Blood cell intelligent detection device and blood cell image splicing method

Abstract

The invention relates to the technical field of blood cell detection, and provides an intelligent blood cell detection device and a blood cell image splicing method.The intelligent blood cell detection device comprises a device shell, a base, a micro-motion positioning assembly, an imaging support, an observation assembly, a placement table and a two-way positioning assembly, the base is fixedly installed in the device shell through the damping foot pad, the micro-motion position adjusting assembly is installed on the base, the imaging support is installed on the micro-motion position adjusting assembly, the observation assembly is installed on the imaging support, the placing table is arranged on the base, and the bidirectional position adjusting assembly is installed on the base. According to the technical scheme, the problems that in the prior art, when a specimen is subtly observed, the observation position is difficult to accurately adjust in a small distance, adjacent images are difficult to obtain, and meanwhile the images are difficult to accurately splice are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of blood cell detection. Specifically, it relates to an intelligent blood cell detection device and a method for splicing blood cell images. Background Art

[0002] In the current medical industry, the detection methods for blood and parasites are as follows: the collected blood specimens are attached to an optical slide, and an intelligent camera is used to image the specimens. Operators judge the blood conditions based on the imaged morphology with their own professional knowledge and experience. This is a traditional detection method that completely relies on the professional knowledge and experience of the operators and is difficult to accurately judge the state of the imaging results.

[0003] After retrieval, the patent document with the publication number CN202111345140.9 proposes a detection and analysis method for blood or parasites, which uses an adjustment structure in three directions to adjust the observation position of the intelligent camera to achieve automatic focusing on the imaging of blood cells or parasite eggs without manual intervention in focusing; and it can perform intelligent scanning of multiple frames of images to form a sufficiently large imaging area for blood or parasites.

[0004] During the use of an intelligent blood camera, in order to obtain complete blood cell information, it is usually necessary to take multiple images of precisely adjacent local fields of view with the intelligent camera and splice them to construct a blood cell image with a larger range and higher resolution. Since the intelligent camera is very sensitive to small changes in the sample position, in the operation of the intelligent camera, adjusting the focal length is a key step to obtain a clear image. When making a small-distance adjustment to the observation position, any minor malfunction or uncalibration of the device may cause a deviation in the observation position, making it difficult to accurately adjust to obtain the images required for splicing.

[0005] At the same time, there are many challenges in the splicing process of blood cell images taken by an intelligent camera. During the imaging process of the intelligent camera, due to the physical limitations of the optical system, the depth of field of each blood cell image is limited, resulting in cells in the edge area of the image being often blurrier than those in the central area; the objective lens of the intelligent camera focuses well at the center position of the field of view, while at the edge of the field of view, due to the influence of spherical aberration and field curvature, the focus of the edge cells may shift, resulting in a decrease in clarity; in addition, limited by the light source structure and changes in the sample thickness, the illumination in the edge area of the image is usually weak, resulting in a decrease in contrast and an increase in noise, further exacerbating the edge blur problem. Since the clarity of the edge areas of each image decreases, discontinuous transition areas are likely to occur during splicing, affecting the overall image quality. Summary of the Invention

[0006] The present invention provides an intelligent blood cell detection device and a method for splicing blood cell images, which solves the problem in the prior art that it is difficult to accurately adjust the observation position at a small distance when observing a specimen in detail, affecting the accuracy of observation and increasing the inconvenience of detection.

[0007] The technical solution of the present invention is as follows:

[0008] An intelligent blood cell detection device, including a device housing, an observation window is provided on the device housing, and further includes:

[0009] A base, the base is fixedly installed inside the device housing through shock-absorbing foot pads;

[0010] A micro-position adjustment component, the micro-position adjustment component is installed on the base;

[0011] An imaging bracket, the imaging bracket is installed on the micro-position adjustment component, and the micro-position adjustment component is used to adjust the position of the imaging bracket;

[0012] An observation component, the observation component is installed on the imaging bracket and is used to observe the specimen;

[0013] A placement table, the placement table is arranged on the base;

[0014] A two-way position adjustment component, the two-way position adjustment component is installed on the base and is used to adjust the position of the placement table.

[0015] On the basis of the foregoing solution, the micro-position adjustment component includes:

[0016] A carrier frame, the carrier frame is fixedly installed on the base, and the carrier frame is slidably connected to the imaging bracket;

[0017] A fixed frame, the fixed frame is fixedly installed inside the carrier frame;

[0018] A sliding frame, the sliding frame is slidably installed inside the fixed frame, and the sliding frame is fixedly connected to the bottom of the imaging bracket;

[0019] A differential part, the differential part is installed between the fixed frame and the sliding frame and is used to finely adjust the relative position between the sliding frame and the fixed frame;

[0020] A driving part, the driving part is installed inside the carrier frame and is used to drive the differential part to act.

[0021] On the basis of the foregoing solution, the differential part includes:

[0022] A first nut, the first nut is fixedly installed on the fixed frame;

[0023] The first screw rod, which is threadedly arranged inside the first nut;

[0024] The second nut, which is fixedly installed on the sliding frame;

[0025] The second screw rod, which is threadedly arranged inside the second nut and fixedly connected to the first screw rod;

[0026] Wherein, the second screw rod and the first screw rod have different pitches and opposite helix directions.

[0027] On the basis of the foregoing solution, the driving part includes:

[0028] The sliding shaft, which is fixedly installed on the first screw rod;

[0029] The transmission shaft, which is coaxially and slidably installed on the sliding shaft;

[0030] The first motor, which is fixedly installed on the bearing frame, and the output end of the first motor is fixedly connected to the transmission shaft.

[0031] On the basis of the foregoing solution, the observation assembly includes:

[0032] The mounting frame, which is fixedly installed on the imaging support;

[0033] The sliding frame, which is slidably installed on the mounting frame;

[0034] The camera, which is fixedly installed on the upper part of the sliding frame;

[0035] The lens, which is fixedly installed on the sliding frame and installed on the camera;

[0036] The control part, which is installed on the mounting frame and used to control the position of the sliding frame;

[0037] The observation part, which is installed on the placement table and used to place the specimen to be observed.

[0038] On the basis of the foregoing solution, the control part includes:

[0039] The second motor, which is fixedly installed on the mounting frame;

[0040] The first lead screw, which is fixedly installed on the output end of the second motor;

[0041] The first transmission nut, which is threadedly sleeved on the first lead screw and fixedly connected to the sliding frame.

[0042] Based on the foregoing solution, the observation part includes:

[0043] A placement groove, which is provided on the placement table;

[0044] A light-transmitting hole, which is provided on the placement groove;

[0045] A light source, which is fixedly installed on the base.

[0046] Based on the foregoing solution, the bidirectional position adjustment assembly includes:

[0047] A lateral position adjustment frame, which is fixedly installed on the base;

[0048] A first connecting frame, which is slidably installed on the lateral position adjustment frame;

[0049] A longitudinal position adjustment part, which is installed on the first connecting frame and is used to adjust the position of the placement table;

[0050] A power part, which is installed on the lateral position adjustment frame and is used to adjust the position of the placement table.

