31results about How to "Clear edges" patented technology

The invention discloses a cell image segmentation method, and relates to the technical field of cell analysis, and the method comprises the following steps: S1, image acquisition; s2, processing optical artifacts generated by imaging of the phase contrast microscope; s3, processing interference of impurities and debris in the cell image; in the method, a phase contrast microscope and an optical coherence tomography microscope are combined to obtain a cell image, three-dimensional structure information of an OCT image is utilized to assist in correcting artifacts generated by imaging of the phase contrast microscope, and the artifacts are processed based on phase recovery and filtering and texture feature matching and repairing. According to the method, the interference of impurities and debris is removed by applying a diffusion and repair method based on a partial differential equation and a machine learning classification and reconstruction method, so that the cell image quality is effectively improved, the interference of artifacts and impurities on subsequent analysis is reduced, and a reliable image data basis is provided for cell image segmentation and cell morphology analysis.

The invention belongs to the technical field of magnetic particle imaging, particularly relates to a time sequence magnetic nanoparticle imaging device and method for detecting nerve response activity, and aims to solve the problems that the sensitivity of magnetic particle signal detection is not high, a large coil assembly is needed for achieving high magnetic field intensity, the calculation amount is large, and the calculation mode is complex. According to the device, a signal generator outputs a voltage signal, amplifies the voltage signal and then accesses a selection coil and a driving coil; the two selection coils are arranged in a Maxwell configuration and have the same axis; the driving coils are arranged at intervals of 90 degrees from top to bottom, from left to right and from front to back by taking the FFR as the center; the excitation coil, the receiving coil and the selection coil are coaxial; and the high-pass filter filters the signal received by the receiving coil and imports the signal into a computer through a signal acquisition card for time sequence fMPI image sequence reconstruction. According to the invention, the magnetic particle signal detection sensitivity is improved, the volume of the coil assembly is reduced, the calculation complexity and the calculation amount are simplified, and the magnetic particle imaging robustness is improved.

A heel backer shaper comprises a case, a last-compressing ironing base, a last rotation mechanism, a shaping wheel, an automatic selector switch, and a footboard. One side of the upper half of the case is not closed. The last rotation mechanism is disposed on the upper half of the case. The last rotation mechanism comprises a cylinder and a rotary table; the rotary table is disposed at the top of the cylinder; the last-compressing ironing base is disposed on the top inner wall of the case and right faces the rotary table. During heel shaping, the heel backer shaper can adjust the operating performances, such as the position of the shaping wheel on a backer, knocking force, ironing temperature and delay time, according to the quality conditions of a shoe so as to acquire optimal shaping effect; during normal working, the heel backer shaper can operate in an automatic manner; an electrothermal plate on the last-compressing ironing base is controlled via a three-compartment closed-loop controlled silicon voltage regulator, temperature control precision is greatly improved, a temperature difference can be controlled within a 10 DEG C range, and the plane of the shaped backer is flat, with clear edge and smooth wrinkle-free surface.

The invention relates to a lightweight semantic segmentation method for strawberry flower and fruit recognition in a complex scene, and the method comprises the following steps: obtaining a data set of strawberry images, and carrying out the image collection of strawberry flowers and fruits in a scene covering different illumination, time and plant states; carrying out data processing, and carrying out classification and mask labeling on the screened strawberry images; a deep and narrow backbone network is designed, and data features of rich and complex strawberry images are obtained; introducing more attention to the boundary of the strawberry image through an edge attention mechanism guiding model to obtain a clear edge; a pyramid pooling module is introduced at a de-wharf, and global information is perceived by aggregating strawberry image features of different scales; the strawberry image is segmented through the convolutional neural network, so that strawberry image segmentation in a complex environment is dealt with, and the lightweight and accuracy of the model are improved.

The invention discloses a pellet positioning circle measurement defect detection method and device based on an SOD-YOLOv5 model, and the method comprises the following steps: placing a plurality of pellets in grid holes of a pellet box, precisely positioning the center of each pellet through a microscope lens, detecting a pellet image through the SOD-YOLOv5 model, calculating the center offset, and determining the defect of the pellet positioning circle measurement defect. Fine positioning of the pellet is realized; then, the maximum inscribed circle of the pellet is extracted, edge detection is carried out along the circumference, and a circular contour is fitted through a least square method so as to calculate geometrical characteristics such as roundness and diameter; identifying and positioning the surface defects of the pellet by using an SOD-YOLOv5 model and a difference method; a microscope lens automatically moves to the position of each hole, and multi-pellet detection is carried out in sequence; the device can adapt to various colors and illumination conditions, high-precision, automatic and batch detection is realized, and the reliability of a detection result is ensured.

