1048 results about "Image matching" patented technology

The invention discloses an automobile part quality detection method and system based on artificial intelligence visual inspection, and belongs to the field of artificial intelligence machine visual inspection, and the method comprises the steps: firstly, carrying out the registration of a collected RGB image and a depth image, and extracting a part region through a saliency detection network; two-dimensional key points are extracted based on an RGB region image and are matched with key points of a three-dimensional model, an initial three-dimensional attitude is obtained by adopting a PnP algorithm, iterative registration is performed with the three-dimensional model in combination with a point cloud generated by a depth image, and a fine three-dimensional attitude is obtained. And calculating a geometric transformation matrix from the part to a standard front view attitude according to the attitude, and performing attitude correction on the RGB and depth region image. And then matching the corrected image with a standard template image by using a feature detection and matching network so as to correct the position of the detection window. And finally, the three-dimensional size of the part is calculated in the corrected detection window in combination with the depth value, and tolerance judgment is carried out. And the precision, the robustness and the automation level of online detection of the automobile parts can be obviously improved.

The invention relates to an automatic visual alignment method and system based on a photoetching machine and a storage medium, and relates to the technical field of optical alignment. The automatic visual alignment method comprises the following steps: sequentially focusing image taking areas according to a preset multi-light-source visual focusing mechanism, recording corresponding global focusing parameters, and collecting corresponding original single-layer images; screening all the original single-layer images according to a preset image definition threshold value to obtain a to-be-detected image, and extracting corresponding key channel features; calculating a comprehensive feature error of the key channel features according to a preset image hierarchy template, and judging an initial focusing state in combination with a preset parameter error interval; correcting the global focusing parameter according to the initial focusing state in combination with a preset cooperative calibration mechanism, generating a corresponding control signal, and cooperatively tuning the corresponding visual equipment; through multi-light-source multi-time visual image capturing, the problem of poor alignment precision caused by light source errors and image capturing errors is reduced, the image matching accuracy is improved, and the visual alignment precision is improved.

The invention relates to the technical field of artificial intelligence, can be applied to business scenes such as financial science and technology and medical health, and discloses an image generation method, device and equipment based on theme information and a medium. The method comprises the steps that input information is analyzed to generate theme information and copywriting information, a cue word set is generated, and a figure image set and a background image set are generated; fitting the segmented figure image with the background image to form a head image candidate set, and selecting a head image matching template frame to generate a basic image; decomposing the copywriting to generate a sub-module initial picture set, and adding a gradual change effect to form a sub-module picture set; the adjusted sub-module pictures are obtained based on size adjustment, and a pre-synthesized image is generated through splicing; and identifying the blank area to draw a title text to obtain a final image. Through theme analysis, template matching, image splicing, modular copywriting processing and blank drawing, poster generation efficiency is improved, layout flexibility is enhanced, and visual unification is realized.

The invention relates to a weak supervision text-pedestrian image matching method based on local and global dual-granularity identity association. The method comprises the following steps: generating an information asymmetric sample pair which comprises an information imbalance image and an information imbalance text description; extracting original image features, information imbalance image features, original text description features and information imbalance text description features; constructing a global image feature library and a global text description feature library; training a cross-modal image-text feature alignment capability based on a contrast learning loss function; constructing an intra-batch cross-modal identity association relationship in the local granularity, and reinforcing the fine-granularity feature difference capturing capability through inter-modal identity constraint; a dynamic cross-modal identity association network is constructed by taking a visual modal as an anchor point in global granularity, and weak association sample recognition sensitivity is improved in combination with a confidence coefficient dynamic adjustment mechanism; fusing information asymmetric sample pairs to construct a consistency learning mechanism; according to the method, the cross-modal matching precision is remarkably improved, and an efficient solution is provided for text-pedestrian image matching.