[0051] Based on the foregoing solution, the longitudinal position adjustment part includes:

[0052] A through groove, which is provided on the imaging support frame

[0053] A longitudinal position adjustment frame, which is fixedly installed on the first connecting frame;

[0054] A second connecting frame, which is slidably installed on the longitudinal position adjustment frame, and the second connecting frame is fixedly connected to the placement table.

[0055] Based on the foregoing solution, the power part includes:

[0056] A second lead screw, which is rotatably installed on the lateral position adjustment frame;

[0057] A second transmission nut, which is threadedly sleeved on the second lead screw, and the second transmission nut is fixedly connected to the first connecting frame;

[0058] A third motor, which is fixedly installed on the lateral position adjustment frame, and the output end of the third motor is fixedly connected to the second lead screw;

[0059] A third lead screw, which is rotatably installed on the longitudinal position adjustment frame;

[0060] The third driving nut is threadedly sleeved on the third lead screw, and the third driving nut is fixedly connected to the second connecting frame;

[0061] The fourth motor is fixedly installed on the longitudinal position adjusting frame, and the output end of the fourth motor is fixedly connected to the third lead screw.

[0062] On the basis of the foregoing device, the present invention further proposes a method for splicing blood cell images, including the following steps:

[0063] Step S1, collect the blood cell images captured by the intelligent camera and preprocess them; input the preprocessed blood cell images into the first generation network, and use the adversarial learning mechanism to enhance the image quality to generate the first enhanced image

[0064] Step S2, adopt the cell adaptive enhancement network to jointly analyze the central cells and edge cells in the same image, and use the structural information of the central cells to optimize the clarity of the edge cells;

[0065] Step S3, adopt the second generation network to extract features from the optimized image to obtain the second enhanced image;

[0066] Step S4, adopt the improved segmentation network to segment the cell structure of the second enhanced image to obtain the blood cell feature map;

[0067] Step S5, based on the blood cell feature map, adopt the image stitching method based on feature matching to align multiple blood cell images

[0068] The working principle and beneficial effects of the present invention are as follows:

[0069] 1. In the present invention, when making a small position adjustment to the imaging bracket through the sliding bracket, it is necessary to drive the first lead screw and the second lead screw to rotate simultaneously. The first lead screw and the second lead screw have opposite helix directions. At this time, the pitch difference between the first lead screw and the second lead screw will cause a precise small movement of the sliding bracket, thereby driving the imaging bracket to generate displacement, improving the accuracy during the small-distance adjustment of the observation position when observing the specimen, and facilitating the acquisition of images at different positions on the specimen.

[0070] 2. In the present invention, the third motor drives the second lead screw to rotate, thereby driving the first connecting frame to move through the second driving nut, and then driving the longitudinal position adjusting frame to move along the axis direction of the second lead screw to adjust the horizontal position of the specimen. The fourth motor drives the third lead screw to rotate, thereby driving the second connecting frame to move through the third driving nut, and then driving the placement table to move along the axis direction of the third lead screw to adjust the longitudinal position of the specimen, realizing the precise movement of the specimen in the X and Y directions, and facilitating the multi-angle and multi-position observation of blood cells.

[0071] 3. In the present invention, the position of the imaging bracket is adjusted through the fine adjustment component to achieve high-precision positioning of the observation component, ensuring the accuracy and clarity of observation. Through the setting of the bidirectional adjustment component, the position of the specimen is adjusted. In combination with the setting of the shock-absorbing foot pads, the interference of external vibration on the detection process is effectively reduced, and the stability and reliability of detection are improved.

[0072] 4. The cell adaptive enhancement network is adopted to optimize the edge cells by using the clear central cell information in the same image, avoiding enhancement artifacts caused by excessive differences in cell morphology between different images and improving the recognizability of edge cells. Through local contrast adaptive enhancement, the visibility of cells in low-light regions is enhanced, ensuring the overall clarity of the spliced image.

[0073] 5. An improved segmentation network is adopted, which combines multi-scale image fusion, topology structure affinity graph optimization and spatial constraint morphology adaptation, can accurately segment adjacent cells, reduce cell misconnection or over-segmentation, and improve the consistency of the blood cell region. By learning the topological relationship of adjacent blood cells through the graph neural network, accurate segmentation can still be ensured in the densely distributed blood cell region, improving the segmentation accuracy and providing high-quality input data for splicing. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] The present invention will be further described in detail below in conjunction with the drawings and specific embodiments.

[0075] Figure 1 It is a schematic structural diagram of the whole in the present invention;

[0076] Figure 2 It is a schematic cross-sectional structural diagram of the cooperation of the fine movement component, the observation component and the bidirectional adjustment component in the present invention;

[0077] Figure 3 It is a schematic structural diagram of the cooperation of the fine movement component, the observation component and the bidirectional adjustment component in the present invention;

[0078] Figure 4 It is a schematic cross-sectional structural diagram of the fine adjustment component in the present invention;

[0079] Figure 5 It is a schematic cross-sectional structural diagram of the cooperation of the differential part and the driving part in the present invention;

[0080] Figure 6 It is a schematic structural diagram of the observation component in the present invention;

[0081] Figure 7 It is a schematic cross-sectional structural diagram of the control part in the present invention;

[0082] Figure 8 It is a schematic cross-sectional structural diagram of the bidirectional adjustment component in the present invention;

[0083] Figure 9 This is a schematic diagram of blood cell splicing according to the present invention.

[0084] In the figure: 1, device housing; 2, observation window; 3, base; 4, shock-absorbing foot pad; 5, imaging bracket; 6, placement table; 7, carrier; 8, fixing bracket; 9, sliding bracket; 10, first nut; 11, first screw; 12, second nut; 13, second screw; 14, sliding shaft; 15, transmission shaft; 16, first motor; 17, mounting bracket; 18, sliding frame; 19, camera; 20, lens; 21, second motor; 22, first lead screw; 23, first transmission nut; 24, placement groove; 25, light-transmitting hole; 26, light source; 27, lateral position adjustment bracket; 28, first connecting bracket; 29, through groove; 30, longitudinal position adjustment bracket; 31, second connecting bracket; 32, second lead screw; 33, second transmission nut; 34, third motor; 35, third lead screw; 36, third transmission nut; 37, fourth motor. Detailed implementation manners

[0085] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0086] As Figures 1 to 8 shown, this embodiment provides an intelligent blood cell detection device, including a device housing 1, an observation window 2 is opened on the device housing 1, and further includes a base 3, a fine position adjustment component, an imaging bracket 5, an observation component, a placement table 6 and a two-way position adjustment component. The base 3 is fixedly installed inside the device housing 1 through shock-absorbing foot pads 4. The fine position adjustment component is installed on the base 3, and the imaging bracket 5 is installed on the fine position adjustment component. The fine position adjustment component is used to adjust the position of the imaging bracket 5. The fine position adjustment component includes a carrier 7, a fixing bracket 8, a sliding bracket 9, a differential part and a driving part. The carrier 7 is fixedly installed on the base 3. The carrier 7 is slidably connected to the imaging bracket 5. The fixing bracket 8 is fixedly installed inside the carrier 7. A sliding bracket 9 is slidably installed inside the fixing bracket 8. The sliding bracket 9 is fixedly connected to the bottom of the imaging bracket 5. A differential part is installed between the fixing bracket 8 and the sliding bracket 9 for finely adjusting the relative position between the sliding bracket 9 and the fixing bracket 8. The driving part is installed inside the carrier 7 for driving the differential part to act.