The invention discloses an automatic die-cutting detection device, which relates to the technical field of die-cutting detection, and comprises a detection mechanism, wherein the detection mechanism comprises a detection head and a backlight source, wherein a piece to be tested is installed between the detection head and the backlight source; a relaxation mechanism, which is installed on the detection mechanism and is used to relax a detection section at the top of the piece to be tested; a detection section clamping mechanism, which is installed on the top of the relaxation mechanism and is used to clamp two sides of the piece to be tested to form a detection section, wherein the piece to be tested is located between the detection section clamping mechanisms; a wrinkle processing mechanism, which is installed on the detection section clamping mechanism and is used to eliminate wrinkles formed by the stress on the piece to be tested; the invention processes the detection section through the wrinkle processing mechanism, thereby solving the influence of wrinkles on detection, and relaxes the stressed piece to be tested through the detection section clamping mechanism and the relaxation mechanism, thereby avoiding deformation of the piece to be tested and affecting light transmission, making the edge of the piece to be tested clearer and avoiding misjudgment.

The invention discloses an image enhancement method and device based on double-teacher knowledge distillation and a storage medium, and belongs to the technical field of digital image processing. According to the method, an underwater image enhancement network for double-teacher knowledge distillation is provided, and an image processing problem is divided into a plurality of small problems by the framework; a Freq-UIE network and a vision-based Fourier processing module are designed, and frequency domain information is added to different layers of the network, so that the effect of edge and detail enhancement is achieved; in order to better match and distill features of multiple teachers and students, a knowledge collection and distillation module is provided, and the features of the students and the multiple teachers are processed on each scale. Experimental results prove that compared with the prior art, the method has the advantages that the detail and texture information of the image can be well enhanced, the color recovery is more accurate, and the edge is clearer.

The invention belongs to the technical field of graphene, and particularly relates to graphene leather capable of realizing temperature-sensitive color development and a preparation method of the graphene leather. As the graphene heating functional layer adopts the high-thixotropic silver paste, the printed circuit pattern is clear in edge, uniform in thickness and extremely high in consistency with the designed pattern; and a clear interface is formed between a silver paste circuit and leather fibers, and improper penetration is avoided. And the circuit resistance value consistency is excellent (the deviation is controlled within + / -5%), so that rapid, uniform and stable planar heating can be realized after the graphene heating layer is connected, and the local overheating hidden danger is perfectly eliminated. According to the present invention, the natural texture and the soft touch feeling of the leather are maintained while the efficient and safe heating performance with the visual color developing function is achieved.

The application relates to the technical field of coating devices, and discloses a fireproof layer coating machine for fireproof glass production and a production process, wherein the fireproof layer coating machine for fireproof glass production comprises an equipment frame, a conveying assembly fixedly installed in the equipment frame, a bottom frame mechanism arranged above the conveying assembly and a coating mechanism, and the inside of the bottom frame mechanism supports a glass sample; a three-stage nested floating structure formed by first, second and third sliding sleeves provides a multi-degree-of-freedom alignment capability for a coating roller; when special-shaped glass passes, the coating roller can passively and adaptively float along the curved surface profile of the special-shaped glass, and always adhere to the surface of the glass; meanwhile, the whole coating head is flexibly hung by a plurality of supporting springs, overall buffering is provided, the problems of accumulation at protruding positions and coating leakage at recessed positions are solved, uniform transfer of fireproof coating on a special-shaped surface is ensured, and finally, a high-quality coating with consistent thickness and continuous integrity is obtained.

The invention provides a cross-domain transfer learning laser ultrasonic layering defect detection method suitable for a composite material and application thereof.The method comprises the steps that firstly, multi-channel input containing a frequency domain, a time-frequency domain and Koopman operator characteristics is constructed for the problem of data distribution deviation caused by the batch difference of the composite material; secondly, extracting and fusing high-dimensional spatial-temporal features through a multi-channel attention mechanism and a graph convolutional network; and finally, migrating the defect features of the source domain to the target domain in combination with dynamic distribution self-adaption and domain adversarial training strategies. The method can effectively overcome model performance reduction caused by material attribute change, realizes high-precision layering defect identification and imaging under the condition of no target domain label, and is suitable for composite material detection in a complex industrial scene.

