424 results about "Single pixel" patented technology

The embodiment of the invention provides an optical detection device which comprises an imaging assembly and a photoelectric detection device, and the imaging assembly is used for forming an image on an image plane of a to-be-detected area of a wafer. And the photoelectric detection device is used for receiving images and is arranged at the image plane. The photoelectric detection device comprises a light beam separation assembly and a sensor assembly, the light beam separation assembly is used for dividing a continuous first light signal in an image into multiple beams of second light signals, and the light beam separation assembly comprises a microlens array. The sensor assembly is used for receiving multiple beams of second optical signals, and each pixel unit of each sensor in the sensor assembly is used for receiving one beam of second optical signal in the multiple beams of second optical signals. The sensor assembly comprises a single-pixel sensor, and the single-pixel sensor corresponds to one micro lens in the micro lens array; and / or, the sensor assembly comprises a multi-pixel sensor, the multi-pixel sensor corresponds to a plurality of microlenses in the microlens array, and spatial segmentation is beneficial to improving the signal-to-noise ratio.

The invention relates to the technical field of MIP devices, solves the problems that in the prior art, a single-pixel MIP device is large in pin number, low in screen body contrast ratio and high in manufacturing cost, and provides an all-in-one MIP device and a manufacturing method thereof.The manufacturing method comprises the steps that a substrate to be processed is obtained, and bonding pads and copolar lines are formed on the second surface of the substrate through deposition; depositing and forming an insulating layer in a non-pad area in the second surface; forming a through TGV conductive through hole in the substrate along the thickness direction, and forming a conductive metal layer in the TGV conductive through hole; forming back pin metals in one-to-one correspondence with the TGV conductive through holes on the first surface to form a pin array; welding the LED chips on the bonding pad in a flip-chip manner, and arranging the LED chips at intervals to form pixel units; and a packaging layer is deposited and formed on the LED chip and the gap area of the LED chip, and the all-in-one MIP device is obtained. The number of the pins is reduced, the contrast ratio of the screen body is improved, and the manufacturing cost is reduced.

The invention discloses an ultra-low illumination-oriented SPAD active imaging data compression method, which comprises the following steps of: performing wavelet decomposition on histogram data of a single pixel in a time-frequency domain based on wavelet transform to obtain low-frequency data of each pixel after wavelet decomposition; by taking each pixel as a center, performing non-maximum suppression and data enhancement on the low-frequency data after the wavelet decomposition of the current pixel by using adjacent pixels to obtain processed compressed data; customizing different Gaussian kernel parameters for each pixel according to the possibility that each pixel is located at the boundary, and performing Gaussian filtering on the processed compressed data of each pixel to obtain filtered data; and performing depth estimation on the filtered data to obtain a final depth image. According to the method, the depth reconstruction performance of the laser pulse can be improved by utilizing the multi-resolution characteristic of wavelet transform, the space-time correlation of signal photons and the smoothness of Gaussian filtering.

The invention discloses a feature hierarchical attention fusion and dynamic optimization method for single-stage target detection, and relates to the technical field of computer vision. The method comprises the steps of collecting image data, preprocessing an image and dividing the image into a training set, a test set and a verification set; the HAF-DRFPN neck network uses dynamic up-sampling single pixel point sampling to recover the feature resolution; feature extraction is optimized by using a gating residual mechanism and depth separable convolution; a feature refining feed-forward network is introduced through multi-layer perception generation weight, and feature expression is enhanced; respectively capturing and fusing superficial details and deep semantic information by utilizing channel and space attention and cross attention; a multi-branch dynamic sampling stacking framework is adopted, and cross-scale features are adaptively fused and optimized; multi-module cooperative processing is carried out on the whole architecture, and multi-scale features with higher discrimination are provided for the detection head; according to the method, the YOLO12 trunk and the detection head are connected with the HAF-DRFPN to form the model HAF-DRNet suitable for target detection, so that the flexibility and accuracy of single-stage target detection are improved.

