Holographic microplastic screening method and system based on deep learning

Abstract

The invention provides a holographic microplastic screening method and system based on deep learning. The method comprises the following steps: acquiring a white light image of a training sample; acquiring a first holographic image of the training sample; performing deep learning training on the white light image and the first holographic image to obtain a holographic image reconstruction model; acquiring a second holographic image of the to-be-detected sample; reconstructing the second holographic image through the holographic image reconstruction model to obtain a high-quality reconstructed image; and analyzing the high-quality reconstructed image to obtain micro-plastic information, and screening micro-plastic particles. According to the holographic micro-plastic screening method based on deep learning, a high-speed holographic technology and high-quality deep learning holographic reconstruction are combined, and the micro-plastic screening system is high in throughput and high in precision at the same time. According to the screening system, the structure and the operation process are simplified, and the screening efficiency is improved on the premise of reducing the cost.

Description

Technical Field

[0001] This invention relates to the technical field of microplastic detection, and in particular to a holographic microplastic screening method and system based on deep learning. Background Technology

[0002] Microplastics, as an emerging environmental pollutant, are increasingly attracting attention due to their potential risks to ecosystems and human health. Efficient and accurate screening and identification of microplastics from environmental samples (such as water samples and sediments) is a current research focus. Among existing technologies, automated screening based on optical microscopy is the mainstream method, which can be mainly divided into the following two categories: a) Microscopic imaging techniques based on fluorescent labeling and high-magnification objectives: Specific Structure and Principle: This technology is one of the current "gold standard" methods for laboratory detection of microplastics. The basic procedure involves first staining the sample with a fluorescent dye (such as Nile Red), which specifically adsorbs onto the polymer and emits fluorescence. Then, the sample is placed on a glass slide and scanned using an upright or inverted fluorescence microscope equipped with a high numerical aperture objective lens, a precision motorized platform, a fluorescence light source, and a cooled CCD camera. The system controls the platform movement to stitch together multiple fields of view, acquiring a wide range of fluorescence and white light transmission images.

[0003] Current status: This method can provide high-contrast, high-resolution fluorescence images and high-quality white light absorption images, which can clearly show the morphology, size and fluorescence characteristics of particles with high accuracy.

[0004] b) Lens-free holographic imaging techniques: Specific Structure and Principle: This technology is an emerging computational imaging method. A typical device includes a monochromatic light source (such as an LED), a sample slide, and an image sensor (such as a CMOS sensor). It eliminates the need for traditional microscope objectives; the sample is directly illuminated by the light source, and its diffraction pattern (i.e., a hologram) is directly recorded by the image sensor below. Subsequently, computer algorithms (such as angular spectral analysis, convolution, etc.) are used to digitally reconstruct the recorded hologram, recovering the amplitude and phase information of the sample, thereby obtaining its morphological image.

[0005] Current Status: Lensless holographic imaging technology has attracted attention due to its large field of view, low cost, and simple and compact structure. It can acquire sample information over a large area of ​​several square centimeters in a single exposure, greatly improving imaging throughput and making it suitable for rapid initial screening of massive environmental samples.

[0006] While both of these technical approaches have their advantages, they both have inherent and irreconcilable drawbacks, which are analyzed in detail below: Regarding the above technique a) (fluorescence microscopy): Main drawbacks: low imaging throughput, slow screening speed, and severe aberrations when imaging over a large area; The problem stems from the fact that its high imaging quality heavily relies on high-magnification, small-field-of-view microscope objectives. To cover a large area of ​​the sample on the slide, hundreds or even thousands of steps and multi-field image stitching must be performed using a precision motorized platform. This process is extremely time-consuming and demands extremely high mechanical stability and control precision from the system. Furthermore, the high cost of core components such as high numerical aperture objectives and fluorescence filter sets makes the entire system prohibitively expensive. Therefore, this technology is essentially a "time-for-quality" approach, unable to meet the urgent need for rapid, high-throughput screening of large numbers of samples. Conversely, when using small numerical aperture lenses, aberrations often prevent the imaging area from being sufficiently large.