The invention relates to the technical field of road construction, in particular to a spraying and marking device for road construction, which comprises a frame, a template mechanism and a height adjusting mechanism, and is characterized in that a template is arranged on a coiled material, and the shape of a preformed hole directly facing a notch is switched through a driving mechanism. The template marking device can carry more templates at the same time, so that the marking work is more convenient, and the marking efficiency is improved. And by arranging the multiple template mechanisms and the multiple injection assemblies, the scribing efficiency is improved. The absorbing body is arranged in the preformed hole of the coiled material and matched with the ejection assembly, so that the coating is more conveniently transferred to the road, and the marking efficiency is improved. The coating on the surface of the shielding body is scraped through the scraping assembly, coating waste is avoided, the pressing degree of the marbles at the edge positions of the preformed holes is smaller than that of the marbles in the middle positions of the preformed holes through the adjusting assembly, the coating at the edges of the preformed holes is prevented from being diffused, the edges of marked characters or patterns are clearer, and the working efficiency is improved. And the scribing effect is optimized.

This invention provides a metal mask for silicon-based liquid crystal light valve chips, relating to the field of semiconductor optoelectronic device manufacturing technology. It includes a metal substrate and a pixel aperture array, alignment mark area, stress buffer, and reinforced border disposed on the metal substrate, adaptable to mass production evaporation and photolithography processes for LCoS light valve chips. The metal substrate consists of four layers stacked sequentially from bottom to top: a substrate layer, a molybdenum transition layer, a Mo-Ta-N functional pattern layer, and an Al2O3-SiO2 composite passivation layer. The molybdenum transition layer is deposited on the substrate surface using magnetron sputtering, with a surface roughness Ra < 0.15 nm. The Mo-Ta-N functional pattern layer is deposited on the surface of the molybdenum transition layer using reactive magnetron sputtering. The near-vertical, low-roughness aperture sidewalls of this invention result in shadow-free evaporation, uniform electrodes, and clear edges, improving the light utilization and contrast of the LCoS chip. It also increases the utilization rate of the evaporation material and reduces material loss during mass production.

The application discloses a display panel, a display device and a preparation method of the display panel. The display panel comprises a substrate, an encapsulation layer arranged on one side of the substrate, the encapsulation layer comprising a plurality of encapsulation grooves, a first electrode layer arranged on the side of the encapsulation layer away from the substrate, the first electrode layer being provided with a plurality of first electrode openings, the orthographic projection of the first electrode openings overlapping the orthographic projection of the encapsulation grooves in the plane of the substrate, a high-refractive layer comprising a plurality of high-refractive parts, at least a part of each high-refractive part being arranged in the encapsulation groove, the refractive index of the high-refractive part being greater than the refractive index of the encapsulation layer, an electrochromic layer arranged on the side of the first electrode layer away from the substrate, and a second electrode layer arranged on the side of the electrochromic layer away from the substrate, the second electrode layer and the first electrode layer forming an electric field to control the color change of the electrochromic layer. The display panel, the display device and the preparation method of the display panel disclosed by the application can improve the wetting problem of the electrochromic layer.

The invention relates to a coal mine driving face key target detection method, and belongs to the technical field of target detection. Comprising the steps of performing foreground and background segmentation on an original image to obtain a foreground image and a background image; performing image enhancement on the background image through a CLAHE algorithm; denoising processing is carried out on the foreground image, image enhancement is carried out on the denoised foreground image through an improved CLAHE algorithm, and the improved CLAHE algorithm adaptively adjusts the value of an enhancement parameter according to an image quality index; fusing the enhanced background image and the enhanced foreground image to obtain a target image; and inputting the target image into a pre-trained improved YOLOv8n model for key target detection. The image enhanced by the method is high in contrast, low in noise and clear in edge; the improved YOLOv8n model has a larger receptive field, context information can be better extracted, the detection precision of multi-target shielding can be enhanced, and the convergence speed and the regression precision are increased.