The invention relates to a pavement crack image skeleton extraction method based on an improved Zhang-Suen refinement algorithm. The method comprises the following steps: S1, carrying out binarization processing on a pavement crack image; s2, carrying out boundary expansion on the crack binary image; s3, carrying out run-length coding on the crack binary image in the horizontal direction, and carrying out compression storage on continuous non-zero pixel segments in each row; s4, using an improved Zhang-Suen refinement algorithm to process the crack binary image processed in the step S3; s5, removing boundary extension; s6, performing simplification processing on the preliminarily refined crack skeleton image by using four predefined 3 * 3 L-shaped structure templates; s7, removing burrs in the image; and S8, outputting a crack skeleton image. The technical problems that according to an existing skeleton extraction method, a topological structure is prone to being damaged, the calculation efficiency is low, the extraction result cannot be completely a skeleton with the single pixel width, and burrs exist in the skeleton extraction result are solved.

The invention discloses an electric power meter intelligent reading system based on image recognition, and relates to the technical field of computer vision. The method is used for solving the problems of poor instrument reading stability and low range identification accuracy in a complex environment. The method comprises the following steps: firstly, performing multi-scale retina processing and four-direction gradient integral response analysis through an anti-shielding dial plate detection module, generating a probability heat map based on symmetry distribution, and outputting an accurate dial plate region segmentation mask through channel fusion; secondly, extracting a single-pixel wide pointer skeleton through non-maximum suppression and morphological refinement by using a meta-learning skeleton extraction module, and obtaining scale line distribution characteristics based on radial gradient analysis; thirdly, performing three-level progressive verification through a context range analysis module in combination with a geometrical relationship, a knowledge graph and a logical relationship, and outputting a range parameter; multi-dimensional information fusion is adopted to evaluate segmentation quality and skeleton integrity and verify consistency, reliable reading is generated, credibility evaluation is carried out, and high-precision stable reading in a complex environment is realized.

The invention provides an infrared image sensor and a pixel-level reading circuit thereof, and relates to the field of infrared image sensors. The pixel-level reading circuit is integrated in a single pixel unit, and comprises a direct injection input-level circuit used for converting an infrared light-generated current generated by an infrared image sensor into an analog voltage signal; the pulse frequency modulation analog-to-digital converter is connected with the direct injection input stage circuit and is used for converting the analog voltage signal into a pulse signal; the SR latch is electrically connected with the pulse frequency modulation analog-to-digital converter and is used for carrying out accumulative counting on the pulse signals so as to form digital code words reflecting the infrared light intensity; and the digital logic control unit is electrically connected with the SR latch and is used for transmitting the digital code word to a reading bus of the infrared image sensor, so that the output of a digital signal is realized, and the reading rate of the image sensor is improved.

The embodiment of the invention provides a photoelectric detection device and an optical detection device. The photoelectric detection device comprises a light beam separation assembly and a sensor assembly. The light beam separation assembly is used for dividing a continuous first light signal into a plurality of beams of second light signals at different spatial positions. The sensor assembly is used for receiving multiple beams of second optical signals. The sensor assembly includes a plurality of sensors, the plurality of sensors including single-pixel sensors. The continuous first optical signal is divided into a plurality of small discrete second optical signals through the light beam separation assembly, so that the light sensing area of each sensor corresponding to a sample is further reduced, the haze noise on a single sensor is further reduced through the small light sensing area, and the haze noise on the single sensor is further reduced under the condition that the original first optical signal is not changed. Noise on a single sensor is reduced, and the signal-to-noise ratio of the photoelectric detection device is improved.

The invention relates to the technical field of optical imaging, and discloses a single-pixel color imaging system based on multi-wavelength radio frequency marking. The system comprises a light source module used for generating three laser beams with three optical primary colors; the radio frequency driving module is used for generating a radio frequency complex frequency signal; the light path adjusting module is used for performing heterodyne interference processing on the three laser beams according to the radio frequency complex frequency signal, respectively forming a light spot array and a reference light source and then combining the beams, and generating a combined beam to be incident on a measured target; each laser beam corresponds to one group of sub light spot arrays; sub light spot arrays with aligned spatial positions exist among the three groups of sub light spot arrays; the detected target detection module is used for collecting light data generated after a detected target reflects the combined light beam; and the image inversion module is used for performing image inversion processing on the optical data to generate a true color image of the measured target. According to the system, the light spot array of multi-wavelength laser incidence can be registered in the spatial position.