[0087] Specifically, when it is necessary to observe a blood specimen, first place the specimen on the observation component, then start the two-way positioning component to move the specimen to the observation position, and then adjust the relative position between the observation component and the specimen until the observation component moves to the optimal observation position. Then, observe the specimen and collect relevant data such as images. During this process, it is necessary to adjust the micro-position of the specimen to obtain clearer and more accurate observation results. At this time, start the driving part, and the driving part drives the differential part to act, thereby driving the sliding frame 9 to move slightly within the fixed frame 8, so as to drive the imaging bracket 5 to move a small distance through the sliding frame 9, so as to achieve fine positioning of blood cells at different positions in the specimen, ensure the imaging quality, and improve the observation range.

[0088] As described above, such as Figure 4 , Figure 5 shown, the differential part includes a first nut 10, a first screw 11, a second nut 12 and a second screw 13. The first nut 10 is fixedly installed on the fixed frame 8, the first screw 11 is threadedly arranged inside the first nut 10, the second nut 12 is fixedly installed on the sliding frame 9, and the second screw 13 is threadedly arranged inside the second nut 12. The second screw 13 is fixedly connected to the first screw 11. Among them, the pitch of the second screw 13 is different from that of the first screw 11 and the rotation directions are opposite.

[0089] Specifically, when the imaging bracket 5 is slightly adjusted in position by the sliding frame 9, it is necessary to drive the first screw 11 and the second screw 13 to rotate simultaneously. During rotation, under the cooperation of the first nut 10 and the first screw 11 and the cooperation of the second nut 12 and the second screw 13, the pitch difference between the first screw 11 and the second screw 13 will cause precise and slight movement of the sliding frame 9, thereby pushing the imaging bracket 5 to generate displacement, improving the accuracy during the process of adjusting the observation position when observing the specimen.

[0090] As described above, such as Figure 4 , Figure 5 shown, the driving part includes a sliding shaft 14, a transmission shaft 15 and a first motor 16. A sliding shaft 14 is fixedly installed on the first screw 11, the transmission shaft 15 is coaxially and slidably installed on the sliding shaft 14, and the first motor 16 is fixedly installed on the carrier 7. The output end of the first motor 16 is fixedly connected to the transmission shaft 15;

[0091] Specifically, when it is necessary to drive the first screw 11 and the second screw 13 to rotate, start the first motor 16. The first motor 16 drives the transmission shaft 15 to rotate, and the transmission shaft 15 drives the sliding shaft 14 to rotate. Since the first screw 11 and the second screw 13 will have axial displacement during rotation, during the rotation process, the sliding shaft 14 will also move axially within the transmission shaft 15 without affecting normal rotation, ensuring the stable and efficient operation of the entire system.

[0092] As Figure 6 , Figure 7 shown, the observation assembly is installed on the imaging bracket 5 for observing the specimen. The observation assembly includes a mounting frame 17, a sliding frame 18, a camera 19, a lens 20, a control unit and an observation unit. The mounting frame 17 is fixedly installed on the imaging bracket 5. The sliding frame 18 is slidably installed on the mounting frame 17. The camera 19 is fixedly installed on the upper part of the sliding frame 18. The lens 20 is fixedly installed on the sliding frame 18. The lens 20 is installed on the camera 19. The control unit is installed on the mounting frame 17 for controlling the position of the sliding frame 18. The observation unit is installed on the placement table 6 for placing the specimen to be observed.

[0093] Specifically, when observing the specimen, it is necessary to drive the observation unit to move below the lens 20, and then through the setting of the control unit, drive the sliding frame 18 to move on the mounting frame 17. At the same time, the sliding frame 18 will drive the camera 19 and the lens 20 to move synchronously, and collect the images of the specimen at the same time. That is, the automatic difference method is used to compare the previous and the next images until an image with the best focusing effect is determined, so as to adjust the observation position, ensure the best focusing position between the lens 20 and the specimen, so as to ensure that the imaging quality is not affected, so as to obtain clear image information, and then turn off the control unit and start observing the specimen.

[0094] As described above, as Figure 7 shown, the control unit includes a second motor 21, a first lead screw 22 and a first transmission nut 23. The second motor 21 is fixedly installed on the mounting frame 17. The first lead screw 22 is fixedly installed at the output end of the second motor 21. The first transmission nut 23 is threadedly sleeved on the first lead screw 22. The first transmission nut 23 is fixedly connected to the sliding frame 18.

[0095] Specifically, when adjusting the position of the sliding frame 18, start the second motor 21. The second motor 21 drives the first lead screw 22 to rotate, thereby driving the first transmission nut 23 to move, thereby driving the sliding frame 18 to linearly move on the mounting frame 17, realizing the adjustment of the position of the lens 20. After the adjustment is completed, turn off the second motor 21 to fix the relative position between the sliding frame 18 and the mounting frame 17.

[0096] As described above, as Figure 2 , Figure 8 shown, the observation unit includes a placement groove 24, a light transmission hole 25 and a light source 26. The placement table 6 is provided with a placement groove 24. The placement groove 24 is provided with a light transmission hole 25. The light source 26 is fixedly installed on the base 3.

[0097] Specifically, when observing a specimen, first place the specimen in the placement groove 24 and fix its relative position. Then, after the specimen is moved to the observation position, at this time, the light transmission hole 25 is located above the light source 26. When observing the specimen, start the light source 26. The light source 26 emits light, which irradiates the specimen through the light transmission hole 25, thereby improving the visibility of its details and facilitating the observation of blood cells in the specimen.

[0098] As Figure 8 shown, the placement table 6 is arranged on the base 3, and the two-way position adjustment component is installed on the base 3 for adjusting the position of the placement table 6. The two-way position adjustment component includes a transverse position adjustment frame 27, a first connecting frame 28, a longitudinal position adjustment part, and a power part. The transverse position adjustment frame 27 is fixedly installed on the base 3, and the first connecting frame 28 is slidably installed on the transverse position adjustment frame 27. The longitudinal position adjustment part is installed on the first connecting frame 28 for adjusting the position of the placement table 6, and the power part is installed on the transverse position adjustment frame 27 for adjusting the position of the placement table 6.