The utility model discloses a device for collecting mosquito egg images of a mosquito and egg trapping device in multiple scenes, the main body structure of the device is composed of an industrial camera, a light supplementing lamp panel and a macro lens, and the device is characterized in that the industrial camera is embedded into a slot position of the light supplementing lamp panel and is fixed through a buckle; the light supplementing lamp panel is in a circular ring shape, and light supplementing lamps evenly distributed along the circular ring are arranged on the light supplementing lamp panel. The device body can directly cover the upper portion of the mosquito and egg trapping device with the cover opened to shoot filter paper in the mosquito and egg trapping device, and can also be fixed to the telescopic supporting vertical plate to shoot filter paper in the mosquito and egg trapping device or taken out independently. The device has the flexibility of adapting to multi-scene application, allows operators to efficiently obtain clear and countable filter paper pictures of mosquito eggs under the condition of not using a stereoscope or a microscope for observation, is convenient for subsequent work such as manual or machine identification and counting of egg granules and picture storage, and is high in practicability. And the efficiency of identifying and counting the eggs after the mosquito and egg trapping device is recovered is effectively improved.

The present application relates to the technical field of mineral material microcharacterization, and in particular to a method for dispersing clay mineral monolayer suitable for transmission electron microscope observation. The method for dispersing clay mineral monolayer based on stage alcohol intercalation and exfoliation comprises the following steps: dispersing clay mineral in deionized water to obtain a clay mineral suspension system by stirring; adding small molecule alcohol to the suspension system first and then adding macromolecular alcohol, and stirring to obtain a dilute dispersion system; performing ultrasonic treatment to make the clay mineral form a dispersion system mainly in monolayer or few-layer structure; dropping onto a carrier film for transmission electron microscope to make the lamella spread on the surface of the carrier film; and observing by using a transmission electron microscope. The present application is conducive to obtaining clay mineral monolayer samples with good dispersion and complete structure, and is suitable for morphology and structure characterization under transmission electron microscope.

This invention discloses a method for preparing patterned colloidal crystal films with hydrophilic-hydrophobic conversion effects, specifically relating to the field of colloidal crystals. The specific steps are as follows: A colloidal microsphere emulsion of a specific concentration is prepared as a coating solution; a wire rod is used to apply a constant force to the substrate surface to form a film at a uniform speed; after the solvent evaporates, a hydrophilic colloidal crystal film is formed; and then it is placed at a constant temperature for a specific post-treatment time to obtain a hydrophobic colloidal crystal film. This method utilizes the change in the hydrophilic functional groups on the surface of polymer colloidal microspheres under high-temperature conditions to achieve hydrophilic-hydrophobic conversion of the colloidal crystal film at low cost. The hydrophilic-hydrophobic conversion colloidal crystal film prepared by this invention has a simple process and low cost, and can be used in fields such as anti-counterfeiting patterns and sensor devices.

The embodiment of the utility model relates to a composite parabola curved surface reflection cup and a flashlight with the reflection cup. A first parabola curved surface and N second parabola curved surfaces formed by recessing the first parabola curved surface are formed on the composite parabola curved surface reflection cup; every two adjacent second parabolic curved surfaces are separated from each other or the edges of the second parabolic curved surfaces are overlapped; the corresponding focus of each second parabola curved surface and the focus of the first parabola curved surface are positioned on the same plane; the edge of the top end of each second parabola curved surface is overlapped with the edge of the top end of the first parabola curved surface; and the bottom end opening of each second parabolic curved surface is communicated with the bottom end opening of the first parabolic curved surface. Light emitted by the LED lamp beads can be reflected by the corresponding parabolic curved surfaces, light condensation is accurate, meanwhile, mutual interference of the light is avoided, and light condensation light spots which are different in shape or color (corresponding to the color of light emitted by the LED lamp beads), clear in edge and uniform in intensity distribution can be obtained.

This invention relates to the field of laser marking technology, and provides a control method and system for an aluminum rod laser marking machine. The method includes: identifying the contour, placement position, and height / undulation of the aluminum rod; real-time acquisition of the actual distance data between the surface of the marking area of ​​the aluminum rod and the laser head, and calculating the basic compensation amount of the laser focus based on the actual distance data; simultaneously compensating for the closed-loop system delay between the signal acquisition of the ranging module and the execution response of the XY-axis servo drive module; obtaining the final control parameters of the XY-axis dynamic displacement based on the basic compensation amount and the delay compensation; and transmitting the final control parameters to the XY-axis servo drive module to drive the laser marking module to perform XY-axis dynamic displacement, marking the aluminum rod in a non-contact manner. This invention enables precise control of the XY-axis servo drive module, achieving dynamic tracking and adhesion of the laser focus to the surface of the aluminum rod, which improves the versatility and adaptability of the equipment.