The invention discloses a self-adaptive underwater single-pixel imaging technology based on Fourier modulation, belongs to the technical field of computational optical imaging, is used for ocean detection and improvement of ocean surveillance and defense efficiency, and adopts a self-designed LED array to replace an expensive DMD and a four-quadrant detector composed of common photodiodes to replace an expensive barrel detector. And high-speed illumination and parallel acceptance are realized while the cost is greatly reduced. The technical route comprises a high-speed underwater single-pixel imaging device, a high-speed Fourier single-pixel imaging algorithm and an adaptive sampling strategy based on Fourier single-pixel imaging. The high-speed single-pixel imaging device comprises a self-designed LED array light modulation module, a converging lens module and a receiving module. The high-speed Fourier single-pixel imaging algorithm comprises a binarization algorithm based on error diffusion and an algorithm based on adaptive compressed sensing. According to the innovative self-adaptive sampling strategy based on Fourier single-pixel imaging, a specific sampling path is generated by fusing circular sampling and random sampling according to the characteristics of an image in a Fourier spectrum, so that Fourier coefficients which have great influence on image quality are acquired as much as possible under water with low sampling rate and high turbidity, and the image quality is improved. Therefore, the high-efficiency imaging of the underwater target object is realized, and the imaging speed and quality are both considered.

The invention discloses a machine vision-based micro-plastic intelligent identification and measurement method, which comprises the following steps: filtering a water sample containing micro-plastic or not containing micro-plastic, and acquiring images under a uniform amplification factor by using a microscope to obtain a micro-plastic image to be detected and a contrast image; calculating the absolute value of the color centroid value of the contrast image and the color difference value of the micro-plastic image, generating a difference image, and carrying out denoising, graying and binarization processing on the difference image to obtain a binary image with highlighted micro-plastic; extracting pixel coordinates of the micro-plastics, setting a threshold value to carry out connected clustering, and distinguishing the micro-plastics and image impurities in combination with a preset micro-plastic recognition threshold value to realize micro-plastic recognition and quantity statistics; and further shooting a standard objective lens micrometer image, calculating the actual physical size represented by a single pixel, and automatically calculating the area and the length of the micro-plastic by combining each clustering enclosing rectangle and the number of the pixels. The method is accurate in recognition and high in operation automation degree, and can be widely applied to intelligent recognition and analysis of water body micro-plastics.

The invention discloses a pulse terahertz time-domain spectroscopy system of a single photoconductive antenna and a detection method, and relates to the technical field of terahertz, the system comprises a laser pulse generation module, a single photoconductive antenna module, a compression sampling module, a signal reconstruction module and a control unit; by integrating the random light delay modulation unit and the single-pixel detector, the system realizes random time interval sampling of terahertz signals, effectively breaks through the limitation of a traditional sampling mode, improves the flexibility and efficiency of signal acquisition, and combines a signal reconstruction module of an FPGA + GPU heterogeneous computing architecture, so that the system has a wide application prospect. The system can adaptively construct a sparse base and perform accurate sparse representation on a measurement vector, so that an original terahertz time domain signal is efficiently reconstructed by optimizing a target function and an adaptive iterative solution algorithm, the signal reconstruction precision is improved, the data processing complexity is also reduced, and the method is suitable for large-scale popularization and application. And powerful support is provided for practical application of the terahertz technology.

The invention provides an industrial anomaly classification method and system based on a multi-modal large language model, and belongs to the technical field of visual detection. The visual pre-screening module adopts a BYOL unsupervised feature extractor and a local block matching technology to calculate the minimum Euclidean distance between an image to be detected and a normal sample so as to generate an abnormal thermodynamic diagram, and efficient preliminary screening is realized; the coordinate set is converted into a binary mask through an abnormal contour marking module, and an accurate single-pixel contour is generated through morphological gradient and SIFT displacement correction; fusing a CLIP visual encoder and a GPT text encoder in a multi-modal decision engine, realizing visual-semantic alignment through cross-modal fusion and a multi-head attention mechanism, and supporting alpha hybrid dynamic rule update; and a dynamic classification executor calculates an abnormal region risk index based on the risk grading model, and generates a grading response action and a structured report. According to the invention, the classification precision is improved, so that the method can adapt to complex and diversified industrial abnormal scenes.