The invention provides an inspection task dynamic scheduling system based on endurance prediction and a load balancing method, and belongs to the technical field of train inspection. The system comprises an endurance prediction module, a task decision-making module, a load balancing scheduling module, a credibility calculation module, a position correction module and the like. Residual endurance time and stop time are generated through an endurance prediction module, and a task decision module generates a task transfer data packet at the stop time; the load balancing scheduling module screens candidate robots based on the total demand energy consumption and the time adaptability, and determines a replacing robot by combining the comprehensive load value of task interval correction; the credibility calculation module generates a position credibility score through image matching and track calibration, and the position correction module dynamically adjusts the position to guarantee the precision. According to the method, the adaptation precision and the position evaluation accuracy of the candidate robots are improved, the global resource configuration is optimized, the task interruption rate is reduced, and the inspection task execution efficiency and stability are enhanced.

The invention relates to the technical field of multi-modal image matching, and discloses a multi-modal image matching method based on a parallel multi-scale cascade Transform. The method comprises the following steps of: extracting coarse-level and fine-level features from an input multi-modal image pair by using a feature extractor; performing association modeling on the coarse-level features through a parallel multi-scale cascade Transform, and dynamically adjusting channel weights by using a channel weight adaptive module in the coarse-level features so as to optimize feature representation; a local feature dynamic enhancement module is used for enhancing local information expression, and the detail capture capability and the multi-scene adaptability of features are improved; constructing a score matrix based on the enhanced coarse-level features, and realizing coarse matching through a mutual nearest neighbor criterion; by introducing a refining layer, a rough matching result is gradually optimized and adjusted on fine-level features, and finally high-precision matching is achieved; the method shows excellent performance in a multi-modal image matching task, can effectively cope with cross-modal characteristic changes in a complex scene, and is suitable for various application scenes in computer vision.

The invention discloses a cylindrical roller trajectory tracking method and device, and the cylindrical roller trajectory tracking method comprises the following steps: obtaining an actual measurement displacement sequence of a key point on the end surface of a roller under a global coordinate system based on a multi-reference matching digital image correlation method; drawing a key point displacement curve about key points and time according to the actually measured displacement sequence; judging the motion state of a roller according to the shape characteristics of the key point displacement curve, dividing a smooth part in the key point displacement curve into a rolling area, and dividing a part with jumping or sharp points into a sliding area; calculating the actual revolution angular velocity and the actual rotation angular velocity of the roller based on the actually measured displacement sequence corresponding to the sliding area; according to the method and the device, the image matching precision can be improved, meanwhile, the rolling and sliding components of the roller are accurately distinguished, the precision of the track of the cylindrical roller is improved, and high-precision and visual experimental data are provided for existing dynamics and tribology models.

The invention discloses an AI large model-based arrival person screening method and device, a medium and equipment, and belongs to the field of screening, and the method comprises the steps of firstly obtaining screening conditions input by a user, including text content, image data and qualification information requirements, and to-be-screened arrival person data; next, performing word segmentation, keyword extraction and semantic matching on the text content of the person arrival data by utilizing a preset semantic analysis model in combination with BiLSTM, CRF and LDA technologies, and outputting a semantic score; meanwhile, note styles are recognized through the text classification model, logic judgment is conducted, and styles and logic verification scores are obtained. In addition, element detection and style verification are carried out on the image data through the multi-modal recognition model, and image matching scores are output; the qualification scoring module calculates qualification scores according to a preset weight formula. And finally, fusing the multi-dimensional scores to generate a comprehensive score, and carrying out accurate screening on the arriving persons according to the comprehensive score.

The invention provides a camera pose estimation method based on 2D Gaussian splash. The camera pose estimation method comprises the following steps: acquiring a depth map and a normal map from a training image pose based on a pre-trained 2D Gaussian splash model; acquiring a grid ray starting point based on the depth map; generating a main ray and a hemispherical ray for each sampling point based on the normal diagram and the ray starting point; constructing a ray and image matching network and a lightweight convolutional neural network, and performing network training through a loss function; estimating the position of the camera through the trained ray and image matching network and the lightweight convolutional neural network based on the ray features of the main ray and the hemispherical ray and the image features of the query image, and constructing a rotation matrix to realize initial camera pose estimation; and optimizing the initial camera pose by adopting a depth-guided pose optimization method to obtain a final optimized camera 6D pose estimation result. The method gets rid of the dependence of an initial value, and has the advantages of high precision, strong robustness, efficient calculation and wide application.

The invention discloses an automatic generation method for health science popularization articles, and belongs to the crossing field of computer technology and health science popularization. The method comprises the following steps that S1, health authority information is collected through a multi-source crawler, and an original material pool is obtained through Sension-BERT vectorization duplicate removal and BioBERT medical entity labeling; s2, classifying nine types of health themes by using a RoBERTa fine tuning model; s3, multiple AI model supplementary materials are made into a structured package; s4, predetermining optimal selection questions in combination with AI three-dimensional scoring and editing; s5, calling the multi-dimensional knowledge base to generate a professional knowledge packet; s6, generating an outline and performing three-dimensional five-score system auditing of knowledge point fullness and the like; s7, the LLM expands and writes the text and marks knowledge sources; s8, performing double-stage auditing; s9, anthropomorphic draft moistening; s10, intelligently illustrating the picture; s11, when the matching degree is smaller than a preset threshold value, automatically updating the knowledge base; and S12, integrating and outputting. According to the method, the health science popularization article is generated, the manual participation time is shortened to be within 30 minutes, the medical error rate is reduced to be below 5%, the daily output of a 10-person team is improved by 3-5 times, the annual human cost is reduced by 60% or above, and the large-scale and high-quality science popularization requirements are met.

Disclosed herein is a method for estimating a depth of an image. The method may include acquiring a first depth image based on left and right images collected from a stereo camera, estimating a second depth image based on any one of the left and right images, generating a depth estimation model based on first depth values within a depth range recognizable by the stereo camera in the first depth image and on second depth values of the second depth image matching the first depth values, and inputting third depth values of a range intended to be estimated in the second depth image to the depth estimation model, thereby estimating fourth depth values of the first depth image matching the third depth values.

The invention belongs to the technical field of computer vision, and particularly relates to a blind hole plate hole depth measuring method and system based on distributed cameras 2D and 3D, and the method comprises the steps: projecting laser stripes through a laser, and synchronously collecting a laser stripe image and a blind hole plate 2D gray scale image through the distributed cameras; according to pixel gray scale distribution in a preset window on the laser stripes, adaptively selecting a Gaussian fitting method or a gray scale centroid method to carry out sub-pixel-level center extraction, and generating 3D height data of the blind hole plate; based on the 3D height data and the positioning holes, blind hole center positioning is carried out through affine transformation and image matching; and combining the 3D height data and the 2D grayscale image, judging the measurability of the blind hole according to 2D grayscale features, selectively starting hole depth calculation based on the 3D height data, and comprehensively evaluating the quality of the blind hole. According to the invention, the problems of specular reflection and deep hole measurement are effectively solved, and the precision and efficiency of blind hole depth measurement are remarkably improved.

The invention discloses an image processing method and device, a storage medium, equipment and a program product, which are applied to a style migration scene. The method comprises the following steps: obtaining a cue word, an original image containing a target subject and a background image with a target background style, wherein the cue word is used for prompting to generate an image containing the target subject and with the target background style; splitting the original image to obtain a target main body area image, a mask image and an edge image; carrying out splicing processing on the target main body area graph, the mask graph and the edge graph to obtain a spliced image; according to the spliced image, the cue word and the background image, style migration processing is carried out, a target image matched with the cue word is obtained, and the target image comprises a target body and has a target background style. Through double guidance of the cue word and the background image, the original image is subjected to style migration to generate the high-quality target image, and the target main body in the target image and the target background style have a high fusion degree.