[0007] Regarding the above technology b) (lensless holographic imaging technology): Main drawbacks: low image quality (especially resolution and contrast), severe reconstruction artifacts, and lack of chemical composition information.

[0008] Cause of the problem: Image quality depends on the algorithm: Holograms are not intuitive sample images; their quality depends entirely on the accuracy of the digital reconstruction algorithm. The reconstruction process is sensitive to noise and easily affected by twin-image artifacts, speckle noise, etc., resulting in blurred edges, loss of detail, and low signal-to-noise ratio in the reconstructed amplitude (absorption) and phase images, making them difficult to use for accurate morphological analysis and identification of small particles.

[0009] Lack of specific identification capability: Lens-less holographic imaging can only provide physical morphological information of the sample (size, shape, transparency), and cannot distinguish between microplastics and non-plastic particles (such as diatoms, mineral particles, organic debris, etc.), resulting in a high false detection rate. Therefore, this technology is essentially a "quality for throughput" method; although fast, it is not accurate enough to be used as a final confirmation method. Summary of the Invention

[0010] The purpose of this invention is to provide a holographic microplastic screening method and system based on deep learning to solve at least some of the above-mentioned technical problems.

[0011] The first aspect of this invention provides a deep learning-based holographic microplastic screening method, comprising: Obtain white light images of the training samples; Obtain the first holographic image of the training sample; The white light image and the first holographic image are subjected to deep learning training to obtain a holographic image reconstruction model; Acquire the second holographic image of the sample to be tested; The second holographic image is reconstructed using the holographic image reconstruction model to obtain a high-quality reconstructed image; Microplastic information is obtained from the high-quality reconstructed images, and microplastic particles are screened.

[0012] Furthermore, acquiring the white light image of the training sample includes: The control light source module emits white light to illuminate the training sample placed on the first sample displacement stage; A movement command is issued to the first sample displacement stage. During the movement of the first sample displacement stage, multiple white light images with different fields of view are acquired on the second sample displacement stage.

[0013] The white light image is acquired by an imaging chip mounted on the second sample displacement stage. During acquisition, the light source module emits white light, which is collimated by the first lens and then illuminates the training sample on the first sample displacement stage. The image is then focused and imaged by the second lens.

[0014] Furthermore, acquiring the first holographic image of the training sample includes: The control light source module emits coherent light to illuminate the training sample placed on the second sample displacement stage; A movement command is issued to the second sample displacement stage, and multiple first holographic images with different object distances are acquired on the second sample displacement stage during the movement.

[0015] Furthermore, the step of training the white light image and the first holographic image using deep learning to obtain a holographic image reconstruction model includes: The white light image is stitched together using correlation calculation to obtain a large field-of-view, high-quality white light image corresponding to the first holographic image. The first holographic image and the large field-of-view high-quality white light image are registered at the pixel level using feature recognition algorithms and correlation coefficient calculation algorithms. Simultaneously crop the first holographic image and the large field-of-view high-quality white light image to form a holographic-white light training data pair; The training data pairs are fed into a deep learning image generation network for cross-domain learning, and the network learns to obtain a mapping from holographic images to high-quality white light images; Save the weight file of the learned holographic image reconstruction deep learning model to form a holographic image reconstruction model.

[0016] Furthermore, acquiring the second holographic image of the sample to be tested includes: The control light source module emits coherent light to illuminate the sample to be tested placed on the second sample displacement stage; A movement command is issued to the second sample displacement stage, and multiple second holographic images with different object distances are acquired on the second sample displacement stage during the movement of the second sample displacement stage; During the acquisition of the second holographic image, the light source module emits coherent light, which is collimated by the first lens and illuminates the sample to be tested on the second sample displacement stage. The image is then acquired by the imaging chip located on the second sample displacement stage.

[0017] Furthermore, the step of obtaining microplastic information from the high-quality reconstructed image and screening for microplastic particles includes: Adaptive threshold segmentation is performed on the high-quality reconstructed image to obtain a binary image; The binary image was statistically analyzed for microplastic shapes using a connected component analysis algorithm. Microplastic screening information is obtained based on the microplastic statistics.