The invention provides a fast color single-pixel imaging system and method based on an FPGA and a DLP projector, the fast color single-pixel imaging system comprises the DLP projector, a single-pixel photoelectric detector, an AD data acquisition card, a Raspberry Pi and the FPGA, the Raspberry Pi is adopted as an upper computer, the FPGA and the AD data acquisition card are dispatched to work cooperatively, and the DLP projector and the single-pixel photoelectric detector are integrated. The rapid and stable projection of the substrate pattern and the accurate sampling of the corresponding reflected light intensity information are realized, and the efficient reconstruction of the color image of the target scene is further realized. In a data acquisition stage of an imaging system, an HDMI image data output task is executed by the FPGA, and the FPGA outputs an image frame synchronization signal, so that the stability problem and the signal synchronization problem when a substrate pattern is played at a high frame rate (60Hz) are solved by making full use of the advantages of stable and controllable sequential logic and no software level scheduling delay of the FPGA, and the stability problem and the signal synchronization problem when the substrate pattern is played at a high frame rate (60Hz) are solved. And the influence of image tearing, frame loss and frame repetition on image reconstruction is eliminated.

The invention discloses a curved broken filament cluster-shaped region closing and defect area quantification method, which comprises the following steps of: firstly, performing image acquisition on a target region containing a curved broken filament defect to obtain an original image; preprocessing the original image to obtain a binary defect area image, and performing skeletonization processing to obtain a defect skeleton curve of a single pixel width; segmenting the skeleton curve and carrying out arc fitting to obtain a fitting circle and a starting point, an ending point and a midpoint of an arc section; judging whether the arc section exceeds a semicircle or not by constructing a triangle, and complementing the arc section exceeding the semicircle by using a fitting circle to form a closed area; and finally, combining all the closed areas to generate a defect mask, calculating an actual area and judging whether the defect is qualified or not according to a preset threshold value. According to the method, the bent broken filaments are subjected to reasonable ball-shaped region closing treatment, so that the quantized defect area is highly consistent with a manual visual inspection result.

The invention relates to a skeleton line intersection region redundancy restoration method based on distance constraint and center substitution, and belongs to the field of image processing, and the restoration method comprises the steps: extracting a 3 * 3 neighborhood of each foreground pixel point in a skeletonized image, temporarily setting a center pixel as a background value, and detecting an intersection point through an 8-neighborhood connection marking algorithm; realizing cross point clustering through Euclidean distance calculation and a breadth-first search algorithm based on the spatial proximity relationship of the cross points; and carrying out group screening, ignoring a group only containing a single cross point, and carrying out topology reconstruction on a group containing a plurality of cross points. According to the method, the redundant pixel cluster is accurately identified, the quality of the skeleton line is improved, the redundant pixels in the diagonal line area are accurately identified and cleared through the template matching algorithm, the skeleton line is ensured to strictly meet the single-pixel width requirement, and false endpoints and line length measurement errors are avoided.

The invention relates to the technical field of defect detection, in particular to a defect detection method and system based on a microwave frequency crystal resonator, and the method comprises the steps: forming a microwave probe through a microwave frequency crystal resonator array, placing a to-be-detected material below the microwave probe, and setting a scanning area and scanning step lengths in two directions on the surface of the material; controlling the probe to perform point-by-point scanning according to step length, and collecting a plurality of reflection coefficients of each sampling point to form an echo signal matrix; performing singular value decomposition on the matrix, applying an adjustable coefficient to a first singular value and a second singular value, and adaptively selecting an optimal coefficient according to imaging quality to reconstruct an echo signal; performing two-dimensional Fourier transform on the reconstructed matrix, constructing a spatial filter in combination with probe spacing and defect depth to perform phase compensation, performing two-dimensional Fourier transform to obtain a defect image, and performing threshold segmentation on the defect image to obtain a binary image only containing defects; and analyzing the connected regions to obtain the gravity center position of each defect and the defect area determined by the pixel number and the single pixel area. The problems of serious clutter interference, insufficient defect imaging contrast ratio, difficulty in quantitative evaluation of defect size and the like in the existing microwave detection can be solved.