[0099] Specifically, when the specimen is placed well, it is necessary to quickly move the specimen to the observation position. At this time, start the power part. The power part pushes the first connecting frame 28 on the transverse position adjustment frame 27 to move, thereby driving the longitudinal position adjustment part to move and adjusting the transverse position of the placement table 6. Then start the longitudinal position adjustment part to adjust the longitudinal position of the placement table 6 until the positions of the light transmission hole 25 and the light source 26 correspond. At this time, the specimen is in the observation position. Then turn off the power part to lock the position, and then the observation of the specimen can be started.

[0100] As described above, as Figure 8 shown, the longitudinal position adjustment part includes a through groove 29, a longitudinal position adjustment frame 30, and a second connecting frame 31. A through groove 29 is formed on the imaging support 5. The longitudinal position adjustment frame 30 is fixedly installed on the first connecting frame 28, and the second connecting frame 31 is slidably installed on the longitudinal position adjustment frame 30. The second connecting frame 31 is fixedly connected to the placement table 6.

[0101] Specifically, when adjusting the longitudinal position of the placement table 6, push the second connecting frame 31 on the longitudinal position adjustment frame 30, and then the placement table 6 can be pushed to move longitudinally, thereby adjusting the longitudinal position of the placement table 6.

[0102] As described above, as Figure 8As shown in the figure, the power unit includes a second lead screw 32, a second transmission nut 33, a third motor 34, a third lead screw 35, a third transmission nut 36, and a fourth motor 37. The second lead screw 32 is rotatably installed on the lateral adjustment frame 27. The second transmission nut 33 is threadedly sleeved on the second lead screw 32. The second transmission nut 33 is fixedly connected to the first connection frame 28. The third motor 34 is fixedly installed on the lateral adjustment frame 27. The output end of the third motor 34 is fixedly connected to the second lead screw 32. The third lead screw 35 is rotatably installed on the longitudinal adjustment frame 30. The third transmission nut 36 is threadedly sleeved on the third lead screw 35. The third transmission nut 36 is fixedly connected to the second connection frame 31. The fourth motor 37 is fixedly installed on the longitudinal adjustment frame 30. The output end of the fourth motor 37 is fixedly connected to the third lead screw 35.

[0103] Specifically, when adjusting the lateral position of the specimen on the placement table 6, the third motor 34 is started. The third motor 34 drives the second lead screw 32 to rotate, thereby driving the second transmission nut 33 to move. The second transmission nut 33 drives the first connection frame 28 to move, thereby driving the longitudinal adjustment frame 30 to move along the axis direction of the second lead screw 32 to adjust the lateral position of the specimen. When adjusting the longitudinal position of the specimen, the fourth motor 37 is started. The fourth motor 37 drives the third lead screw 35 to rotate, thereby driving the third transmission nut 36 to move. The third transmission nut 36 drives the second connection frame 31 to move, thereby driving the placement table 6 to move along the axis direction of the third lead screw 35 to adjust the longitudinal position of the specimen.

[0104] The working principle or usage process of this application is as follows:

[0105] When it is necessary to observe a blood specimen, first place the specimen in the placement groove 24 and fix the relative position of the specimen. Then move the specimen to the observation position. When adjusting the lateral position of the specimen on the placement table 6, the third motor 34 is started. The third motor 34 drives the second lead screw 32 to rotate, thereby driving the second transmission nut 33 to move. The second transmission nut 33 drives the first connection frame 28 to move, thereby driving the longitudinal adjustment frame 30 to move along the axis direction of the second lead screw 32 to adjust the lateral position of the specimen. When adjusting the longitudinal position of the specimen, the fourth motor 37 is started. The fourth motor 37 drives the third lead screw 35 to rotate, thereby driving the third transmission nut 36 to move. The third transmission nut 36 drives the second connection frame 31 to move, thereby driving the placement table 6 to move along the axis direction of the third lead screw 35 to adjust the longitudinal position of the specimen until the positions of the light-transmitting hole 25 and the light source 26 correspond. At this time, the specimen is in the observation position. Subsequently, the power unit is turned off to lock the position, and then the observation of the specimen can be started.

[0106] When observing the specimen, it is necessary to adjust the position of the lens 20. At this time, the second motor 21 is started, and the second motor 21 drives the first lead screw 22 to rotate, thereby driving the first transmission nut 23 to move, and then driving the sliding frame 18 to move on the mounting frame 17. At this time, the sliding frame 18 will drive the camera 19 and the lens 20 to move synchronously. At this time, the light source 26 is started, and the light source 26 emits light and irradiates the specimen through the light transmission hole 25 until the observation position is adjusted to ensure the best focus position between the lens 20 and the specimen, so as to ensure that the imaging quality is not affected, and clear image information is obtained. Then the second motor 21 is turned off, and the relative position between the sliding frame 18 and the mounting frame 17 is fixed, and then the blood cells in the specimen can be observed and relevant data such as images can be collected.

[0107] In order to obtain clearer and more accurate observation results, it is also necessary to adjust the micro position of the specimen during the observation process. At this time, the first motor 16 is started, and the first motor 16 drives the transmission shaft 15 to rotate. The transmission shaft 15 drives the sliding shaft 14 to rotate. Since the first screw 11 and the second screw 13 will have an axial displacement when rotating, during the rotation process, the sliding shaft 14 will also move axially within the transmission shaft 15 without affecting the normal rotation, ensuring the stable and efficient operation of the entire system. Under the cooperation of the first nut 10 and the first screw 11 and the cooperation of the second nut 12 and the second screw 13, the pitch difference between the first screw 11 and the second screw 13 will cause a precise micro movement of the sliding frame 9, thereby driving the imaging support 5 to generate a displacement, improving the accuracy during the process of adjusting the observation position when observing the specimen, so as to realize the fine positioning of blood cells at different positions in the specimen, ensure the imaging quality, and thus increase the observation range.

[0108] As Figure 9 shown, by adjusting the horizontal (X) and vertical (Y) movements of the specimen on the placement table 6, multiple cell images can be collected. The collected images will be sequentially encoded and named by the algorithm. The algorithm will name them into a puzzle layout of XxY. The boundary of each frame of the collected image coincides with the corresponding image overlapping area, and the algorithm will perform puzzle recognition according to the image overlapping area. Taking red blood cells with a size of 5-7um as an example, the size of a single frame of the captured image is about 0.2*0.4mm, and the X / Y walking distance will be set within a shooting range of about 0.18*0.38mm, so that there is an overlap of about 0.01mm at the boundary of each captured image. However, as described in the background art, due to the decrease in clarity of the edge regions of each image, discontinuous transition regions are likely to occur during splicing, affecting the overall image quality.

[0109] To solve the problems in the splicing process, based on the device in the foregoing example, the present invention further proposes a method for splicing blood cell images, which includes the following steps:

[0110] Step S1, collect the blood cell images captured by the intelligent camera and preprocess them; input the preprocessed blood cell images into the first generation network, and use the adversarial learning mechanism to enhance the image quality and generate the first enhanced image.

[0111] Step S2, adopt the cell adaptive enhancement network to jointly analyze the central cells and edge cells in the same image, and use the structural information of the central cells to optimize the clarity of the edge cells.

[0112] Step S3, adopt the second generation network to extract features from the optimized image and obtain the second enhanced image.

[0113] Step S4, adopt the improved segmentation network to segment the cell structure of the second enhanced image and obtain the blood cell feature map.

[0114] Step S5, based on the blood cell feature map, adopt the image stitching method based on feature matching to align multiple blood cell images.

[0115] The steps in the above method are further explained below.

[0116] Step S1: Collect the blood cell images captured by the intelligent camera and preprocess them; input the preprocessed blood cell images into the first generation network, and use the adversarial learning mechanism to enhance the image quality and generate the first enhanced image.

[0117] First, through an optical intelligent camera, a phase contrast intelligent camera, a fluorescence intelligent camera or other imaging devices capable of obtaining blood cell images, where the intelligent camera consists of a high-resolution camera + a high-power lens, image collection is performed on the blood sample to obtain microscopic blood cell images under different fields of view.

[0118] For the collected blood cell images, due to possible problems such as imaging blur, uneven illumination, lens distortion, and color deviation during the imaging process of the intelligent camera, the present invention preprocesses the original blood cell images, including but not limited to the following operations: noise suppression processing, brightness and contrast equalization, color deviation correction, and lens distortion correction.

[0119] Input the preprocessed blood cell images into the first generation network. The first generation network adopts an adversarial learning mechanism, including a generator and a discriminator, to enhance the image quality and generate the first enhanced image.

[0120] Among them, the generator optimizes the boundaries, textures, and internal structures of blood cells through multi-layer feature extraction modules such as convolutional neural networks (CNNs) and attention mechanisms, and generates enhanced cell images; the discriminator compares the generated images with the original images and optimizes the capabilities of the generator through loss functions to ensure that the enhanced images can not only remove noise and artifacts but also preserve the true morphology and structural information of the cells.

[0121] In an alternative implementation method, the first generation network specifically includes:

[0122] An image quality improvement module for removing problems such as imaging blur, uneven illumination, and lens distortion.

[0123] This module is used to improve the overall clarity of blood cell images, enhance contrast, and reduce image quality degradation caused by factors such as uneven illumination, imaging noise, and lens distortion. The specific implementation may include:

[0124] A multi-scale denoising network that combines spatial and frequency domain features to perform adaptive denoising on Gaussian noise, Poisson noise, or electronic sensor noise in blood cell images.

[0125] Illumination equalization correction, through methods such as Gamma correction and adaptive brightness adjustment, to improve the visibility of blood cells in different regions, making the cells in low-illumination regions and high-illumination regions at the same brightness level and improving the stability of subsequent analysis.

[0126] Texture enhancement processing, using a convolutional neural network (CNN) to extract cell boundary and internal structure features, and through the feature learning ability of the deep learning network, making details such as the texture and staining particles inside the cells clearer.

[0127] A cell morphology optimization module for repairing cell boundaries and improving segmentation accuracy.

[0128] This module is used to optimize the boundary features of blood cells, improve the integrity of cell contours, and reduce edge blur problems caused by optical imaging limitations. The specific implementation may include:

[0129] Morphology optimization based on edge detection, using a deep learning network based on edge detection, such as an enhanced U-Net or a Transformer-based edge detection model, to extract cell contour information and optimize the blurred areas using morphology repair methods.

[0130] Cell reconstruction based on morphological adaptation. For cell regions with blurred edges, a method based on same-image enhancement is used, that is, using the clear cell information in the central region of the image to perform adaptive morphological reconstruction on the cells in the edge region, thereby complementing the information missing due to optical imaging limitations and improving the clarity of edge cells.

[0131] Edge enhancement based on the attention mechanism, through an adaptive attention module, enhances the features of the edge region of blood cells, making the edge information more obvious and improving the accuracy of subsequent segmentation and stitching.

[0132] After the above image quality improvement and cell morphology optimization processes, a first enhanced image is obtained. This enhanced image has a more uniform light distribution, clearer cell contours, less noise interference, and high detail retention ability in the edge region.

[0133] The first enhanced image provides high-quality input data for subsequent blood cell feature extraction and image stitching, ensuring the stability of the stitching process and the integrity of the stitching result.

[0134] Step S2: Use a cell adaptive enhancement network to jointly analyze the central cells and edge cells in the same image, and optimize the clarity of the edge cells using the structural information of the central cells.

[0135] Since the blood cell images captured by intelligent cameras usually have the problem of blurred edges, directly using traditional global enhancement methods may cause loss of cell details. Especially when the thickness of the blood sample varies greatly, there are significant differences in the imaging quality of edge cells and central cells in different pictures, thus affecting the accuracy and consistency of image stitching.

[0136] Therefore, by jointly analyzing the central cells and edge cells in the same image, and optimizing the clarity of the edge region cells using the structural information of the central region cells, the overall quality of the microscopic image is improved, providing high-quality input data for subsequent cell segmentation and image stitching.

[0137] In an optional implementation method, the implementation of the cell adaptive enhancement network may include:

[0138] A cell feature matching module, which is used to select cells with similar features such as morphology, color, and light in the same image, and compensate and enhance the edge cells with the central region cells as a reference.

[0139] Since the blood cells in the same image usually have relatively consistent lighting conditions and sample thickness, the present invention does not use the information of other images when enhancing the edge cells, but compensates and enhances the edge cells based on the central region cells in the same image.

[0140] The cell feature matching module is used to select cells with similar features such as morphology, color, and light in the same image, and enhance the cells in the edge region with the cells in the central region as a reference. Its specific processing may include:

[0141] Cell morphology matching, calculate the geometric features of the cell contour (including area, perimeter, shape factor, etc.), and select the central cell with the closest morphology to the edge cell through a morphological similarity measurement method (such as Hausdorff distance).

[0142] Color feature matching, using the color histogram statistical method, match the staining conditions of the central cell and the edge cell to ensure that no color distortion is introduced during the enhancement process.

[0143] Lighting balance matching, calculate the local lighting distribution of the cell area, and select the central cell with the closest lighting conditions to reduce the impact of lighting deviation on the enhancement of edge cells.

[0144] Through the above feature matching process, ensure that the selected central cell can be used as an enhancement reference for the edge cell, thereby improving the clarity and recognizability of the edge cell.

[0145] Cell edge detail migration module, used to enhance the resolution and detail clarity of edge cells based on the texture features of high-quality central cells.

[0146] After obtaining the matched central cell, based on the texture features of the high-quality central cell, perform detail migration and enhancement on the edge cell to improve the resolution and detail clarity of the edge cell. Specifically, it can include:

[0147] Texture feature extraction, use a convolutional neural network to extract features from key structural regions such as the cell membrane, cytoplasm, and cell nucleus of the central cell, and adopt a multi-scale feature fusion method to effectively extract detail information at different scales.

[0148] Feature migration based on deep learning, use a feature pyramid network or an attention mechanism to construct a cell detail mapping relationship to ensure that no texture distortion or information loss occurs during the feature migration process from the central cell to the edge cell.

[0149] Edge reconstruction based on gradient optimization, calculate the difference in boundary gradients between the edge cell and the central cell, and adopt a gradient compensation method based on the Laplace operator to enhance the contour details of the edge cell to make it have a clearer edge structure.

[0150] Through the above steps, the morphological features of the edge cell are more consistent with the central cell, reducing the edge blur problem and improving the overall resolution of the microscopic image.

[0151] Adaptive contrast adjustment module, adaptively adjust the brightness and contrast of the edge cell according to the lighting difference between the edge cell and the central cell to reduce image non-uniformity.

[0152] Due to the possible uneven distribution of the imaging light source of the intelligent camera, the brightness of cells in the edge area is relatively low or the contrast is weak, which affects the accuracy of subsequent cell segmentation and stitching. Therefore, in view of the lighting difference between edge cells and central cells, the brightness and contrast of edge cells are adaptively adjusted to reduce image non-uniformity. The specific implementation may include:

[0153] Local brightness normalization, calculating the brightness distribution curve in the same image, and removing local extreme brightness points based on the Gaussian filtering method to make the lighting change tend to be smooth.

[0154] Contrast optimization based on histogram matching, calculating the brightness histogram of central cells, and using the histogram matching method to adjust the brightness of edge cells to a level similar to that of central cells to ensure overall uniform lighting.

[0155] Brightness enhancement based on Gamma correction, for different types of blood cells, using the Gamma correction method to non-linearly adjust the brightness of edge cells to ensure their visual clarity, while avoiding artifacts caused by over-enhancement.

[0156] Through the above methods, the lighting balance of blood cells is enhanced, the brightness consistency of the entire image is improved, and obvious lighting transition traces will not appear after stitching.

[0157] After the above cell feature matching, detail migration, and adaptive contrast adjustment processing, the clarity of edge cells is improved, and their morphological features, texture details, and brightness contrast are highly consistent with those of central cells.

[0158] The optimized image provides high-quality input for subsequent blood cell segmentation, feature extraction, and image stitching, improves the matchability of edge cells during stitching, reduces the stitching error rate, and enhances the overall visual effect of the final stitched image and the accuracy of medical analysis.

[0159] Step S3: Use the second generation network to extract features from the optimized image to obtain a second enhanced image.

[0160] Due to the physical limitations of intelligent camera imaging, in blood cell images, the boundary information of some cells is weak. Especially in the case of uneven lighting, focal length offset, or sample thickness change, it may cause cell edges to be blurred and the contrast to be low, affecting subsequent cell segmentation and feature extraction.

[0161] Therefore, a deep learning network is used to extract features from the optimized blood cell image, and further enhance the cell boundary, local contrast, and visual clarity of key regions, thereby obtaining a second enhanced image.

[0162] In an alternative implementation method, the second generation network consists of a multi-scale feature extraction module, an attention mechanism module, and a local contrast enhancement module, which jointly enhance the features of blood cell images and improve the usability of the imaging data of the intelligent camera.

[0163] The multi-scale feature extraction module is used to extract cell boundary information at different levels simultaneously.

[0164] Since blood cell images contain cell structures of different sizes, and the cell boundaries may exhibit different levels of clarity due to lighting or imaging noise, therefore, extracting cell boundary information at different levels simultaneously to ensure the detail integrity of the overall image.

[0165] The core functions of the multi-scale feature extraction module are as follows:

[0166] The multi-scale convolutional neural network uses convolutional layers with different receptive fields to extract large, medium, and small scale features from blood cell images to capture cell boundary information at different levels.

[0167] Based on the feature pyramid for feature fusion, using top-down and bottom-up fusion strategies to ensure that different scale information of cell boundaries can be fully utilized and fine structure information can be retained in high-resolution images.

[0168] Local adaptive filtering, for cell boundaries in different regions, uses adaptive filtering technology to remove high-frequency noise, while enhancing boundary clarity and improving the visibility of cell contours.

[0169] This module can integrate cell boundary features at different scales to ensure that cells of different sizes and under different lighting conditions can be accurately detected, improving the accuracy of subsequent cell segmentation and feature matching.

[0170] The attention mechanism module is used to enhance the edge clarity of the target cell region.

[0171] Since there are problems such as complex background and local blurring in blood cell images, extracting features only through convolution operations may cause the mixing of target cell and background information, affecting the clarity of cell boundaries. Therefore, adaptively focus on the target cell region while suppressing background noise to improve the clarity of the target region.

[0172] The core functions of the attention mechanism module are as follows:

[0173] Spatial attention, based on the pixel-level feature map, calculates the difference between the cell region and the background region, and adaptively assigns attention weights to make the model pay more attention to the cell boundary region.

[0174] Channel attention, through global feature statistics, weights the information of different channels, enhances the focus on key structures such as cell membrane, cell nucleus, cytoplasm, and improves the contrast of cell areas.

[0175] Multi-layer self-attention, using the Transformer structure or self-attention mechanism, enables the model to establish long-distance dependencies on feature maps of different scales and ensure the stability of boundary information.

[0176] This module can significantly improve the visibility of the target cell area, making the boundaries of blood cells clearer, while effectively suppressing background noise and improving the overall image quality.

[0177] The local contrast enhancement module is used to improve the regional contrast of smart camera images.

[0178] Since the illumination distribution of blood cell images may be uneven, the edge areas of some cells may have insufficient contrast due to weak illumination, which affects subsequent image analysis. Therefore, the contrast of the cell area is adaptively adjusted to improve the visual clarity of the smart camera image.

[0179] The core functions of the local contrast enhancement module are as follows:

[0180] Based on regional histogram equalization, an adaptive local histogram equalization method is used to address the illumination differences in different regions to ensure that the brightness of all cell regions is balanced and avoid artifacts caused by over-enhancement.

[0181] Based on the contrast adjustment of gamma correction, nonlinear gamma transformation is used to enhance the contrast of darker cell areas, making the edge details of the cells clearer.

[0182] High dynamic range enhancement uses an HDR mapping algorithm to enhance the visibility of blood cells without losing details in overexposed or underexposed areas that may appear in smart camera imaging.

[0183] This module can enhance the contrast of local areas without affecting the overall image balance, improve the visibility of blood cell images taken by smart cameras, make cell boundaries more prominent, and improve stitching quality.

[0184] After the above multi-scale feature extraction, attention mechanism optimization and local contrast enhancement processing, a second enhanced image is obtained. Compared with the original image, the enhanced image has the following advantages:

[0185] The cell boundaries are clearer and the blurred areas are reduced; the contrast of the target cell area is higher, which enhances the visual recognizability; the separation between the cell area and the background area is improved, providing a more stable input for subsequent segmentation and splicing.

[0186] The second enhanced image can be used for subsequent splicing of blood cell images, automatic cell classification, and medical diagnostic analysis, ensuring the high quality of blood intelligent camera image data during the high-precision splicing process.

[0187] Step S4: Use an improved segmentation network to segment the cell structure of the second enhanced image to obtain a blood cell feature map.

[0188] During the imaging process of the intelligent camera, the morphological characteristics of blood cells may change due to factors such as lighting, staining, and focal length offset, resulting in blurred boundaries of some cells, cell overlap, or incomplete morphology, affecting the accurate identification of blood cells and the subsequent splicing effect.

[0189] Therefore, through multi-scale feature extraction, topological structure analysis, and spatial constraint optimization, the blood cells in the second enhanced image are accurately segmented to generate a blood cell feature map, so as to improve the segmentation accuracy of blood cells and ensure the alignment consistency of each cell region during the subsequent splicing process.

[0190] The improved segmentation network consists of a multi-scale image fusion module, an affinity graph module based on topological structure, and a morphological adaptive module based on spatial constraints, jointly realizing the accurate recognition and optimization of blood cell regions.

[0191] Among them:

[0192] The multi-scale image fusion module is used to simultaneously utilize local and global features to improve the recognition ability of cell regions.

[0193] Blood cell images usually contain both local microscopic features (such as cell membrane boundaries and nuclear structures) and global macroscopic features (such as cell arrangement relationships and cell distribution patterns). Single-scale feature extraction methods are difficult to take into account cell information at different levels.

[0194] The multi-scale image fusion module adopts the method of multi-resolution feature extraction and cross-scale information fusion to simultaneously extract and enhance the boundary features of cell regions at the cell level and tissue level, improving the recognition ability of the segmentation network for blood cells.

[0195] In an optional implementation manner, the functions of the multi-scale image fusion module are as follows:

[0196] The multi-scale convolutional neural network uses receptive fields of different sizes to hierarchically extract high-frequency edge features and low-frequency morphological information of blood cell images to ensure the integrity of cell structures at each scale.

[0197] Feature pyramid fusion uses a strategy that combines bottom-up feature extraction and top-down detail compensation to fuse cell boundary information at multiple scales, enabling the segmentation network to accurately identify the contours of individual cells while maintaining the spatial consistency between cell regions.

[0198] Morphological compensation based on variational autoencoder uses a variational autoencoder to perform feature compensation on low-resolution regions, reducing the problem of missing cell boundaries caused by focal length changes or image blur and improving the integrity of segmentation.

[0199] This module can effectively combine local and global information to ensure that the segmentation of blood cell regions has both high-resolution details and the coherence of the overall morphology, providing a more stable distribution of cell regions for subsequent stitching.

[0200] The affinity graph module based on topological structure is used to identify the relationships between adjacent blood cells and improve the segmentation accuracy.

[0201] Blood cells in microscopic images usually exist in a state of being closely arranged, partially overlapping or in contact. Simple segmentation methods based on pixel-level features may misconnect or split adjacent cells, affecting the accuracy of segmentation. Therefore, further identifying the relationships between adjacent blood cells, optimizing the segmentation boundaries, and improving the segmentation accuracy.

[0202] The core functions of the affinity graph module based on topological structure are as follows:

[0203] Construct a cell affinity graph. Using cell contour features, boundary gradient information, and cell density distribution, construct a topological relationship graph to describe the connection relationships between adjacent cells.

[0204] Optimize cell segmentation based on graph convolutional network. Take the constructed affinity graph as input, and use graph neural network to calculate the association degree between cells to ensure that adjacent cells can be correctly separated while avoiding the mis-splitting of individual cells.

[0205] Based on maximum affinity learning, adopt a maximum affinity learning loss function to minimize the cell segmentation error rate, enabling the segmentation network to effectively adapt to blood cells of different morphologies and improve the segmentation stability of complex cell structures.

[0206] This module can accurately identify the topological relationships between adjacent cells, ensure that blood cells are not misconnected or mis-split, and improve the integrity and segmentation accuracy of blood cell regions.

[0207] The morphological adaptive module based on spatial constraints is used to perform morphological optimization when the boundaries of blood cells are blurred.

[0208] Since the boundaries of blood cells may become unclear due to problems such as blurred imaging, cell deformation, or uneven staining, traditional segmentation methods are prone to segmentation errors when dealing with cells with incomplete boundaries. Therefore, during the segmentation process, the morphological boundaries of cells are dynamically adjusted to improve the accuracy of segmentation.

[0209] The core functions of the morphology adaptive module based on spatial constraints are as follows:

[0210] Boundary optimization based on morphological filtering uses morphological dilation + morphological erosion combined with gradient information to perform morphological adaptive optimization on cells with blurred boundaries, making the cell boundaries clearer.

[0211] Morphology correction based on self-supervised learning uses a self-supervised deep learning model to automatically learn the morphological priors of normal blood cells through pre-trained data, and correct abnormal morphologies (such as blurred morphology and light staining), making the segmentation results more consistent with the true structure of blood cells.

[0212] Spatial constraint loss function. During the model training process, a spatial constraint loss function is introduced to ensure that the cell boundaries are consistent with their local context, preventing the occurrence of cell boundary breaks or overly smooth segmentation.

[0213] This module can dynamically adjust the boundary morphology of blood cells, improve the adaptability of segmentation to blurred regions, ensure that the final blood cell feature map has clear contours, and improve the accuracy of subsequent stitching.

[0214] After multi-scale feature extraction, topological structure optimization, and morphological adaptive correction, the final blood cell feature map is obtained. This feature map can be used for subsequent blood cell stitching, automatic classification, and disease analysis, ensuring a stable cell region distribution during the high-precision stitching of blood intelligent camera image data and improving the stitching quality.

[0215] Step S5: Based on the blood cell feature map, use an image stitching method based on feature matching to align multiple blood cell images.

[0216] Among them, the stitching method includes:

[0217] The image registration module based on key point detection is used to identify similar features between cells to improve the stitching accuracy.

[0218] Since there may be focal length deviations, cell distribution changes, and morphological deformations between different images, traditional direct alignment methods are difficult to guarantee the stitching accuracy. Therefore, similar features between cells are identified to ensure the accurate alignment of images in different fields of view.

[0219] The core functions of the image registration module based on key point detection are as follows:

[0220] Feature point extraction based on Scale-Invariant Feature Transform (SIFT) is used to detect key points in the cell region by SIFT, ensuring that the feature points of the same cell can be accurately identified even under illumination, rotation, or scale changes.

[0221] Real-time feature matching based on Speeded-Up Robust Features (SURF). On the basis of SIFT detection, the ORB algorithm is used for accelerated matching to improve the computational efficiency and ensure that the alignment operation can be completed in real time during the stitching process.

[0222] Geometric correction based on affine transformation. For the problem of cell morphological deformation caused by optical distortion or sample non-uniformity, the affine transformation model is used for correction to ensure that the cell morphology of adjacent images remains consistent and improve the stability of stitching.

[0223] An image fusion module based on deep learning is used to eliminate the stitching boundary and improve the continuity of the overall image.

[0224] After image registration, if stitching is directly performed, due to factors such as cell staining differences, uneven illumination, and contrast changes, there may be obvious color differences and discontinuities in the stitching boundary region, thus affecting the visual effect of the overall image and the stability of medical analysis. Further:

[0225] Seamless transition learning based on Generative Adversarial Network (GAN). An adversarial learning mechanism is adopted to enable the image content in the stitching region to be automatically adjusted to match the adjacent region, thereby eliminating the obvious transition region at the stitching boundary.

[0226] Texture matching based on self-attention mechanism. A deep learning model is used to extract the texture features of the cell region and perform adaptive adjustment in the stitching region, so that the cell structure in the boundary region will not mutate due to stitching.

[0227] Progressive stitching based on multi-resolution fusion. A multi-resolution method is adopted. First, large-scale matching is performed at the low-resolution level, and then the boundary features of the cells are refined at the high-resolution level, thus realizing progressive seamless fusion.

[0228] This module can effectively reduce the mutation at the stitching boundary, improve the overall consistency of the image, and ensure that the stitched blood cell image has high usability in both visual and medical analysis.

[0229] The finally generated high-resolution stitched blood cell image after image registration, deep learning fusion, and gradient optimization transition processing.

[0230] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent blood cell detection device, comprising a device housing (1), and an observation window (2) is provided on the device housing (1), characterized in that, Further included are: A base (3), which is fixedly installed inside the device housing (1) through shock-absorbing foot pads (4); A fine adjustment component, which is installed on the base (3); An imaging support (5), which is installed on the fine adjustment component, and the fine adjustment component is used to adjust the position of the imaging support (5); An observation component, which is installed on the imaging support (5) and is used to observe specimens; A placement table (6), which is arranged on the base (3); A two-way adjustment component, which is installed on the base (3) and is used to adjust the position of the placement table (6).

2. The intelligent blood cell detection device according to claim 1, characterized in that, The fine adjustment component includes: A carrier (7), which is fixedly installed on the base (3), and the carrier (7) is slidably connected to the imaging support (5); A fixed frame (8), which is fixedly installed inside the carrier (7); A sliding frame (9), the sliding frame (9) is slidably installed inside the fixed frame (8), and the sliding frame (9) is fixedly connected to the bottom of the imaging support (5); A differential part, which is installed between the fixed frame (8) and the sliding frame (9) and is used to finely adjust the relative position between the sliding frame (9) and the fixed frame (8); A driving part, which is installed inside the carrier (7) and is used to drive the differential part to act.

3. The intelligent blood cell detection device according to claim 2, characterized in that, The differential part includes: A first nut (10), which is fixedly installed on the fixed frame (8); A first screw (11), which is threadedly arranged inside the first nut (10); A second nut (12), which is fixedly installed on the sliding frame (9); A second screw (13), which is threadedly arranged inside the second nut (12), and the second screw (13) is fixedly connected to the first screw (11); Wherein, the pitch of the second screw (13) is different from that of the first screw (11) and their helix directions are opposite.

4. The intelligent blood cell detection device according to claim 3, wherein, The driving part includes: A sliding shaft (14), which is fixedly installed on the first screw (11); A transmission shaft (15), which is coaxially and slidably installed on the sliding shaft (14); A first motor (16), which is fixedly installed on the carrier (7), and the output end of the first motor (16) is fixedly connected to the transmission shaft (15).

5. An intelligent blood cell detection device according to claim 4, characterized in that, The observation component includes: A mounting frame (17), which is fixedly installed on the imaging support (5); A sliding frame (18), which is slidably installed on the mounting frame (17); A camera (19), which is fixedly installed on the upper part of the sliding frame (18); A lens (20), which is fixedly installed on the sliding frame (18), and the lens (20) is installed on the camera (19); A control unit, which is installed on the mounting bracket (17) and is used to control the position of the sliding bracket (18); An observation unit, which is installed on the placement table (6) and is used to place the specimen to be observed.

6. The intelligent blood cell detection device according to claim 5, wherein, The control unit includes: A second motor (21), which is fixedly installed on the mounting bracket (17); A first lead screw (22), which is fixedly installed at the output end of the second motor (21); A first transmission nut (23), which is threadedly sleeved on the first lead screw (22), and the first transmission nut (23) is fixedly connected to the sliding bracket (18).

7. An intelligent blood cell detection device according to claim 6, characterized in that, The observation unit includes: A placement groove (24), which is formed on the placement table (6); A light-transmitting hole (25), which is formed on the placement groove (24); A light source (26), which is fixedly installed on the base (3).

8. An intelligent blood cell detection device according to claim 7, characterized in that, The two-way position adjustment assembly includes: A lateral position adjustment bracket (27), which is fixedly installed on the base (3); A first connecting bracket (28), which is slidably installed on the lateral position adjustment bracket (27); A longitudinal position adjustment part, which is installed on the first connecting bracket (28) and is used to adjust the position of the placement table (6); A power part, which is installed on the lateral position adjustment bracket (27) and is used to adjust the position of the placement table (6).

9. The intelligent blood cell detection device according to claim 8, characterized in that, The longitudinal position adjustment part includes: A through groove (29), which is formed on the imaging bracket (5) A longitudinal position adjustment bracket (30), which is fixedly installed on the first connecting bracket (28); A second connecting bracket (31), which is slidably installed on the longitudinal position adjustment bracket (30), and the second connecting bracket (31) is fixedly connected to the placement table (6).

10. An intelligent blood cell detection device according to claim 9, characterized in that, The power part includes: A second lead screw (32), which is rotatably installed on the lateral position adjustment bracket (27); A second transmission nut (33), which is threadedly sleeved on the second lead screw (32), and the second transmission nut (33) is fixedly connected to the first connecting bracket (28); A third motor (34), which is fixedly installed on the lateral position adjustment bracket (27), and the output end of the third motor (34) is fixedly connected to the second lead screw (32); A third lead screw (35), which is rotatably installed on the longitudinal position adjustment bracket (30); A third transmission nut (36), which is threadedly sleeved on the third lead screw (35), and the third transmission nut (36) is fixedly connected to the second connecting bracket (31); A fourth motor (37), which is fixedly installed on the longitudinal position adjustment bracket (30), and the output end of the fourth motor (37) is fixedly connected to the third lead screw (35).

11. A method for splicing blood cell images, applied to the blood cell intelligent detection device according to any one of claims 1-10, characterized in that, The method includes the following steps: Step S1, collect the blood cell images captured by the intelligent camera and preprocess them; input the preprocessed blood cell images into the first generation network, and use the adversarial learning mechanism to enhance the image quality to generate the first enhanced image Step S2, adopt the cell adaptive enhancement network to jointly analyze the central cells and edge cells in the same image, and use the structural information of the central cells to optimize the clarity of the edge cells; Step S3, adopt the second generation network to extract features from the optimized image to obtain the second enhanced image; Step S4, adopt the improved segmentation network to segment the cell structure of the second enhanced image to obtain the blood cell feature map; Step S5, based on the blood cell feature map, adopt the image stitching method based on feature matching to align multiple blood cell images.

Images