[0018] A second aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the above-described deep learning-based holographic microplastic screening method.

[0019] A third aspect of the present invention provides a deep learning-based holographic microplastic screening system, comprising an imaging device and the aforementioned electronic device, wherein the imaging device and the electronic device are electrically connected.

[0020] Furthermore, the imaging device includes a light source module and a first lens, a first sample displacement stage, a second lens, and a second sample displacement stage arranged sequentially along the light source emission direction of the light source module, wherein an imaging chip is provided on the second sample displacement stage.

[0021] Furthermore, the first lens, the first sample displacement stage, the second lens, and the second sample displacement stage constitute a white light imaging module, the light source module is used to emit white light to illuminate the sample placed on the first sample displacement stage, and the imaging chip is used to acquire the white light image of the sample; the first lens and the second sample displacement stage constitute a holographic imaging module, the light source module is used to emit coherent light to illuminate the sample placed on the second sample displacement stage, and the imaging chip is used to acquire the holographic image of the sample.

[0022] The beneficial effects of this plan are as follows: This solution, based on a deep learning-based holographic microplastic screening method, addresses the issues of poor holographic image reconstruction quality and low microscopic counting throughput, achieving both high throughput and high precision in microplastic screening. The screening system is not a simple parallel connection of two devices; rather, it features a redesigned structure, simplifying the design and operational procedures, thereby improving screening efficiency while reducing costs. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the screening device structure under white light acquisition mode; Figure 2 This is a top view schematic diagram of the screening device under white light acquisition mode; Figure 3 This is a schematic diagram of the optical path principle of a screening device in white light acquisition mode; Figure 4 This is a schematic diagram of the screening device structure in coaxial holographic acquisition mode; Figure 5 This is a top view of the screening device in coaxial holographic acquisition mode. Figure 6 This is a schematic diagram of the optical path principle of a screening device in coaxial holographic acquisition mode; Figure 7 This is a flowchart illustrating a deep learning-based holographic microplastic screening method. Figure 8 This is a schematic diagram of the first holographic image; Figure 9 A schematic diagram of a large-format white light image; Figure 10 A schematic diagram of a holographic computational image; Figure 11 A schematic diagram of a high-quality reconstructed image; Figure 12 This is a schematic diagram of a large-format, high-quality stitched image. Figure 13 This is a schematic diagram of the deep learning training process; Figure 14 A schematic diagram of the process for screening microplastic particles; Figure 15 This is a schematic diagram of the module structure of an electronic device provided in an embodiment of the present invention.

[0024] Explanation of reference numerals in the attached figures: 10. Light source module; 11. Low-pass aperture; 12. First lens; 13. Second lens; 14. First sample displacement stage; 15. Second sample displacement stage; 20. Imaging chip; 30. Electronic device; 301. Memory; 302. Processor; 303. Network interface. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0026] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "front," "rear," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention; the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance; furthermore, unless otherwise explicitly specified and limited, the terms "installed," "connected," and "joined" should be interpreted broadly, for example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium, or they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the terms in this invention can be understood according to the specific circumstances.

[0027] The current technological landscape presents a paradoxical situation where "quality" and "throughput" are mutually exclusive. Researchers often have to choose between the two based on their needs: either opting for high-quality, low-speed microscopy for precise authentication, or choosing low-quality, high-speed holography for rapid initial screening. While simply paralleling the two devices can yield complementary data, it faces problems such as complex system integration, difficulties in data registration, cumbersome operational procedures, and a significant increase in overall cost, failing to fundamentally resolve the conflict between efficiency and cost.

[0028] See patent publication number CN 116026729A, entitled "A Portable Microplastic Detection Device Based on Digital Coaxial Holographic Microscopy." The method disclosed in this patent is applicable to the detection of microplastics in water; however, this method cannot distinguish microplastics from impurities, resulting in significant false positives. The device contains multiple lenses and imaging components, making it costly.

[0029] This solution discloses a deep learning-based holographic microplastic screening system, including an imaging device and an electronic device, wherein the imaging device and the electronic device are electrically connected.