The invention relates to a gallium oxide defect detection method. The method comprises the following steps: step 1, obtaining exciting light with a proper wavelength; 2, modulating the exciting light obtained in the step 1 so as to enable the exciting light to accurately enter a microscope system with proper energy; 3, irradiating the modulated exciting light to a gallium oxide sample on an electric displacement table through a microscope system, collecting fluorescence generated after the gallium oxide sample is excited through the microscope system, and transmitting the fluorescence to a spectrograph system to obtain a corresponding fluorescence spectrogram under a preset single pixel point position; and 4, setting a spectrograph system to collect the fluorescence intensity under a specific wavelength, and driving the gallium oxide sample to move through an electric displacement table so as to scan the fluorescence intensity of all pixel point positions in a sample detection range, thereby obtaining a gallium oxide defect fluorescence intensity imaging graph in the sample detection range. The method has the advantage of high sensitivity, can realize rapid nondestructive detection, and also reduces the detection cost at the same time.

The invention provides an IMU sensor space-time fusion industrial defect detection system and method based on an SORT algorithm. The method comprises the steps that S1, system initialization and space-time calibration are carried out, a time synchronization model is established, the single pixel size dpixel is calculated, and IMU and camera coordinate system registration is completed; s2, collecting asynchronous data, and obtaining working parameters of a camera and the speed vt of a conveyor belt; s3, constructing an SORT state vector and predicting an IMU enhancement state, detecting defects of a current frame by adopting YOLOV8, and generating a detection set Dt; s4, executing EKF space-time fusion, and predicting and updating a defect state covariance and a position; s5, quantifying the position uncertainty sigma pos, calculating an optimal acquisition period Topt, controlling the line frequency of the camera, and carrying out confidence verification and misjudgment elimination; s6, cloud federated filtering fuses the edge and the cloud state, IMU zero offset is estimated, and cooperation of multiple detection devices and system expansibility enhancement are achieved.

The invention discloses a multi-gray-scale cholesterol liquid crystal display device and an array substrate thereof, a plurality of pixel units are arranged on a substrate body, each pixel unit comprises a plurality of sub-units, so that the different pixel units present the same or different gray scales, each sub-unit comprises a sub-pixel electrode and a TFT device connected with the sub-pixel electrode, and the TFT device is connected with the sub-pixel electrode. The areas of the sub-pixel electrodes of at least two sub-units are different, and the width-to-length ratios of the channel structures of the TFT devices of at least two sub-units are different. A single pixel unit is divided into a plurality of sub-pixels with different areas, different reflectivity can be achieved by controlling the areas or the sum of the areas of whether the sub-pixels are lightened or not, then the purpose of gray scale display is achieved, meanwhile, a suitable TFT device is configured for each sub-pixel, the sub-pixels with different areas are made to keep the same high charging rate, and the display effect is improved. And pixel opening compression is reduced as much as possible, so that the gray scale display effect is ensured.

A spectral sensing device and associated methods for enhanced diagnostic capabilities across medical fields are disclosed. The device incorporates multispectral, hyperspectral, and photoplethysmography technologies to analyze biological features including tongue, oral cavity, saliva, teeth, and other tissues. The system transforms any standard RGB camera into a multispectral imaging device by training machine learning models to associate RGB values with spectral values inside and outside the visible spectrum using measurements from a single pixel detector spectroradiometer. A multimodal neural network with separate processing branches analyzes spectral images, non-spectral images, and tabular data to predict microbiome composition. The system generates health risk assessments by comparing predicted microbial abundances to thresholds associated with various medical conditions. Near-infrared sensing enables blood flow monitoring, subcutaneous imaging, and assessment of physiological parameters. Convolutional neural networks enable spectral reconstruction without requiring the physical device, while supporting continuous monitoring in various form factors.