The present invention discloses a multi-frame blind restoration method for an adaptive optical system corrected image, which specifically comprises the following steps: step 1: using a gradient filter to extract high-frequency information of an f-th frame degraded image g to obtain the extracted high-frequency information; establishing an objective function based on the extracted high-frequency information to estimate a point spread function corresponding to the f-frame degraded image; dividing the objective function into a first sub-problem and a second sub-problem based on an alternating iterative algorithm; using an iterative algorithm to perform a k-th iterative calculation on the first sub-problem to obtain a high-frequency restored image x k , based on x k The iterative algorithm is used to perform the k-th iteration calculation on the second subproblem to obtain the point spread function h k ; According to the point spread function obtained by calculation, the f-th frame original image is restored by using a multi-frame image restoration algorithm containing Poisson noise. The restored image of the present invention has better visual effects and evaluation indicators.

The invention belongs to the field of colloidal gold detection, and particularly discloses a digital identification method suitable for colloidal gold detection card detection, which comprises the following specific steps: S1, image acquisition; s2, image preprocessing; s3, area detection; s4, feature extraction; s5, data comparison and analysis; s6, integrating the system; a complete application program is integrated, a mobile phone or an iPad is utilized to shoot a colloidal gold detection card, then an image is uploaded to a cloud server to be processed, one or more areas containing C and T lines of the colloidal gold detection card are identified, the length, width and color values of the C and T lines are read and calculated, then the extracted C and T line values are compared with a standard dictionary, and the detection result of the colloidal gold detection card is obtained. The content of the target object in the sample is judged, data is uploaded to a food safety big data platform, dependence on professional detection instruments is not needed, operation is easy, food enterprises, farmer markets, food workshops and families can conveniently conduct self-inspection screening on food, and food safety is protected.

This invention relates to a Gaussian super-resolution reconstruction method based on physically consistent sparsity and frequency awareness, comprising: performing feature extraction and saliency prediction on the spacecraft image to be processed to obtain a multi-scale deep feature tensor and a foreground saliency mask; using the foreground saliency mask to perform spatial feature modulation on the multi-scale deep feature tensor to obtain a modulated multi-scale deep feature tensor; decoding the feature vector corresponding to each spatial position of the modulated multi-scale deep feature tensor to predict the corresponding Gaussian kernel parameters; based on the foreground saliency mask, performing a physically consistent hard gating operation on the opacity of all Gaussian kernel parameters to generate a sparsified set of Gaussian kernel parameters; and based on the sparsified set of Gaussian kernel parameters, obtaining the final super-resolution result image through differentiable Gaussian sputtering. This method can obtain high-quality arbitrary-scale super-resolution reconstruction results while meeting the high-performance requirements of spaceborne edge devices.

The present invention relates to the use of a circular RNA molecule in the preparation of a drug for treating muscle injury, wherein the circular RNA molecule is circTXNIP, and its buffalo source sequence is shown in SEQ ID NO: 1. Overexpression of circTXNIP promotes the generation of circTXNIP-encoded peptides, thereby increasing the proliferation and differentiation of myoblasts; the circTXNIP forms a circTXNIP nucleic acid complex with liposomes as a component for preparing a drug for treating muscle injury. The circTXNIP-encoded peptide generated after the translation of the circTXNIP provided by the present invention can promote the repair of muscle injury, so circTXNIP is expected to be developed as a nucleic acid drug for the preparation of drugs for treating muscle injury, and can be used for treating muscle injury by local intramuscular injection.

The invention relates to the technical field of aviation rubber product forming, in particular to an eight-position fan-shaped bonding ultrathin combined filter disc deburring tool and method.The device comprises an upper die and a lower die, a positioning groove is formed in the middle area of the lower die, a positioning boss is arranged in the center of the bottom of the upper die, and the positioning boss is in sliding fit with the positioning groove to achieve center positioning; the upper die is of a fan structure and comprises eight fan blades arranged around the positioning boss at equal intervals, and the fan blades are attached to the eight glass fiber reinforced phenolic resin areas of the eight-position fan-shaped bonded ultrathin combined filter disc in a one-to-one correspondence mode. It is guaranteed that the phenolic glass fiber compression molding size of the deburred part is accurate, and the edge is clear and uniform; and an electric grinder is matched with a conical carborundum grinding needle to replace a file, so that the labor intensity is reduced, and the deburring efficiency is improved by more than 5 times.

