (en) * | 2017-01-16 | 2017-06-06 | 暨南大学 | One kind is based on Legendre's Fourier polynomial cylinder interference joining method and system |
| EP3432215A1 (en) * | 2017-07-17 | 2019-01-23 | Siemens Healthcare GmbH | Automated measurement based on deep learning |
| CN107808122A (en) * | 2017-09-30 | 2018-03-16 | 中国科学院长春光学精密机械与物理研究所 | Method for tracking target and device |
| CN109291657A (en) * | 2018-09-11 | 2019-02-01 | 东华大学 | Laser Jet system is identified based on convolutional neural networks space structure part industry Internet of Things |
| CN109949363A (en) * | 2019-03-26 | 2019-06-28 | 北京遥感设备研究所 | A kind of object recognition and detection system suitable for Terahertz bionic compound eyes imaging system |
Non-Patent Citations (4)
| Title |
|---|
| Automatic extraction of impervious surfaces from high resolution remote sensing images based on deep learning;Fenghua Huang等;《J. Vis. Commun. Image R.》;20181117;第453-461页 * |
| Training convolutional neural network from multi-domain contour images for 3D shape retrieval;Zhu Zongxiao等;《PATTERN RECOGNITION LETTERS》;20170901;第119卷;第41-48页 * |
| 一种光场图像空间和角度分辨率重建方法;郑祥祥等;《电脑知识与技术》;20170430;第13卷(第12期);第171-173页 * |
| 计算机再现全息与辅助球面混合补偿检测凸非球面方法研究;李明等;《光学学报》;20151130;第35卷(第11期);第1122001-1~1122001-8页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110487211A (en) | 2019-11-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110487211B (en) | Aspheric element surface shape detection method, device and equipment and readable storage medium | |
| CN110335297B (en) | Point cloud registration method based on feature extraction | |
| US7619749B2 (en) | Methods and apparatus of aligning surfaces | |
| CN110188321B (en) | Primary and secondary mirror calibration method based on neural network algorithm | |
| O’Leary et al. | Direct and specific least-square fitting of hyperbolæ and ellipses | |
| CN104019799B (en) | A kind of relative orientation method utilizing local parameter optimization to calculate basis matrix | |
| US20150292979A1 (en) | Analysing and machining an optical profile | |
| Zhang et al. | A novel method for repeatedly generating speckle patterns used in digital image correlation | |
| CN109099857B (en) | Subaperture splicing method based on SURF feature matching | |
| CN104050643A (en) | Remote-sensing image relative correction method and system integrating geometry and radiation | |
| CN103258327B (en) | A kind of single-point calibration method based on two degrees of freedom video camera | |
| Ding et al. | A robust registration algorithm of point clouds based on adaptive distance function for surface inspection | |
| CN109341588B (en) | Binocular structured light three-system method visual angle weighted three-dimensional contour measurement method | |
| CN106023146B (en) | For the related unilateral self-calibration bundle adjustment method in field in photogrammetric | |
| CN113468802A (en) | Intelligent optical active debugging method based on point spread function | |
| CN114022552A (en) | Target positioning method and related device integrating laser radar and camera | |
| JP6951469B2 (en) | How to calibrate an optical measuring device | |
| CN108595373B (en) | Uncontrolled DEM registration method | |
| CN115682991A (en) | Method for detecting non-uniformity of curvature radius of each sub-mirror of spliced telescope in orbit | |
| CN110285808A (en) | A kind of instrument magnitude conversion method for star sensor based on infinite dimensional space | |
| Yang et al. | Error suppression for speckle correlation-based 3D shape measurement via dynamic shape function selection | |
| CN108764161B (en) | Remote sensing image processing method and device for breaking pathological singularity caused by sparse array based on polar coordinate system | |
| Stamatopoulos et al. | On the self-calibration of long focal length lenses | |
| CN112199814B (en) | System error self-checking method, device, equipment and medium for measuring system | |
| CN111336963B (en) | Error compensation method and measurement error evaluation method based on biaxial symmetry standard device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |