CN112914497B - Xerophthalmia machine inspection device and using method - Google Patents

Abstract

The invention relates to a xerophthalmia machine inspection device and a xerophthalmia machine inspection method, which at least comprise an analysis storage module, wherein the analysis storage module can obtain an original data packet according to a collected lesion tissue picture, and can preprocess collected image data and construct a hybrid network algorithm model on the basis of establishing an information base; the analysis storage module trains and tests the established hybrid network algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition when the algorithm model is established, thereby improving the accuracy of the calculation or prediction result of the algorithm model; when the image data transmitted by the optical imaging component is received, a group of newly acquired image data and stored sample information are compared and analyzed through a hybrid network algorithm model, and when the image data which is not stored is acquired, the newly acquired image data is learned and stored in an information base.

Description

Xerophthalmia machine inspection device and using method

Technical Field

The invention relates to the technical field of medical detection, in particular to a xerophthalmia machine detection device and a manufacturing method thereof.

Background

At present, due to the daily life requirements of people in China, the popularization of electronic equipment, the reduction of air quality and other factors, the number of ophthalmic patients in China is caused to show a trend of rapid increase recently, especially, people use electronic equipment such as mobile phones for a long time, the workload of eyes is increased more and more, the number of times of eyes being hit due to long-time injection to an electronic screen is reduced, in addition, due to the arrangement of intelligent home and working environment, people are often in a dry air-conditioned room and an electronic radiation environment, tear evaporation is caused to be too fast, and the possibility of eye discomfort and ophthalmic diseases of people is further increased remarkably. Among them, dry eye disease is an ophthalmic disease whose prevalence rate is increasing rapidly at present, and is often accompanied by various ophthalmic diseases such as ocular discomfort and lesions of ocular surface tissues. Thus, when conditions such as a decrease in the number of eye transients and rapid tear evaporation occur, dry eye related disorders often further worsen. However, the causes of the dry eye disease are complex and various at present, so that the pathogenesis and the disease symptoms of the dry eye disease are numerous and complicated, especially affected by the environment of regions, the early symptoms of each region have large differences, but at present, a complete and comprehensive judgment standard is not provided, and a dry eye diagnosis device capable of comprehensively detecting and summarizing and analyzing various disease symptoms is not provided. The existing product mainly diagnoses the loaded disease symptoms as the basis for assisting medical care personnel, but cannot reasonably and effectively diagnose the generated new early disease symptoms, and the system also has no self-learning capability and cannot learn and store according to the captured new early disease symptoms, so that the diagnosis capability of the system is improved.

Chinese patent CN111161278A discloses a fundus image focus segmentation method based on deep network convergence, which comprises the steps of carrying out manual segmentation on the focus outline of an obtained fundus focus image, and constructing a training set and a testing set according to the segmentation result; then, after a deep aggregation network module is added in a backbone network of the U-Net model, the obtained U-Net model is transferred to focus segmentation of the eye fundus image, the U-Net model is trained, and the trained U-Net model is used as an eye fundus image focus segmentation model; and finally, segmenting the fundus image to be segmented by using the fundus image focus segmentation model.

In the prior art, when the acquired image is processed by using a neural network, irrelevant factors at the periphery of the image are usually removed by using the neural network, and a more accurate and effective focus segmentation image is obtained by improving a neural algorithm, however, the prior art does not adopt a technical scheme of learning, storing and updating a database of the acquired case with specific parameters which is not incorporated into an information base, and the case with the specific parameters which appears for the first time cannot be efficiently and accurately identified, so that a calculation model with strong learning capability and analysis capability is required, the acquired latest version of data can be stored and updated while a novel case is identified and analyzed, an intelligent judgment list of the eye image of a patient under examination is enhanced, and a more complete detection auxiliary function is provided for medical care personnel.

Furthermore, on the one hand, due to the differences in understanding to those skilled in the art; on the other hand, since the inventor has studied a lot of documents and patents when making the present invention, but the space is not limited to the details and contents listed in the above, however, the present invention is by no means free of the features of the prior art, but the present invention has been provided with all the features of the prior art, and the applicant reserves the right to increase the related prior art in the background.

Disclosure of Invention

Aiming at the defects of the prior art, the application provides a xerophthalmia machine inspection device, which is based on the image recognition and the time sequence progressive data processing learning capacity of various neural networks and memory networks by constructing a hybrid network algorithm model, so that the intelligent judgment of an intelligent system on the tissue picture of a patient under inspection is increased, and a more perfect detection auxiliary function is provided for medical staff.

The technical scheme of the invention is to provide a dry eye mechanical examination device, which at least comprises an analysis storage module, an illumination assembly and an optical imaging assembly, wherein the analysis storage module can store and learn collected diseased tissue pictures of a dry eye patient, the illumination assembly forms a dry eye detection handheld device, and part of imaging channels are arranged in the illumination assembly; the analysis storage module can obtain an original data packet according to the collected lesion tissue picture, preprocess the collected image data and construct a hybrid network algorithm model on the basis of establishing an information base; the analysis storage module trains and tests the established hybrid network algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition when the algorithm model is established, thereby improving the accuracy of the calculation or prediction result of the algorithm model; when the image data transmitted by the optical imaging component is received, a group of newly acquired image data and the stored sample information are compared and analyzed through a hybrid network algorithm model, and when the new image data which is not stored is acquired, the newly acquired image data is learned and stored in an information base.

According to a preferred embodiment, the information base stores data of a plurality of groups of images of tear film rupture along time input by the optical imaging assembly or the data port, samples which can be set to be a plurality of groups of different calibration parameters according to different collected tear film rupture positions and rupture process durations are analyzed and processed in a data mode to establish a hybrid network algorithm model; in the process of constructing the hybrid network algorithm model, at least two selected neural network systems are modeled in a mode of being connected in series, wherein the selected neural network systems can at least respectively extract time domain information from a time dimension and extract space domain information from the same frame.

According to a preferred embodiment, the constructed spatial domain information extraction model comprises an input layer, a convolution layer, a full connection layer and an input layer, wherein the full connection layer is built at the last part of the spatial domain information extraction model, and the output layer is arranged at the downstream of the full connection layer; the constructed time domain information extraction model comprises a long and short memory layer and at least one full connection layer; when the images acquired by the light imaging assembly are compared and analyzed, the characteristic information in the acquired images is quickly identified, and the domain information extraction model adopts a repeated module chain form to quickly extend and supplement the relevant data of the characteristic information; according to the obtained image information of the specific parameters which are not recorded in the process of identification and analysis, learning is carried out, feedback data are obtained, correction is carried out according to the accuracy of the data, and therefore the model is continuously perfected.

According to a preferred embodiment, the hybrid network algorithm model performs information extraction of the collected images in a mode of performing cross extraction on changed parameter features in a group of image data, so that when one domain of information is extracted, too much information of other domains is not lost, so that features on each image in a group of tear film images are changed, the changed features are extracted in sequence, and the extracted features are converted from bottom-level features to high-level semantic features; the analysis storage module can select at least two network systems from the existing network systems such as CNN, LSTM or RNN according to different requirements of image data processing executed by the analysis storage module to construct a hybrid network algorithm model capable of extracting specified features.

According to a preferred embodiment, the lighting assembly comprises a lighting source, a housing and a light modulation module which can be inserted into the housing to change the wavelength of the light emitted by the light source, wherein a light source panel formed by the lighting source is arranged in the internal cavity channel of the housing, and the light modulation module adjusts the wavelength of the light emitted by the lighting source in a manner of being selectively inserted into the internal cavity channel of the housing so as to obtain the light with different wavelengths and colors as required; the light source panel formed by the illumination light source is provided with a plurality of groups of monochromatic light LED lamp units which are annularly arranged in a mode of obtaining light with uniform brightness irradiated along the extending direction of the light channel formed by the shell, a plurality of groups of annular lamps are arranged in a mode of forming concentric circles, and the center of the light source panel is provided with a through hole for installing an imaging channel of the optical imaging assembly; the lighting assembly further comprises a light diffusion plate parallel to the cross section of the shell, the light diffusion plate is embedded in the light channel formed inside the shell, the light diffusion plate is perpendicular to the extending direction of the light channel, so that light emitted by the lighting source can penetrate through the light diffusion plate along the extending direction of the light channel, light emitted by the lighting source sequentially penetrates through the conversion plate and the light diffusion plate in the shell, the light converted by the conversion plate can further pass through the light diffusion plate to be subjected to light uniformity adjustment, and the light further uniformly diffused is emitted out of the light channel. Through the annular light source that arranges that only sets up single wavelength, greatly reduced the cost that the light source set up can launch comparatively even light simultaneously, simultaneously through setting up the homogeneity that the improvement light diffusion that the light diffuser plate can be further for the light that shines or the luminance of projection image is more even.

According to a preferred embodiment, at least two light modulation modules which are positioned on a set cross section of the shell and are the same as the cross section and are communicated with the interior of the shell are arranged on the set cross section of the shell, each light modulation module comprises a limiting shell, a conversion plate and a pressing assembly, the converting plate is arranged in the limiting shell, the limiting shell is arranged on the surface of the shell, an inner cavity of the limiting shell is communicated with an inner cavity of the shell, and the pressing assembly is connected with the limiting shell in a manner that part of a body of the pressing assembly is inserted from an opening at the top of the limiting shell and is connected with the conversion plate; the pressing assembly can push the conversion plate to be inserted into the shell from the limiting shell under the action of first external force pressing, so that light rays in the shell are converted; the pressing assembly can pull the conversion plate to return to the limiting shell from the shell after being pressed by second external force. Through setting up the light conversion module that can carry out the conversion to light source light, can reduce the demand of lamp light source quantity when reducing lamp light source control switch to the realization only needs the light source setting of single wavelength just can acquire the demand of the different colour light of multiple wavelength.

According to a preferred embodiment, the two light modulation modules are respectively provided with conversion plates with different light conversion effects, and the two conversion plates are respectively provided with a first conversion sheet and a second conversion sheet, wherein when the first conversion sheet is inserted into the light channel in the housing, the first conversion sheet can convert the first light emitted by the illumination light source into a second light; when the second conversion sheet is inserted into the light channel in the shell, the second conversion sheet can convert the first light emitted by the illumination light source into third light; when the first conversion sheet and the second conversion sheet are not inserted into the light channel, the first light emitted by the illumination light source directly irradiates the eye surface; the size of the first conversion sheet and the size of the second conversion sheet are equal to the cross sectional area of the illumination channel in the shell, through holes communicated with the imaging channel are respectively arranged at the respective face center positions of the first conversion sheet and the second conversion sheet, and a shielding coating which can be mutually butted with the pipeline wall of the imaging channel to form a lightless channel is coated on the annular surface of each through hole. Light modulation module sets up the conversion piece of multiple light conversion effect, can make things convenient for the operator to select the conversion piece of needs to obtain and set for light, especially the colour of different conversion pieces, difference such as sign, regulation mode during operation is different, can avoid the operator to make the operation of mistake under unconsciousness or inertia thinking effect effectively, in addition, the face heart at the conversion piece sets up the through hole that can communicate the formation of image passageway, especially coat the picture layer of shading on the annular wall in hole, can avoid the light in the light passageway directly to enter into the formation of image passageway effectively, influence the quality that the formation of image passageway gathered the image, light in the while can be with the formation of image passageway and the light in the light passageway separate completely.

According to a preferred embodiment, the optical imaging assembly comprises an image acquisition camera for receiving and acquiring images of the surface of eyes and tear film and an imaging objective lens for receiving images of the eyes and forming an imaging optical path so as to transmit and project a projection image onto the image acquisition camera, an imaging channel of the imaging objective lens is embedded into the shell in a manner that the axis of the imaging channel coincides with the axis of the shell, a shielding coating capable of shielding light rays emitted by the illumination light source from directly entering the imaging channel from the outer side of the tube wall of the imaging channel is coated on the tube wall of the imaging channel of the imaging objective lens, the tube forming the imaging channel consists of a plurality of sections of coaxial tubes, wherein a gap capable of being embedded into the first conversion sheet or the second conversion sheet is arranged at the connection position of the tubes, and the special light transmission preventing coating is coated in the imaging channel so as to effectively avoid light rays in the light channel from transmitting and influence the quality of the images acquired by the image acquisition camera.

According to a preferred embodiment, a Placido disc for forming a concentric annular projection with alternate light and dark is further arranged at one end of the shell far away from the illumination light source, and the Placido disc is detachably arranged at an opening position of a light channel penetrating through the end of the shell, so that light irradiated from the light channel can irradiate the Placido disc; the machine inspection device further comprises a holding part, the holding part is formed by a first body and a second body which are perpendicular to each other and form an L shape, one end of the first body of the holding part, far away from the second body, is close to the illuminating assembly, and is detachably connected with a port at one end of the illuminating light source, an image acquisition camera for receiving and acquiring images of eyes and tear film surfaces and an imaging objective lens of which part is connected with the image acquisition camera are arranged in the first body, the imaging objective lens penetrates through the end face of the first body and is inserted into the illuminating assembly, one end of the imaging objective lens, far away from the image acquisition camera, is flush with the outer surface of the Placido disc, the holding part can conveniently detect the holding of the device by an operator and the adjustment of a conversion sheet of a corresponding light conversion module, and can effectively install an optical imaging assembly, reduce the occupation of space, and is convenient to carry and use.

The technical scheme of the invention is to provide a manufacturing method of a dry eye disease machine detection device, which at least comprises an analysis storage module, an illumination component and an optical imaging component, wherein the analysis storage module can store and learn the collected dry eye disease patient lesion tissue pictures, the illumination component forms a dry eye disease detection handheld device, and part of imaging channels are arranged in the illumination component;

s1: obtaining an original data packet according to the collected lesion tissue picture;

s2: preprocessing collected image data and building an information base

S3: constructing a hybrid network algorithm model;

s4: training and testing the established hybrid network algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition when the algorithm model is established, thereby improving the accuracy of the calculation or prediction result of the algorithm model;

s5: when the image data transmitted by the optical imaging component is received, a group of newly acquired image data and stored sample information are compared and analyzed through a hybrid network algorithm model, and when the image data which is not stored is acquired, the newly acquired image data is learned and stored in an information base.

Drawings

FIG. 1 is a flow chart of image analysis based on a dry eye mechanical examination device

FIG. 2 is a schematic diagram of the structure of a preferred embodiment of a dry eye mechanical examination device of the present invention;

fig. 3 is a cross-sectional view of a preferred embodiment of a dry eye mechanical examination device of the present invention with a light modulation module attached;

FIG. 4 is a plan view of the illumination source of the preferred embodiment of the dry eye mechanical examination device of the present invention;

fig. 5 is a schematic structural diagram of a light modulation module of a preferred embodiment of a dry eye mechanical examination device of the present invention;

fig. 6 is a schematic structural view of a first adjustment body of a preferred embodiment of a dry eye mechanical examination device of the present invention;

fig. 7 is a schematic structural view of the second regulating body of the preferred embodiment of the dry eye mechanical examination device of the present invention.

List of reference numerals

1: the lighting assembly 2: optical imaging assembly 3: light modulation module

4: the grip portion 5: placido plate 6: analysis storage module

11: illumination light source 12: the housing 13: light diffusion plate

21: image-capturing camera 22: imaging objective lens 31: limit shell

32: the conversion plate 33: the pressing member 41: first body

42: second body 43: battery 44: SD card slot

45: the USB interface 46: the illumination switch 47: camera switch

321: first conversion sheet 322: second conversion sheet 331: first regulating body

332: second regulator 333: pressing the sleeve 334: connecting spring

335: first protrusion 336: saw tooth protrusion 337: hanging edge

Detailed Description

This is described in detail below with reference to fig. 1-7.

A dry eye mechanical examination device may include at least one of the following components: the device comprises an illumination assembly 1, an optical imaging assembly 2, a light modulation module 3, a holding part 4, a Placido plate 5 and an analysis storage module 6.

According to a specific embodiment, as shown in fig. 1, the analysis and storage module 6 is capable of storing and learning the collected images of the lesion tissue of the dry eye patient. In the case where the illumination light emitted from the illumination assembly 1 illuminates the patient's eye, the optical imaging assembly 2 collects an illumination image and/or a projection image generated by the illumination light illuminating the cornea. The optical imaging assembly 2 transmits the acquired image to the analysis and storage module 6 for analysis and processing so as to acquire relevant data with set parameter information. The analysis and storage module 6 can obtain the original data packet according to the collected lesion tissue picture. Preferably, the analysis storage module 6 selects a CNN (convolutional neural network) and an LSTM (long short term memory network) as the basis of the algorithm model, so as to preprocess the collected image data and construct a CNN-LSTM algorithm model on the basis of establishing the information base. The analysis storage module 6 trains and tests the established CNN-LSTM algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the condition of specific parameters existing during the establishment of the algorithm model, thereby improving the accuracy of the calculation or prediction result of the algorithm model. When receiving the image data transmitted by the optical imaging assembly 2, the newly acquired set of image data and the stored sample information are subjected to comparative analysis through the CNN-LSTM algorithm model. When image data which is not stored is acquired, newly acquired image data is learned and stored in an information base. The information base stores data as a series of image data acquired by the plurality of sets of optical imaging assemblies 2 or input by the data port that the tear film is broken over time. Samples can be set as sets of different calibration parameters depending on the location of tear film rupture and the duration of the rupture process that are collected. The CNN-LSTM algorithm model is built by analyzing and digitizing several different samples. In the process of constructing the CNN-LSTM algorithm model, the CNN neural network system and the LSTM neural network system are modeled in a serial mode. The LSTM neural network system extracts time domain information in a time dimension, and the CNN neural network system extracts space domain information in the same frame. The constructed CNN model comprises an input layer, a convolutional layer, a full-link layer and an input layer. The full connection layer is built at the last part of the CNN model, and the output layer is arranged at the downstream of the full connection layer. The constructed LSTM model comprises a long and short memory layer and at least one full connection layer. When the images collected by the light imaging assembly 2 are subjected to comparative analysis, the characteristic information in the collected images is rapidly identified. The LSTM is a fast extensive complement of the data associated with the feature information in the form of a chain of repeating modules. And learning according to the acquired image information of the specific parameters which are not recorded yet in the identification and analysis process, acquiring feedback data, and correcting according to the accuracy of the data to continuously perfect the model. The CNN-LSTM algorithm model carries out information extraction of the collected image according to a mode of carrying out cross extraction on changed parameter characteristics in a group of image data, so that when one domain of information is extracted, not much information of other domains is lost. The features on each image in a group of tear film images are changed, the changed features are extracted sequentially, and the extracted features are converted from the bottom-layer features to the high-layer semantic features. By constructing a CNN-LSTM algorithm model, the intelligent judgment of an intelligent system on the organization picture of a patient under examination is increased based on the image identification and the learning capability of time sequence progressive data processing of CNN and LSTM networks, and a more complete detection auxiliary function is provided for medical personnel. The CNN-LSTM algorithm model is established, detection and analysis can be performed, learning can be performed on different case pictures, and related data mining can be completed, so that the information acquisition efficiency is greatly improved, the problem that traditional case information is difficult to acquire is solved, the information acquisition accuracy is high due to the learning mechanism of the model, and the defects of the model can be perfected through autonomous learning.

As shown in fig. 2, the illumination module 1 includes an illumination light source 11, a housing 12, and a light modulation module 3 that can be inserted into a light passage inside the housing 12 to change the wavelength of light emitted from the illumination light source 11. The axis of the shell 12 forming the light rays diverging towards the designated direction is exactly coincided with the axis of the imaging channel of the optical imaging assembly 2 inserted into the illumination assembly 1, so that the images irradiated or projected by the light rays can be completely collected and transmitted by the imaging channel penetrating through the center of the cross section of the shell 12. A light source panel constituted by the illumination light source 11 is disposed in the internal cavity passage of the housing 13 in a manner perpendicular to the axis of the housing 13. The light modulation module 3 can adjust the wavelength of the light emitted by the illumination light source 11 in a manner of being perpendicular to the axis of the housing 12 and being selectively inserted into the inner cavity channel of the housing 12, so that the machine inspection device can emit light with a set wavelength and color according to requirements, such as cobalt blue light, white light or infrared light. Through setting up the light conversion module that can carry out the conversion to light source light, can reduce the demand of lamp light source quantity when reducing lamp light source control switch to the realization only needs the light source setting of single wavelength just can acquire the demand of the different colour light of multiple wavelength, has still avoided continuous repetitive cycle regulation and control switch to select suitable light and lead to the problem that the device life-span shortens simultaneously.

As shown in fig. 3, the light modulation module 3 includes a limiting case 31, a conversion plate 32 disposed in the limiting case 31, and a pressing member 33. The limiting case 31 is provided on the surface of the case 12 in such a manner that its inner chamber communicates with the light passage of the case 11. The pressing member 33 is inserted into the inner cavity of the stopper housing 31 from the top opening of the stopper housing 31 according to a part of its body and is connected to the conversion plate 32 disposed in the inner cavity of the stopper housing 31. Preferably, the pressing assembly 33 can push the conversion plate 32 to be inserted into the light channel of the housing 12 from the inner cavity of the limiting housing 31 under the pressing action of the first external force, so that the conversion plate 32 inserted into the housing 12 can just convert the light emitted by the illumination light source 11. After the first pressing, the pressing assembly 33 can receive the action of the second external pressing, and the conversion plate 32 returns to the limiting shell 31 along with the return movement of the pressing assembly 33. As shown in fig. 4, the light source panel that the light source constitutes is provided with a plurality of groups and is the monochromatic light LED lamp unit that the annular was arranged according to the mode that can acquire the light that shines out luminance uniformity along the extending direction of the light passageway that the casing constitutes, and cobalt blue light LED lamp is chooseed for use to this monochromatic light LED lamp to make the initial first light that sends be cobalt blue light. When the light modulation module 3 with different conversion plates 33 is used, the light modulation module is selectively inserted into a light channel, so that light with different wavelengths and colors can be finally irradiated by a light source which is set as a cobalt blue chip according to light requirements.

At least two light modulation modules 3 which are positioned on the same plane and are communicated with the inside of the light channel in the shell 12 are arranged on the surface of the shell 12. Preferably, the plane formed by the at least two light modulation modules 3 is exactly parallel to the cross section of the housing 12. The inside of two light modulation modules 3 is provided with the conversion board 32 of different light conversion effects respectively for press and adjust conversion board 32 of any light modulation module 3 to the time of in the light passageway, the light wavelength and the colour after the conversion are single inequality. Preferably, two different conversion plates 32 are provided with a first conversion sheet 321 and a second conversion sheet 322, respectively. Preferably, the second light and the third light converted from the cobalt blue light after passing through the conversion plate 32 are infrared light capable of accurately locating the tear film rupture position through the projection image and calculating the rupture time thereof, and observing the thickness change of the tear film lipid layer and white light for observing the actual condition of the meibomian gland based on the infrared light and the white light, respectively. When the first conversion sheet 321 is inserted into the light passage in the housing 12, the first conversion sheet 321 can convert the cobalt blue light emitted by the illumination light source 11 into a single infrared light having a set wavelength and capable of illuminating and displaying the eye plate gland so as to facilitate the optical imaging assembly 2 to collect the eye image. When the second conversion sheet 322 is inserted into the light passage in the housing 12, the second conversion sheet 322 can convert the cobalt blue light emitted from the illumination light source 11 into white light that can directly observe the actual condition of the meibomian glands. When the first conversion sheet 321 and the second conversion sheet 322 are not inserted into the light channel, the cobalt blue light emitted by the illumination light source 11 can directly irradiate the ocular surface dyed by the fluorescein sodium and the lissamine green, so that images for observing the cornea and conjunctiva damage condition of the eyes can be conveniently acquired; the first and second conversion sheets 321 and 322 have a size equal to the cross-sectional area of the light passage in the housing 12. And the respective face center positions of the first conversion sheet 321 and the second conversion sheet 322 are respectively provided with a through hole communicated with the imaging channel, and the ring surface of the through hole is coated with a shielding coating which can be mutually butted with the wall of the imaging channel to form a lightless channel. Preferably, the two light modulation modules 3 can set the housing to white and red respectively, so that an operator can visually know the color of the light converted by the conversion sheet corresponding to the light modulation modules. And the misoperation of the operator under the action of unconsciousness or inertial thinking can be effectively avoided. In addition, a through hole capable of communicating with the imaging channel is arranged in the center of the face of the conversion sheet. Especially, the shading layer is coated on the annular wall of the hole, so that the condition that light in the light channel directly enters the imaging channel to influence the quality of an image collected by the imaging channel can be effectively avoided, and meanwhile, the light in the imaging channel and the light in the light channel can be completely separated.

As shown in fig. 5, the pressing member 33 includes a first regulating body 331 and a second regulating body 332. The second adjusting body 332 is sleeved on the first adjusting body 331, and the first adjusting body 331 and the second adjusting body 332 are both limited in the movement space by the limiting shell 31, so that the second adjusting body 332 is set at the set position of the limiting shell 31. The first adjustment body 331 is capable of translation along its axial direction. The first adjuster 331 is rotatably connected to the conversion plate 32. When the first adjustment body 331 simultaneously performs relative rotation around the coaxial line and relative reciprocating translation along the axial direction with respect to the second adjustment body 332, the conversion plate 32 performs reciprocating motion driven by the first adjustment body 331 to be inserted into the light path of the housing 12 from the limiting housing 31 or to return into the limiting housing 31 from the light path of the housing 12. So that the light in the light channel can be converted or not converted according to the requirement by adjusting the conversion plate 32. The end of the first adjuster 331 remote from the conversion plate 32 is provided with a pressing lever 333. The pressing sleeve 333 is fitted to one end of the first adjuster 331, and the pressing sleeve 333 is partially inserted into the second adjuster 332. The first adjustment body 331 can only be translated in the axial direction under the co-constraint of the second adjustment body 332 and the pressing sleeve 333. The first regulator body 331 and the pressing sleeve 333 are also provided with a connecting spring 334. When the pressing sleeve 33 is pressed for the first time under the action of the first external force, the first adjuster 331 drives the conversion plate 32 to be inserted into the light path. As shown in fig. 6 to 7, the first boss 335 of the first adjustment body 331 passes over one side inclined surface of the saw tooth boss 336 of the second adjustment body 332 and rotates along the other side inclined surface, and the rotating first boss 335 is hooked to the hooking rib 337 protruded from the outer circumference of the second adjustment body 332. At this time, the first adjustment body 331 and the second adjustment body 332 are fixed relative to each other, and the conversion sheet 32 remains in the light path. When the pressing sleeve (33) is pressed again by the second external force, the first adjuster 331 rotates and causes the first projection 335 to change position. So that the first protrusions 335 are positioned between two adjacent hanging edges 337, and are translated by the resilient force of the connecting spring 334, and the conversion plate 32 is restored to the inside of the limiting shell 31.

The end of the shell 12 far away from the illumination light source 11 is also provided with a Placido plate 5 for forming a concentric annular projection with alternate light and shade. The Placido plate 5 is detachably mounted at the opening of the light passage through the end of the housing 12 so that the light emitted from the light passage can be vertically incident on the Placido plate 5. The illumination light source 11 and the light diffusion plate 13 are provided with through holes at the center of their faces, through which the imaging channel tubes constituting the imaging objective lens 22 can be inserted. The illumination light source 11 at least includes a plate light source composed of three annular LED lamp units. The surface of the illumination light source 11 is embedded in the housing 12 so as to be perpendicular to the axis of the housing 12. The distance between the light diffusion plate 13 and the illumination light source 11 is 10-15cm.

The optical imaging assembly 2 includes an image capture camera 21 for receiving captured eye and tear film surface images and an imaging objective lens 22 for receiving the eye images and forming an imaging optical path for transmitting a projection image onto the image capture camera 21. The imaging channel of the imaging objective 22 is embedded inside the housing 12 in such a way that the axis coincides with the axis of the housing 12. And the tube wall of the imaging channel of the imaging objective lens 22 is coated with a shielding coating which can shield the light emitted by the illumination light source 11 from directly entering the imaging channel from the outer side of the tube wall of the imaging channel. The pipe forming the imaging channel consists of a plurality of sections of coaxial pipes. The pipe connection position is provided with a gap into which the first conversion sheet 321 or the second conversion sheet 322 can be vertically inserted. Light transmission in the light channel can be effectively avoided by coating a special anti-light-transmission coating in the imaging channel, so that the quality of an image acquired by an image acquisition camera is influenced.

The machine inspection device also comprises a holding part 4. The grip portion 4 is integrally formed by a first body 41 and a second body 42 which are perpendicular to each other and constitute an L-shape. The end of the first body 41 of the grip 4 remote from the second body 42 is detachably connected to the port of the lighting assembly 1 near the end of the lighting source 11. An image acquisition camera 21 for receiving and acquiring images of the eye and the surface of the tear film and an imaging objective lens 22 partially connected with the image acquisition camera are arranged in the first body 41, and the imaging objective lens 22 penetrates through the end face of the first body 41 and is inserted into the lighting assembly 1. And the end of the imaging objective lens 22 remote from the image capturing camera 21 is flush with the outer surface of the Placido plate 5. The second body 42 for holding is provided with a battery 43 for supplying power to the image capturing camera 21 and the illumination light source 11, an SD card slot 44 for mounting an SD card storing an original captured image and being in data connection with the image capturing camera 21, and a USB interface 45 for transmitting an image. Preferably, the second body 42 is further provided with a switch for conveniently controlling the illumination assembly 1 and the optical imaging assembly 2. The battery 43 is connected to the illumination module 1 and the imaging module 2 through wires connected to an illumination switch 46 and a camera switch 47, respectively. During the use, turn on lighting switch 46 according to the demand, then carry out eye light projection or light direct irradiation, when needs gather the image, press down camera switch 47 through holding device's forefinger and carry out image acquisition, it can to loosen camera switch 47 after accomplishing the collection.

Example 1

An LSTM network based dry eye screening device may include at least one of the following: the device comprises an illumination assembly 1, an optical imaging assembly 2, a light modulation module 3, a holding part 4, a Placido plate 5 and an analysis storage module 6.

The optical imaging unit 2 is disposed inside the grip 4 and penetrates the axial center of the illumination unit 1. When the light image formed by the illumination light source 11 of the illumination assembly 1 irradiating the Placido plate 5 is positioned and projected on the tear film of the eye of the patient, the light irradiated on the tear film by the Placido plate 5 partially transmitted by the illumination assembly 1 can be reflected by the tear film. At this time, the optical imaging assembly 2 passing through the illumination assembly 1 can directly acquire the tear film surface image. Preferably, the surface of the lighting unit 1 close to the grip portion 4 is provided with an annular screw hole that fits the outer periphery of the housing of the grip portion 4, so that the lighting unit 1 and the grip portion 4 can be screwed together when the outer peripheral surface of the housing at one end of the grip portion 4 in the axial direction is provided with a male screw. The image acquired by the optical imaging assembly 2 can be sent directly to a designated PC having a dry eye detection unit.

Preferably, the optical imaging assembly 2 includes an image capture camera 21 for receiving an image of the surface of the tear film and an imaging objective 22 for receiving the projected pattern formed on the tear film and forming an imaging optical path for transmitting the projected pattern to the image capture camera 21. The image acquisition camera 21 and the imaging objective lens 22 are both arranged inside the housing of the grip 4 along the projection imaging optical path. The imaging objective lens 22 penetrates through the end shell of the holding part 4 and is partially embedded and installed at the axial center position of the illumination assembly 1, so that the holding part 4 is connected with the illumination assembly 1. When the lighting assembly 1 projects the image of the Placido plate 5 on the tear film of the eye of the patient, the light transmitted on the tear film from the gap of the Placido plate 5 by the lighting assembly 1 is reflected by the tear film, and then the image of the tear film surface is transmitted to the image acquisition camera 21 through the imaging objective lens 22 embedded on the axis position of the lighting assembly 1. When the tear film of the eye of the patient is complete and normal, the tear film image of the surface of the eye of the patient acquired by the image acquisition camera 21 is an image with light and shade lines spaced and a criss-cross line profile complete. When the tear film on the surface of the human eye is broken, the line contour on the tear film of the human eye collected by the image collecting camera is distorted and broken, and cannot form a complete image with uniformly distributed light and shade intervals. Furthermore, a series of imaging photos of the tear film on the surface of the eye of the patient, which are acquired within a set time by the image acquisition camera 21, from complete to rupture can be calculated and acquired through subsequent processing of the image algorithm of the corresponding LSTM network, so that whether the tear film rupture time of the patient is within a threshold range of the normal tear film rupture time can be judged, and whether the patient suffers from xerophthalmia is judged.

12 one end of the housing is fitted over the first body 41 of the grip 4. 12 the inner surface of the port of the housing is provided with an internal thread which engages with the outer peripheral surface of the housing of the grip portion 4. When the housing 12 of the illumination module 1 is screwed to the grip 4, an end portion of the imaging objective lens 22 extending to the outside of the grip 4 through the end of the grip 4 is fitted into the housing 12. 12 the housing is uniformly provided with irradiation lamps on the plate surface perpendicular to the axis of the housing. Preferably, the irradiation lamp is provided with a plurality of groups of annular irradiation lamp units from inside to outside along the center of the plate surface. A through hole for mounting the imaging objective lens 22 is provided at the center of the plate surface. Cobalt blue light lamps which are uniformly and annularly arranged are arranged on each annular unit of the board surface, so that the whole board surface can emit light rays with relatively uniform brightness. Preferably, the inner wall of the 12 casing is coated with a black opaque coating, so that the light emitted by the lighting structure inside the 12 casing with the inner cavity can be sufficiently and diffusely reflected in the inner cavity structure of the 12 casing, and the light source can uniformly irradiate towards the set direction. The end of the shell 12 far away from the holding part 4 is also connected with a light modulation module (3) and a Placido plate 5. A light diffusion plate 13 is further disposed between the light modulation module 3 and the Placido plate 5, so that the light of the set wavelength obtained after the light emitted from the lamp unit is converted by the conversion plate 32 of the light modulation module 3 can be adjusted by the light diffusion plate 13 to further improve the diffusion uniformity. The light passing through the light diffusion plate 13 can be uniformly irradiated on the Placido plate 5, so that the defect of uneven light and shade caused by divergence in the light transmission process of the light source is effectively eliminated. During use, the lighting assembly 1 can be controlled and adjusted according to requirements, so that different light rays can be emitted by the lighting assembly. Wherein, the infrared light mode is used for accurately positioning the tear film rupture position and calculating the rupture time thereof, and observing the thickness change of the tear film lipid layer based on the rupture time; the white light mode is used to observe the actual condition of the meibomian glands; the cobalt blue light model was used to examine the corneal/conjunctival lesions.

When the light modulation module is used, different conversion sheets of the light modulation module 3 and different light-transmitting plates of the Placido disc 5 can be adjusted and selected for detection. For example, when a cobalt blue light source switch and an image acquisition camera switch are turned on, and a white light conversion sheet is selected and is not blocked by the Placido plate 5, the lighting assembly emits white light, the white light source is directly aligned to the surface of the tear film of the eyes of the patient, and the optical imaging assembly 2 can conveniently acquire the eye image under the light irradiation. The white light can be convenient for medical personnel to directly observe the actual condition of the meibomian glands around the eyes. When selecting infrared light filter and grid light-passing board, medical personnel carry out accurate location and calculate its break time to tear film break position through the tear film image under the infrared ray irradiation to observe the thickness change of tear film lipid layer based on this. When the injury condition of the keratoconjunctiva of a patient needs to be observed, fluorescein sodium and lissamine green dyeing is firstly carried out on the ocular surface of the patient, then a non-conversion light-transmitting sheet and a non-grid light-transmitting sheet are selected for irradiation, medical staff observe the eye under cobalt blue light, and judge whether the keratoconjunctiva is injured or not by observing the existence of spot dyeing.

Preferably, the eye image that optical imaging subassembly 2 gathered can utilize the singlechip that sets up in the optical imaging subassembly to adopt forms such as WIFI, bluetooth or USB interface to send for the PC that has xerophthalmia detecting element. The machine is examined the device and can be supported multiple transmission, can manage the eye state of an illness of storage patient in real time and gather the result to the PC on, even make things convenient for medical personnel to look over with the modularization storage of patient data information as a result. Preferably, the grip portion 4 is further provided with any one or a combination of any plurality of a battery, an SD card slot, a USB interface, and a wireless transmission signal generator. The battery can be used for directly providing power for the image acquisition camera and the illuminating lamp unit, so that the device is more convenient to use when being held by hands, and the winding of a connecting wire is avoided. The battery can adopt a rechargeable battery used by conventional small-sized electric appliances, and can be directly charged by alternating current when the electric energy storage is low. The SD draw-in groove can be used for installing the SD memory card, can carry out the backup of original image to data when transmission data, avoids when the PC breaks down or data loss, and medical personnel can directly transfer the original detection image data of patient detection time measuring from the SD memory card in the device. Can guarantee through setting up USB interface and wireless transmission signal that dry eye detects handheld imaging device both can be through wireless signal transmission and can also can be through the port transmission to guarantee that handheld imaging device's flexibility ratio is high.

Preferably, after the optical imaging assembly 2 collects the illumination image and/or projection image generated by the illumination light irradiating the cornea, the optical imaging assembly 2 transmits the collected image to the analysis and storage module 6 for analysis and processing, so as to obtain the relevant data with the set parameter information. The analysis storage module 6 can obtain an original data packet according to the collected lesion tissue picture, and can preprocess the collected image data and construct a CNN-LSTM algorithm model on the basis of establishing an information base. The analysis storage module 6 trains and tests the established CNN-LSTM algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition existing during the establishment of the algorithm model, thereby improving the accuracy of the calculation or prediction result of the algorithm model. When receiving the image data transmitted by the optical imaging assembly 2, the newly acquired set of image data and the stored sample information are subjected to comparative analysis through the CNN-LSTM algorithm model. When the image data which is not stored is acquired, newly acquired image data is learned and stored in an information base. The information base stores data as a series of image data acquired by the multiple sets of optical imaging assemblies 2 or input by the data port that tear film breaks over time. Samples can be set as multiple sets of different calibration parameters according to the tear film rupture position and the rupture process duration. The CNN-LSTM algorithm model is built by analyzing and digitizing several different samples. In the process of constructing the CNN-LSTM algorithm model, the CNN neural network system and the LSTM neural network system are modeled in a serial mode. The LSTM neural network system extracts time domain information in a time dimension, and the CNN neural network system extracts space domain information in the same frame. The constructed CNN model comprises an input layer, a convolutional layer, a full-link layer and an input layer. The full connection layer is built at the last part of the CNN model, and the output layer is arranged at the downstream of the full connection layer. The constructed LSTM model comprises a long and a short memory layer and at least one full connection layer. When the images acquired by the light imaging assembly 2 are compared and analyzed, the characteristic information in the acquired images is rapidly identified. The LSTM is a fast extensive complement of the data associated with the feature information in the form of a chain of repeating modules. And learning according to the acquired image information of the specific parameters which are not recorded yet in the identification and analysis process, acquiring feedback data, and correcting according to the accuracy of the data to continuously perfect the model. The CNN-LSTM algorithm model carries out information extraction of the collected image according to a mode of carrying out cross extraction on changed parameter characteristics in a group of image data, so that when one domain of information is extracted, not much information of other domains is lost. The features on each image in a group of tear film images are changed, the changed features are extracted in sequence, and the extracted features are converted from the bottom-layer features to the high-layer semantic features. By constructing a CNN-LSTM algorithm model, the intelligent judgment of an intelligent system on the organization picture of a patient under examination is increased based on the image identification and the learning capability of time sequence progressive data processing of CNN and LSTM networks, and a more complete detection auxiliary function is provided for medical personnel. The CNN-LSTM algorithm model is established, so that learning can be performed on different case pictures and related data mining can be completed while detection and analysis are performed, the information acquisition efficiency is greatly improved, the problem that traditional case information is difficult to acquire is solved, the information acquisition accuracy is high due to the learning mechanism of the model, and the defects of the model can be improved through autonomous learning.

Example 2

A manufacturing method of a dry eye mechanical examination device based on an LSTM network comprises the dry eye detection handheld imaging device and a PC connected with the handheld imaging device. During the use, the image of the patient's eye that handheld image device will gather sends the PC, and corresponding processing analysis is carried out to the PC after accepting the eye projected image to provide the diagnosis suggestion to medical personnel based on corresponding contrastive analysis result, make things convenient for medical personnel to diagnose whether to suffer from the condition such as xerophthalmia according to the patient's eye picture that shows on the PC more accurately.

Be provided with analysis storage module 6 on the PC, analysis storage module 6 can handle and contrastive analysis to the eye image that has obtained, makes things convenient for medical personnel to refer to the diagnosis of carrying out xerophthalmia symptom. The PC machine also comprises a dry eye detection unit which has the measurement function of assisting medical personnel in detecting and judging the tear film rupture time and the tear film rupture site; the detection unit also has the function of measuring the tear film lipid layer by contrasting and observing the thickness change of the tear film lipid layer in a plurality of images. The PC is also internally provided with an imaging module for observing meibomian gland imaging images under the irradiation of white light, a lacrimal river height identification and segmentation module for calculating the lacrimal river height and a display for displaying images acquired by all the handheld imaging devices.

Specifically, first, any one of the conversion plates is inserted into the light tunnel, and the projection of the Placido plate 5 is projected onto the corneal surface of the eye from the center to the periphery so that the entire cornea at the center of the pupil is within the projected image coverage. When the lighting assembly 1 selects the second conversion sheet 322 as the light conversion element, the white light irradiated on the patient's eye can obtain the first eye image, and the arrangement of the meibomian glands can be observed and analyzed according to the first eye image displayed on the PC, and the lacrimal river height can be detected. When the infrared light converted by the first conversion sheet 321 irradiates the eye tear film, a second eye image of the eye tear film of the patient can be acquired, and the PC can accurately locate the tear film rupture position and calculate the rupture time of the tear film rupture position according to the received second eye images, and observe the tear film lipid layer thickness change based on the tear film rupture position and the tear film rupture time. In addition, when the cobalt blue lamp 33 generated by the lamp light source directly irradiates the eye surface dyed with fluorescein sodium and lissamine green, a third eye image of the patient's eyes can be acquired, and whether the stippling exists in the image is observed and analyzed, so that whether the keratoconjunctiva of the patient's eyes has damage or not can be judged. This device is through the integrated design with the power supply system of the light that changes different wavelength and camera with single light source, can adopt the mode of a set of image to discern fast and gather the real-time change characteristic in certain time of patient's tear film, thereby makes things convenient for medical personnel to be the rupture place and the rupture time of tear film according to the accurate analysis and the judgement of a set of tear film images that have the change of contrast. Meanwhile, the tear river height judgment and measurement by medical staff are more standardized, and the subjectivity and the accuracy of the medical staff in disease judgment are improved. In addition, the judgment of the thickness of the lipid layer is more obvious and intuitive, and more help is provided for the medical staff to judge the disease degree. Moreover, the dry eye detection equipment can rapidly obtain a diagnosis result, and the defects of large size and high cost of a desktop device in the prior art are greatly overcome.

The method for detecting the dry eye based on the device integrating the multiple functions of detecting the dry eye disease comprises the following steps:

s1, obtaining an original data packet according to a collected lesion tissue picture;

s2, preprocessing the collected image data and establishing an information base;

s3, constructing a CNN-LSTM algorithm model;

s4, training and testing the established CNN-LSTM algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the condition of specific parameters during the establishment of the algorithm model, thereby improving the accuracy of the calculation or prediction result of the algorithm model;

and S5, when the image data transmitted by the optical imaging component 2 is received, comparing and analyzing a group of newly acquired image data and stored sample information through a CNN-LSTM algorithm model, and when the image data which is not stored is acquired, learning the newly acquired image data and storing the newly acquired image data into an information base.

Wherein S5 at least comprises the following steps:

s51: turning on a power switch of a lamp light source and an image acquisition camera, selecting a second conversion sheet capable of converting a cobalt blue light source into white light, inserting the second conversion sheet into a light channel, irradiating the white light on the surface of a tear film of an eye of a patient by a plain disc to generate a projection with a bright-dark interval, adjusting an imaging switch after positioning is completed, turning on the image acquisition camera, measuring the rupture of the tear film of the examinee and the change of the thickness of a lipid layer by medical personnel, acquiring an image, transmitting and storing the image, and detecting the height of a tear river of the patient by the image transmitted to a PC (personal computer);

s52, secondly, selecting a first conversion sheet capable of converting the cobalt blue light source into infrared light, inserting the first conversion sheet into a light channel, obtaining an image of the eye meibomian gland volume imaging condition of the patient through the light and shade staggered projection generated by the excited infrared light irradiation grid module, transmitting the image to a PC (personal computer) for detection and processing, providing a diagnosis suggestion for medical personnel, and displaying the image on a display;

s53, inserting a second conversion sheet which directly penetrates through the light and has no light conversion function into the light channel, so that the cobalt blue light emitted by the cobalt blue light lamp directly irradiates the eye surface dyed with sodium fluorescein and lissamine green, obtaining an eye image under the cobalt blue light, transmitting the eye image to a PC (personal computer), displaying the eye image on a display, observing and analyzing whether a spot stain exists or not by observing the image under the light source, and further judging whether the keratoconjunctiva has a damage or not;

and S54, finally, turning off the lighting lamp source, completing relevant image data acquisition, and performing comparative analysis on the obtained multiple groups of image photos according to the established CNN-LSTM algorithm model to judge tear film rupture, lipid layer thickness change, meibomian gland imaging condition, lacrimal river height, keratoconjunctival injury condition and the like.

It should be noted that, the three steps S51-S53 are not arranged in the front-back order, and the order of the steps may be changed arbitrarily without affecting the final inspection result. After the tear film rupture, the change of the lipid layer thickness, the meibomian gland imaging condition, the lacrimal river height and the injury condition of the keratoconjunctiva are respectively detected, the dry eye diagnosis result of the patient can be summarized by comprehensively analyzing the detection results.

The technical innovation of the dry eye mechanical examination device based on the LSTM network has many advantages and technical progress for technicians in the same industry nowadays.

Although the present invention has been described with reference to a preferred embodiment, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the invention as defined by the appended claims.

Claims (7)

1. A dry eye mechanical examination device at least comprises an analysis storage module (6) capable of storing and learning collected dry eye patient lesion tissue pictures, an illumination assembly (1) forming a dry eye detection handheld device and an optical imaging assembly (2) with a part of imaging channels arranged in the illumination assembly (1), and is characterized in that under the condition that illumination light emitted by the illumination assembly (1) irradiates the eyes of a patient, the optical imaging assembly (2) collects an illumination image and/or a projection image generated by irradiating the cornea with the illumination light, and the optical imaging assembly (2) transmits the collected image to the analysis storage module (6) for analysis and processing so as to obtain related data with set parameter information;

the analysis storage module (6) can obtain an original data packet according to the collected lesion tissue picture, preprocess the collected image data and construct a hybrid network algorithm model on the basis of establishing an information base; the analysis storage module (6) trains and tests the established hybrid network algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition when the algorithm model is established, thereby improving the accuracy of the calculation or prediction result of the algorithm model;

when image data transmitted by the optical imaging component (2) are received, a group of newly acquired image data and stored sample information are compared and analyzed through a hybrid network algorithm model, and when new image data which are not stored are acquired, the newly acquired image data are learned and stored in an information base;

wherein the content of the first and second substances,

the lighting assembly (1) is internally provided with a light modulation module (3) capable of changing the wavelength of light emitted by a light source, the light emitted by the light source is converted by utilizing conversion plates (32) with different light conversion effects, which are arranged in the light modulation module (3), the two conversion plates (32) are respectively provided with a first conversion sheet (321) and a second conversion sheet (322), wherein,

when the first conversion sheet (321) is inserted into a light channel in a shell (12) of the lighting assembly (1), the first conversion sheet (321) can convert first light emitted by a lighting source (11) of the lighting assembly (1) into second light;

when the second conversion sheet (322) is inserted into the light channel in the shell (12), the second conversion sheet (322) can convert the first light emitted by the illumination light source (11) into third light;

when the first conversion sheet (321) and the second conversion sheet (322) are not inserted into the light channel, the first light emitted by the illumination light source (11) directly irradiates on the ocular surface;

the size of the first conversion sheet (321) and the second conversion sheet (322) is equal to the cross sectional area of an illumination channel in the shell (12), through holes communicated with an imaging channel are respectively formed in the center positions of the faces of the first conversion sheet and the second conversion sheet, and a shielding coating capable of being mutually butted with the pipeline wall of the imaging channel to form a lightless channel is coated on the annular face of each through hole.

2. The dry eye mechanical examination device according to claim 1, wherein the information base stores data for a plurality of sets of image data of tear film rupture with time acquired by the optical imaging assembly (2) or inputted by the data port, and a plurality of sets of samples with different calibration parameters can be set according to different tear film rupture positions and rupture process durations, and a mixed network algorithm model is established by analyzing and performing data processing on a plurality of different samples;

in the process of constructing the hybrid network algorithm model, at least two selected neural network systems are modeled in a mode of being connected in series, wherein the selected neural network systems can at least respectively extract time domain information from a time dimension and extract space domain information from the same frame.

3. The dry eye mechanical examination device according to claim 1 or 2, wherein the constructed spatial domain information extraction model comprises an input layer, a convolution layer, a full connection layer and an input layer, wherein the full connection layer is built at the last part of the spatial domain information extraction model, and the output layer is arranged at the downstream of the full connection layer;

the constructed time domain information extraction model comprises a long and short memory layer and at least one full connection layer;

when the images collected by the optical imaging assembly (2) are compared and analyzed, the characteristic information in the collected images is rapidly identified, and the domain information extraction model rapidly extends and supplements the relevant data of the characteristic information in a repeated module chain mode;

according to the obtained image information of the specific parameters which are not recorded in the process of identification and analysis, learning is carried out, feedback data are obtained, correction is carried out according to the accuracy of the data, and the model is continuously perfected.

4. The dry eye mechanical examination device of claim 3, wherein the hybrid network algorithm model performs the information extraction of the collected images in a manner of performing cross extraction on the changed parameter features in a group of image data, so that when one domain of information is extracted, not much information of other domains is lost, so that the features on each image in a group of tear film images are changed, the changed features are extracted in sequence, and the extracted features are converted from bottom-layer features to high-layer semantic features;

the analysis storage module (6) can select at least two network systems from the existing network systems such as CNN, LSTM or RNN according to different requirements of image data processing executed by the analysis storage module to construct a hybrid network algorithm model capable of extracting specified features.

5. The dry eye mechanical examination device according to claim 1, wherein the optical imaging assembly (2) comprises an image capturing camera (21) for receiving the image of the eye and the tear film surface and an imaging objective lens (22) for receiving the image of the eye and forming an imaging optical path to transmit the projection image to the image capturing camera (21), the imaging channel of the imaging objective lens (22) is embedded in the housing (12) in such a way that the axis of the imaging channel coincides with the axis of the housing (12), the wall of the imaging channel of the imaging objective lens (22) is coated with a shielding coating capable of shielding the light emitted by the illumination light source (11) from directly entering the imaging channel from the outside of the wall of the imaging channel, the imaging channel is formed by a plurality of coaxial pipes, and the connecting position of the pipes is provided with a gap capable of being embedded in the first conversion sheet (321) or the second conversion sheet (322).

6. The dry eye mechanical examination device according to claim 5, wherein a Placido plate (5) for forming concentric annular projections with alternate light and shade is further arranged at one end of the shell (12) far away from the illumination light source (11), the Placido plate (5) is detachably mounted at the opening position of the light passage penetrating the end of the shell (12), so that light irradiated from the light passage can irradiate on the Placido plate (5);

the machine inspection device further comprises a holding part (4), wherein the holding part (4) is formed by a first body (41) and a second body (42) which are perpendicular to each other and form an L shape in an integrated mode, one end, far away from the second body (42), of the first body (41) of the holding part (4) is detachably connected with a port, close to one end of the illumination light source (11), of the illumination assembly (1), an image acquisition camera (21) and an imaging objective lens (22) are arranged in the first body (41) and used for receiving and acquiring images of the surfaces of eyes and tear film, the imaging objective lens (22) is partially connected with the image acquisition camera, penetrates through the end face of the first body (41) and is inserted into the illumination assembly (1), and one end, far away from the image acquisition camera (21), of the imaging objective lens (22) is flush with the outer surface of the Placido disc (5).

7. A use method of a dry eye mechanical examination device at least comprises an analysis storage module (6) capable of storing and learning collected pathological tissue pictures of a dry eye patient, an illumination assembly (1) forming the dry eye mechanical examination handheld device and an optical imaging assembly (2) with part of an imaging channel arranged in the illumination assembly (1), and comprises the following steps:

the adjusted illumination light is used for illuminating the eyes of the patient, the optical imaging assembly (2) collects an illumination image and/or a projection image generated by the illumination light for illuminating the cornea, and the optical imaging assembly (2) transmits the collected image to the analysis storage module (6) for analysis and processing so as to obtain related data with set parameter information;

s1, obtaining an original data packet according to a collected lesion tissue picture;

s2, preprocessing the collected image data and establishing an information base

S3, constructing a hybrid network algorithm model;

s4, training and testing the established hybrid network algorithm model according to the existing data information so as to reduce the data result lacking in the sample data under the specific parameter condition existing in the establishment of the algorithm model, thereby improving the accuracy of the calculation or prediction result of the algorithm model;

s5, when image data transmitted by the optical imaging component (2) are received, a group of newly acquired image data and stored sample information are compared and analyzed through a hybrid network algorithm model, and when image data which are not stored are acquired, the newly acquired image data are learned and stored in an information base;

wherein the content of the first and second substances,

the lighting assembly (1) is internally provided with a light modulation module (3) capable of changing the wavelength of light emitted by a light source, the light emitted by the light source is converted by utilizing conversion plates (32) with different light conversion effects, which are arranged in the light modulation module (3), the two conversion plates (32) are respectively provided with a first conversion sheet (321) and a second conversion sheet (322), wherein,

when the first conversion sheet (321) is inserted into a light channel in a shell (12) of the lighting assembly (1), the first conversion sheet (321) can convert first light emitted by a lighting source (11) of the lighting assembly (1) into second light;

when the second conversion sheet (322) is inserted into the light channel in the shell (12), the second conversion sheet (322) can convert the first light emitted by the illumination light source (11) into third light;

when the first conversion sheet (321) and the second conversion sheet (322) are not inserted into the light channel, the first light emitted by the illumination light source (11) directly irradiates on the ocular surface;

the size of the first conversion sheet (321) and the second conversion sheet (322) is equal to the cross sectional area of an illumination channel in the shell (12), through holes communicated with an imaging channel are respectively formed in the center positions of the faces of the first conversion sheet and the second conversion sheet, and a shielding coating capable of being mutually butted with the pipeline wall of the imaging channel to form a lightless channel is coated on the annular face of each through hole.

Images