According to the multi-mode single-pixel compression imaging system based on the spatial light modulator provided by the embodiment of the invention, a light source supply module in the system can provide a broadband light source, and light emitted by the broadband light source can irradiate a target object and is reflected to a light field modulation module; the light field modulation module performs spectrum separation on the received light signal, and after multiple beams of monochromatic light after separation are incident to the spatial light modulator, the spatial light modulator performs light splitting and parallel modulation; the single-pixel detection module feeds back the collected light intensity data to the data processing module when detecting the modulated light signal; the data processing module issues a mode switching instruction to the spatial light modulator to perform mode switching when determining that the light signal reaches an ideal modulation state, and performs joint optimization modeling and iterative solution calculation based on multi-mode imaging light intensity data to realize synchronous reconstruction of a target object image. The system can realize joint acquisition and fusion reconstruction of multi-modal information.

The invention discloses an automatic detection method and system for the boundary dimension of a shield segment, and relates to the field of civil engineering, and the method comprises the steps: laying quantum dots on the surface of a to-be-detected shield segment, and exciting the quantum dots through ultraviolet laser to generate a fluorescence spectrum; the method comprises the following steps: receiving an optical signal penetrating through a medium gap in a to-be-detected shield segment through a single-pixel barrel detector to form a reflected light intensity sequence, and inputting the reflected light intensity sequence and a speckle pattern into a correlation reconstruction algorithm to generate a three-dimensional point cloud of a shielding area; and inputting the three-dimensional point cloud and the deformation compensation parameters into a deep learning network, generating a deviation thermodynamic diagram of the boundary dimension of the to-be-detected shield segment, automatically marking an area exceeding a tolerance threshold value based on the deviation thermodynamic diagram, and generating a detection result of the boundary dimension of the to-be-detected shield segment. According to the method, the internal medium gap is reconstructed through the single-pixel barrel detector and the basis tracking noise reduction algorithm, in a bolt hole oil stain shielding scene, compared with a traditional method, the reconstruction porosity error is reduced, and online analysis of the full-surface size error is achieved.

According to the image calibration and splicing method based on sub-pixel reconstruction and elliptic coefficient weighting, a super-resolution reconstruction module is introduced before image array splicing, the splicing precision is improved through sub-pixel-level image registration, and the splicing precision better than that of a single pixel unit can be achieved. Aiming at the problems of edge dislocation and stiff transition existing in multi-camera array system view field splicing, the invention provides a weighted splicing algorithm based on an elliptic coefficient, and the algorithm constructs an elliptic gradient weight model, realizes smooth transition of an image boundary region, and remarkably improves the overall visual effect of a panoramic image.

The invention relates to a method (50) comprising the steps of: - using a sensor to acquire (110) an image of the sample composed of grains; - applying (120), to the image (1), a convolutional neural network in order to obtain a contour map (C) associating, with each pixel of the image, a value equal to one unit when that pixel belongs to a contour of a grain and otherwise equal to zero; - applying (130), to the contour map (C), a segmentation algorithm in order to obtain a segmented image (S) comprising a plurality of regions, one region being separated from a neighbouring region by a contour having a thickness of a single pixel; and - applying (140), to the segmented image (S), one or more geometric processing operations in order to calculate physical quantities for each grain so as to characterise the lithology of the sample (3).

The invention provides a display device and a preparation method thereof, and belongs to the technical field of semiconductors. The display device comprises a driving back plate, an insulating reflecting layer, an electrode layer and a plurality of pixel units, the plurality of pixel units are respectively bonded on the driving backboard, the plurality of pixel units are arranged at intervals, each pixel unit comprises an epitaxial layer and a passivation layer, the epitaxial layer comprises a first semiconductor layer, an active layer and a second semiconductor layer which are stacked in sequence, and the passivation layer covers the side surface of the epitaxial layer; the insulating reflecting layer covers one side, far away from the epitaxial layer, of the passivation layer, and a groove is formed between every two adjacent pixel units; the electrode layer is located in the groove. According to the invention, the problem of low occupation ratio of small-angle light extraction efficiency of a single pixel unit can be effectively solved.