The present invention discloses a text image super-resolution method, comprising the steps of: 1. correcting the text image by using a spatial transformation network; 2. extracting text features in the corrected image by using a cyclic residual block; 3. fusing and up-sampling the text features; 4. generating clear and sharp text by using a discriminator-supervised super-resolution network. The method of the present invention has simple steps, reasonable design, and convenient implementation. It can be effectively applied to the super-resolution of text images, and has significant effects. The edges of the characters in the reconstructed text images are clearer and sharper. It has broad application prospects and is easy to promote.

The invention provides an edge optimization method for a canny edge detection algorithm. The method comprises the steps of obtaining coordinates and gradient directions of all edge points by using the canny edge detection algorithm; two end points of the edge section are searched, repeated edge points are searched on the edge section, the repeated edge points are screened and removed, and the double-edge area is optimized; the invention further provides a sub-pixel-level edge optimization method, sub-pixel processing is carried out on the edge, then clustering is carried out on pixel points in the double-edge area, and repeated edge points are reduced; according to the method, repeated edge points are effectively filtered for the problem of double edges detected by canny edges, so that the edges are clearer and more accurate, and the effectiveness of a subsequent template matching algorithm is improved; the method has the characteristics of rapidness and accuracy, and is suitable for optimization processing of the canny edge detection method.

The method for super-resolution reconstruction of panoramic monitoring image of protection system of extra-high voltage converter station belongs to the technical field of power equipment detection, and solves the problems of unclearness and low resolution of the existing panoramic monitoring image, which cannot meet the demand of panoramic monitoring of inspectors; by adopting multi-scale convolution blocks to construct low-order and high-order features of images of various scales in the deep multi-scale residual network model, the phenomenon of incomplete extraction of image details is avoided, the residual learning mechanism is adopted in the network model to retain low-order rough features, reduce the training difficulty, promote the reuse of features, and thus improve the reconstruction capability of the image; the reconstructed image has better structural similarity and peak signal-to-noise ratio performance; the standard data set and the panoramic monitoring image data set of the extra-high voltage converter station are used in sequence for image super-resolution reconstruction and target recognition experiment, and the experimental results show that the high-resolution image reconstructed by the method can meet the demand of panoramic monitoring of inspectors.

The present invention relates to the field of regional segmentation technology, and specifically to a tumor medical image processing method and system for a tumor ablation treatment system. The method performs initial segmentation on the CT image of the diseased part to obtain the tissue region, screens out the initial abnormal region, and screens the final abnormal region based on the difference in the degree of shape change between the initial abnormal region and its analysis region, and the difference in the number of images between the CT image of the diseased part where the initial abnormal region and its analysis region are located, and then screens the mis-segmented region; the gradient value of the edge pixel points of the mis-segmented region is adjusted based on the difference in the position distribution of the edge pixel points of the mis-segmented region and its analysis region, and then the adjusted gradient value is used to re-segment the CT image of the diseased part. The present invention effectively improves the accuracy of regional segmentation of the CT image of the diseased part.

The present application discloses a method, device and storage medium for barcode label recognition applied to AOI, which is used to detect and decode barcode labels. The present application method includes: obtaining a screen image of a screen with a barcode attached; processing the screen image using a Laplace operator; performing grayscale threshold segmentation on the processed image, and removing the noise on the binary image after grayscale threshold segmentation to obtain a backbone binary image; segmenting each connected area in the backbone binary image into independent sub-connected areas; performing feature recognition on each sub-connected area based on Blob analysis to obtain a result image; performing Hough line detection on the result image to obtain the line direction angle of the barcode; creating a rotation matrix and setting the rotation center point and rotation angle; performing adaptive affine transformation on the result image based on the rotation center point and rotation angle; decoding and identifying the barcode.

The application relates to a face and gesture recognition method and product with a fusion attention mechanism. The method comprises the following steps: inputting a camera image into a CenterNet network to generate a face image region of interest and a hand image region of interest; inputting a face image in a database into a twin neural network with a fusion channel and spatial attention mechanism to generate a face image feature, and inputting the face image region of interest into another twin neural network to generate a face image feature; comparing the face image feature and the face image region of interest feature to recognize a face and generate a face recognition result; inputting the hand image region of interest into a ResNet network based on a multi-scale fusion mechanism to perform semantic segmentation and generate a hand binary image; inputting the hand binary image into a classification network to generate a hand recognition result; and controlling an intelligent wheelchair according to the face recognition result or the hand recognition result. The application can improve the efficiency and precision of face and gesture recognition.