[0030] Specifically, the imaging device includes a light source module and a first lens, a first sample displacement stage, a second lens, and a second sample displacement stage arranged sequentially along the light source emission direction of the light source module, wherein an imaging chip is provided on the second sample displacement stage.

[0031] See Figure 1-3 In white light acquisition mode, the first lens, the first sample displacement stage, the second lens, and the second sample displacement stage constitute a white light imaging module. The light source module is used to emit white light to illuminate the sample placed on the first sample displacement stage, and the imaging chip is used to acquire white light images of the sample. See Figure 4-6 In the coaxial holographic acquisition mode, the first sample displacement stage and the second lens need to be removed. The first lens and the second sample displacement stage constitute a holographic imaging module. The light source module is used to emit coherent light to illuminate the sample placed on the second sample displacement stage. The imaging chip is used to acquire the holographic image of the sample.

[0032] The light source module consists of a light source and a low-pass aperture. The light source provides coherent light or white light illumination, and the low-pass aperture is used to limit the coherence length of the light source to obtain uniform illumination.

[0033] A schematic diagram of the optical path principle of white light acquisition mode is shown below. Figure 3 As shown.

[0034] Wherein, f1: front focal length of the first lens (common range: 100 mm-500 mm); 2f2: front focal length of the first lens (common range: 30 mm-500 mm); 2f2: Back focal length of the first lens (common range: 30mm-500mm); Low-pass micro-orifice: The orifice size is typically 5-20μm.

[0035] When the sample is placed on the first sample displacement stage before staining, the light source module, the first lens, the first sample displacement stage, the second lens, and the second sample displacement stage enter the white light acquisition mode. The light source module emits a white light beam, moves the first sample displacement stage, and the imaging chip acquires multiple white light images with different fields of view. A schematic diagram of the optical path principle of the coaxial holographic acquisition mode is shown below. Figure 6 As shown.

[0036] Wherein, f1: Back focal length of the first lens (common range: 5 mm-30 mm); Low-pass aperture: Aperture size is usually 5-20 μm; Δd: Usually 10-100 μm, number of object distances is usually 3-10. d: Object distance range, usually between 500 μm and 3 mm.

[0037] Before staining, when the sample is placed on the holographic imaging module, the light source module and the holographic imaging module form a coaxial holographic acquisition mode. The light source module emits a coherent light beam, and the imaging chip acquires a first holographic image. After staining, when the sample is placed on the holographic imaging module, the light source module and the holographic imaging module form a coaxial holographic acquisition mode. The light source module emits a coherent light beam and moves the second sample displacement stage a distance Δd in a direction perpendicular to the imaging chip. The imaging chip acquires multiple second holographic images with different object distances (typical object distance is 50 micrometers to 5 millimeters).

[0038] The imaging chip in this solution uses an image processor to process images, enabling a deep learning-based holographic microplastic screening method. The image processor includes a deep learning module and an image analysis module. The deep learning module comprises preprocessing, network architecture, and image output sections. The preprocessing section is responsible for normalizing, denoising, and matching images from different modalities, converting the images into a data format that can be learned or input into the network architecture. The network architecture is an end-to-end deep learning image generation network, such as the common Unet or Generative Adversarial Network, responsible for outputting high-quality intensity contrast images from multiple input holographic images. The image output section stitches and enhances the visualization of the network output images. The image analysis module includes image statistics and multivariate analysis modules. Image statistics are responsible for thresholding and statistical analysis of microplastics in the high-quality reconstructed images, while the multivariate analysis module performs connected component analysis on the microplastics to obtain variable information such as microplastic morphology and size.

[0039] See Figure 7 This solution further discloses a holographic microplastic screening method based on deep learning, characterized by comprising: S1. Obtain the white light image of the training sample; A sample slide containing microplastic particles is placed on the first sample displacement stage. A white light beam (LED light source or optical fiber) is controlled as a point light source through a small hole and then magnified and collimated by the first lens into a uniform illumination spot to illuminate the sample slide. The illuminated sample information is recorded by the imaging chip. At the same time, the first sample displacement stage is moved to obtain multiple white light absorption images under different fields of view.

[0040] The acquisition of white light images of training samples includes: S11. Control the light source module to emit white light to illuminate the training sample placed on the first sample displacement stage; S12. A movement command is issued to the first sample displacement stage. During the movement of the first sample displacement stage, multiple white light images with different fields of view are acquired on the second sample displacement stage to obtain multi-field white light images W1, W2, ..., Wn.

[0041] The white light image is acquired by an imaging chip mounted on the second sample displacement stage. During acquisition, the light source module emits white light, which is collimated by the first lens and then illuminates the training sample on the first sample displacement stage. The image is then focused and imaged by the second lens.

[0042] S2. Obtain the first holographic image of the training sample; The second lens is removed, at which point the multimodal imaging module switches to coaxial holographic acquisition mode. The glass slide sample carrying microplastic particles is then placed at the second sample displacement stage; the first sample displacement stage does not participate in imaging. The light source is switched to a coherent illumination beam (LED or fiber optic), which, after passing through the pinhole and the first lens, becomes a uniform coherent illumination spot illuminating the sample glass slide. The second sample displacement stage is moved, causing the sample to move perpendicular to the chip, obtaining holographic images at different object distances (typically 50 micrometers to 5 millimeters).

[0043] S21. Control the light source module to emit coherent light to illuminate the training sample placed on the second sample displacement stage; S22. A movement command is issued to the second sample displacement stage. During the movement of the second sample displacement stage, multiple first holographic images with different object distances are acquired on the second sample displacement stage. (See below) Figure 8 We obtain multi-object distance holographic images Q1, Q2, ..., Qn.

[0044] During the acquisition of the first holographic image, the light source module emits coherent light, which is collimated by the first lens and illuminates the training sample on the second sample displacement stage, and is then acquired by the imaging chip located on the second sample displacement stage.

[0045] S3. Perform deep learning training on the white light image and the first holographic image to obtain a holographic image reconstruction model; the deep learning training step is mainly performed on a computer, server, or other microcomputer platform containing a GPU processor, see [link to documentation]. Figure 13 , .

[0046] S31. Use correlation calculation to stitch the white light image to obtain a large field-of-view high-quality white light image corresponding to the first holographic image; Specifically, image stitching algorithms (such as correlation calculation methods) are used to stitch multi-view white light images W1, W2, ..., Wn into a large-format white light image W. See [link to relevant documentation]. Figure 9 The white light image is the same size as Q1-Qn.

[0047] S32. Perform pixel-level registration of the first holographic image and the large field-of-view high-quality white light image using feature recognition algorithm and correlation coefficient calculation algorithm; Specifically, an image registration algorithm (such as an image feature matching algorithm) is used to align the white light image W and the holographic images Q1-Qn at the pixel level. S33, The first holographic image and the large field-of-view high-quality white light image are simultaneously cropped to form a holographic-white light training data pair; Specifically, the white light image W and the holographic images Q1-Qn are simultaneously cropped into small-format data pairs (q1-qn:w, where q1-qn corresponds to the multi-object distance image in the above figure, and W corresponds to the white light label image in the above figure). The multi-object distance image q1-qn is used to obtain the holographic computation image B through a traditional holographic reconstruction algorithm (such as the GS algorithm), see [link to documentation]. Figure 10 .

[0048] S34. The training data pairs are fed into a deep learning image generation network for cross-domain learning, and the network learns to obtain a mapping from holographic images to high-quality white light images; Specifically, multi-object distance images are fed into a deep learning reconstruction network for training (commonly used networks include Generative Adversarial Networks and UNet). The high-quality reconstructed images output are shown below. Figure 11 Similarity is calculated between the white-light labeled image W and the holographic computation image B, and the resulting loss value is used for network training. After training, all high-quality reconstructed images can be stitched together into a large-format, high-quality stitched image. (See [link to image]). Figure 12 .

[0049] S35. Save the weight file of the learned holographic image reconstruction deep learning model to form a holographic image reconstruction model.

[0050] S4. Acquire the second holographic image of the sample to be tested; place the sample to be tested, which contains microplastic particles, on the slide of the second sample displacement stage to acquire the holographic image.

[0051] S41. Control the light source module to emit coherent light to illuminate the sample to be tested placed on the second sample displacement stage; S42. Issue a movement command to the second sample displacement stage. During the movement of the second sample displacement stage, acquire multiple second holographic images with different object distances on the second sample displacement stage to obtain multi-object distance holographic images q1-qn. During the acquisition of the second holographic image, the light source module emits coherent light, which is collimated by the first lens and illuminates the sample to be tested on the second sample displacement stage. The image is then acquired by the imaging chip located on the second sample displacement stage.

[0052] S5. Reconstruct the second holographic image using the holographic image reconstruction model to obtain a high-quality reconstructed image; put the multi-object distance second holographic image into the holographic image reconstruction model for high-quality image reconstruction to obtain a high-quality reconstructed image C.

[0053] S6. Obtain microplastic information from the high-quality reconstructed image analysis, and screen for microplastic particles. (See [link]) Figure 14 .

[0054] S61. Perform adaptive threshold segmentation on the high-quality reconstructed image to obtain a binary image; Specifically, adaptive thresholding (local adaptive thresholding algorithm) is performed on the high-quality reconstructed image C to obtain the binary image BW.

[0055] S62. Perform microplastic shape statistics on the binary image using a connected component analysis algorithm; Specifically, a connected component analysis algorithm is used to obtain connected component information in the binary image BW, and a contour line angle algorithm is used to distinguish between contacting / overlapping microplastics. The connected components of the distinguished microplastics are then numbered and their information acquired, including perimeter, area, major axis, and minor axis. Roundness information is further calculated. The major axis information is used to perform shape statistics on the microplastics (e.g., major axis greater than 10 micrometers, major axis less than 10 micrometers), and the shape of the microplastics is classified based on roundness and the major axis-minor axis ratio.

[0056] S63. Obtain microplastic screening information based on the microplastic statistics.

[0057] This solution, based on deep learning, combines high-speed holographic technology with high-quality deep learning holographic reconstruction, resulting in a microplastic screening system that simultaneously achieves high throughput and high accuracy. The system simplifies structure and operation, improving screening efficiency while reducing costs.

[0058] The present invention also discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the deep learning-based holographic microplastic screening method as described in any of the above-described methods.

[0059] Please refer to the details. Figure 15 This is a basic structural block diagram of the electronic device in this embodiment.

[0060] The electronic device 30 includes a memory 301, a processor 302, and a network interface 303 that are interconnected via a system bus. It should be noted that only the electronic device 30 with components 301-303 is shown in the figure; however, it should be understood that it is not required to implement all the shown components, and more or fewer components can be implemented alternatively. Those skilled in the art will understand that the electronic device described here is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0061] The electronic device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The electronic device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0062] The memory 301 includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 301 may be an internal storage unit of the electronic device 30, such as the hard disk or memory of the electronic device 30. In other embodiments, the memory 301 may also be an external storage device of the electronic device 30, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc. Of course, the memory 301 may include both internal storage units and external storage devices of the electronic device 30. In this embodiment, the memory 301 is typically used to store the operating system and various application software installed on the electronic device 30, such as program code for distributed information sharing federated learning methods for heterogeneous data statistics. In addition, the memory 301 can also be used to temporarily store various types of data that have been output or will be output.

[0063] In some embodiments, the processor 302 may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 302 is typically used to control the overall operation of the electronic device 30. In this embodiment, the processor 302 is used to run program code stored in the memory 301 or to process data, for example, to run the program code described above for the distributed information sharing federated learning method for heterogeneous data statistics.

[0064] The network interface 303 may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the electronic device 30 and other electronic devices.

[0065] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the distributed information sharing federated learning method for heterogeneous data statistics as described in any of the above descriptions.