The invention relates to a method and device for detecting a stripe lap joint or intersection phenomenon in an image, and belongs to the technical field of image processing, and the method comprises the steps: carrying out the preprocessing of a to-be-detected image, and extracting a binary image of foreground stripes; carrying out refining processing on the binary image to obtain a stripe skeleton image with a single-pixel width; calculating the global average spacing of stripes in the image; traversing pixel points in the stripe skeleton image, and based on eight neighborhood features of the pixel points, preliminarily judging and marking potential lap joints or intersection points as a suspected defect point set; and for each suspected lap joint or intersection point in the suspected defect point set, calculating the maximum connection depth of the suspected lap joint or intersection point, and comparing and verifying the maximum connection depth with a threshold value set based on a global average distance to determine a real lap joint or intersection defect. According to the invention, the problem of high false alarm rate of stripe defect detection in a complex industrial image is solved, and the accuracy and reliability of detection are remarkably improved.

The invention discloses an optical fiber single-pixel imaging method and device based on sparse optimization and neural network fusion, and the method comprises the steps: constructing a collimation illumination system, carrying out the spatial modulation of incident light, introducing a collimation illumination mechanism, and obtaining a spatial speckle illumination light field of a multimode optical fiber; performing sparsity extraction analysis on the spatial speckle illumination light field by using a sparse optimization algorithm, removing a high redundancy mode, and selecting a speckle illumination mode with the lowest cross correlation; a speckle illumination mode after sparse optimization is used as input, a neural network model constructed based on a soft threshold iterative algorithm is trained, and an optical fiber single-pixel image reconstruction model is established; performing sparse optimization speckle illumination on a target object, acquiring corresponding total light intensity response data, inputting the data into the trained optical fiber single-pixel image reconstruction model, realizing image reconstruction and outputting a final target image; the method has anti-defocusing capability and anti-disturbance robustness, and can still keep good optical fiber single-pixel image reconstruction performance under the condition of low sampling rate.

The invention belongs to the technical field of manipulator tool calibration, and particularly relates to a manipulator micro-pen direct writing tool calibration method, device and system based on machine learning and monocular vision. Comprising the steps that an industrial camera and a direct-writing micro pen are installed under a manipulator tail end coordinate system, the manipulator and the industrial camera are connected to a PC end to complete hand-eye calibration, and a hand-eye matrix is obtained; installing different direct-writing micro-pens, collecting data, and obtaining a plurality of groups of micro-pen central point pixel coordinates and space coordinates in a tail end coordinate system to form point pair data; the hand-eye matrix parameters are optimized through the point pair data; establishing an error compensation model through a pinhole camera model and the single-pixel step length adjustment factor, and carrying out collaborative optimization on the single-pixel step length adjustment factor by adopting point pair data and the optimized hand-eye matrix in combination with the error compensation model; and calculating space coordinates under pixel coordinates by using the optimized hand-eye matrix and parameters to complete calibration. According to the invention, high-efficiency and high-precision calibration can be carried out on different models of manipulators and different specifications of micro-pen printing heads.

The invention discloses a correlated imaging impurity detection method and system for drug production, and relates to the technical field of drug impurity detection. The method comprises the following steps: loading a first binary mask matrix, projecting a target product area, controlling a single-pixel detector to detect, determining a first measurement value, and performing lightweight reconstruction to obtain a first preview; generating a second binary mask matrix and performing projection and detection reconstruction by pre-checking the first preview, and performing multi-round iteration until an Nth preview is determined; calling the first preview to the Nth preview from a temporary database of an online detection platform, executing multi-layer compressed sensing and reconstruction, and determining a reconstruction result; and aiming at a reconstruction result, matching in an impurity feature library, and determining an impurity verification result. The technical problems of low impurity detection efficiency and insufficient detection accuracy in the medicine production process in the prior art are solved, and the technical effect of efficient and accurate online detection of medicine impurities in the production link is achieved.