[0066] The aforementioned computer-readable storage medium stores a federated learning method for sharing distributed information on statistical heterogeneity. Through this method, each client shares its local data distribution so that other clients can expand the data to reduce the heterogeneity of the local data. The expanded dataset is then trained into a target task model using federated learning to solve the statistical heterogeneity problem. This is a method to solve the statistical heterogeneity problem from the data level, while ensuring the security of patient data.

[0067] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0068] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A holographic microplastic screening method based on deep learning, characterized in that, include: Obtain white light images of the training samples; Obtain the first holographic image of the training sample; The white light image and the first holographic image are subjected to deep learning training to obtain a holographic image reconstruction model; Acquire the second holographic image of the sample to be tested; The second holographic image is reconstructed using the holographic image reconstruction model to obtain a high-quality reconstructed image; Microplastic information is obtained from the high-quality reconstructed images, and microplastic particles are screened.

2. The deep learning-based holographic microplastic screening method according to claim 1, characterized in that, The acquisition of white light images of training samples includes: The control light source module emits white light to illuminate the training sample placed on the first sample displacement stage; A movement command is issued to the first sample displacement stage. During the movement of the first sample displacement stage, multiple white light images with different fields of view are acquired on the second sample displacement stage.

3. The deep learning-based holographic microplastic screening method according to claim 1, characterized in that, The acquisition of the first holographic image of the training sample includes: The control light source module emits coherent light to illuminate the training sample placed on the second sample displacement stage; A movement command is issued to the second sample displacement stage, and multiple first holographic images with different object distances are acquired on the second sample displacement stage during the movement.

4. The deep learning-based holographic microplastic screening method according to claim 1, characterized in that, The step of training a holographic image reconstruction model by deep learning between the white light image and the first holographic image includes: The white light image is stitched together using correlation calculation to obtain a large field-of-view, high-quality white light image corresponding to the first holographic image. The first holographic image and the large field-of-view high-quality white light image are registered at the pixel level using feature recognition algorithms and correlation coefficient calculation algorithms. Simultaneously crop the first holographic image and the large field-of-view high-quality white light image to form a holographic-white light training data pair; The training data is fed into a deep learning image generation network for cross-domain learning, and the network learns to obtain a mapping from holographic images to high-quality white light images; Save the weight file of the learned holographic image reconstruction deep learning model to form a holographic image reconstruction model.

5. The deep learning-based holographic microplastic screening method according to claim 1, characterized in that, The acquisition of the second holographic image of the sample to be tested includes: The control light source module emits coherent light to illuminate the sample to be tested placed on the second sample displacement stage; A movement command is issued to the second sample displacement stage, and multiple second holographic images with different object distances are acquired on the second sample displacement stage during the movement.

6. The deep learning-based holographic microplastic screening method according to claim 1, characterized in that, The process of obtaining microplastic information from the high-quality reconstructed image and screening for microplastic particles includes: Adaptive threshold segmentation is performed on the high-quality reconstructed image to obtain a binary image; The binary image was statistically analyzed for microplastic shapes using a connected component analysis algorithm. Microplastic screening information is obtained based on the microplastic statistics.

7. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the deep learning-based holographic microplastic screening method as described in any one of claims 1-6.

8. A holographic microplastic screening system based on deep learning, characterized in that, It includes an imaging device and an electronic device as described in claim 7, wherein the imaging device is electrically connected to the electronic device.

9. The deep learning-based holographic microplastic screening system according to claim 8, characterized in that, The imaging device includes a light source module and a first lens, a first sample displacement stage, a second lens, and a second sample displacement stage arranged sequentially along the light source emission direction of the light source module. An imaging chip is provided on the second sample displacement stage.

10. The deep learning-based holographic microplastic screening system according to claim 9, characterized in that, The first lens, the first sample displacement stage, the second lens, and the second sample displacement stage constitute a white light imaging module. The light source module is used to emit white light to illuminate the sample placed on the first sample displacement stage, and the imaging chip is used to acquire the white light image of the sample. The first lens and the second sample displacement stage constitute a holographic imaging module. The light source module is used to emit coherent light to illuminate the sample placed on the second sample displacement stage, and the imaging chip is used to acquire the holographic image of the sample.

Images