An unmanned aerial vehicle target detection tracking method based on intelligent image matching comprises the steps that an unmanned aerial vehicle collects images to make a data set, a model file is obtained through twin neural network training, the model file is loaded to detect a video stream, a current frame target position and detection frame information are obtained, whether the frame is a first frame is judged, and if yes, the target is detected; using a target position information result to initialize a kernel filtering algorithm, and constructing a measurement vector; if not, calculating correlation by using a kernel correlation filter, taking the highest response position as a prediction bit, and constructing a measurement vector; calculating a Kalman gain, and inputting the Kalman gain into a Kalman filter to obtain a predicted state estimated value = (); comparing y-axis deviations of the detector and the tracker, and setting a threshold value for judgment; according to the method, target tracking under resource limitation is realized through light-weight twin network detection in combination with kernel filtering and improved Kalman filtering, the problems of high target loss rate and low precision are solved, and the method is suitable for the fields of military reconnaissance, intelligent autonomous attack and the like.

The invention relates to the crossing field of mineralogy and geochemistry, in particular to a mineral and element relation analysis method for gold-bearing minerals based on image matching and data fusion and application of the mineral and element relation analysis method. Comprising the steps of ore sample processing, gold-bearing mineral image calibration through AMICS, trace element scanning through fixed-point EPMA and EDS, image-level matching and spatial data fusion and gold-bearing mineral-element relation analysis. According to the method, the AMICS mineral map and the EPMA element information are fused, relation analysis of trace elements in the gold ore and visual reconstruction of a mineral-element corresponding mechanism can be achieved, and effective data and image models are provided for microcosmic metallogenic information requirements in digital mine construction.

The invention discloses a text-to-three-dimensional object (3D) generation method based on dual color consistency matching, and aims to solve the problem of 3D asset quality reduction caused by multi-view color inconsistency in the prior art. According to the method, in each training iteration, a group of multi-view images are firstly rendered from a 3D model, and a high-fidelity forward view is specified as a color reference; subsequently, optimization is carried out through a double matching mechanism: 1, matching among images: aligning colors of other views to the reference view by using depth semantic features, thereby breaking circulation of error accumulation; 2, image internal matching: in a single image, color and geometric smoothness in the same semantic region are standardized by improving total variation loss; through the dual mechanism, accumulation and propagation of unnatural color distortion are effectively prevented, color inconsistency is reduced, and the visual fidelity and color consistency of the generated 3D model are remarkably improved.

The invention relates to the technical field of market analysis, in particular to a brand positioning optimization method based on big data, and the method comprises the steps: collecting multi-channel brand performance data, calculating a mean value, extracting a co-occurrence relation between emotion words and labels in comments, judging an emotion and image matching structure, and analyzing competing product label frequency difference and cognitive overlap. And identifying a user portrait and emotion offset, and merging expression gaps to generate an optimization suggestion scheme. According to the method, differentiated collection and unified coding are performed on multi-channel indexes, a brand expression structure capable of cross comparison is constructed, data definition is improved, comment words and brand labels form a co-occurrence matrix, an emotional expression structure under the value dimension is extracted, competitive product analysis locates a cognitive blank on the basis of label difference and coverage degree, and the competitive product analysis accuracy is improved. The emotion deviation value and the label acceptability are introduced as dynamic parameters by user division, and finally a fusion label missing and emotion space is output, expression reconstruction and crowd adaptation are completed, and dynamic linkage and positioning updating of a brand structure are realized.

The invention discloses a feature point detection method based on edge saliency and scale sensitivity. The method comprises the following steps of performing Gaussian smooth denoising on an input image; calculating the edge saliency of the pixel points through a Laplace operator; calculating a texture change degree based on the local gradient magnitude; generating a scale sensitivity weighted value through multi-scale analysis; candidate points are screened in combination with edge saliency, texture weighting and scale weighted values, and mismatching points are removed through non-maximum suppression; and finally, key points are enhanced and marked. According to the method, the edge structure, the texture information and the multi-scale features are fused, the robustness and accuracy of feature point detection are improved, and the method is particularly suitable for image matching and target recognition tasks in complex scenes.

The invention belongs to the technical field of unmanned aerial vehicle autonomous navigation, and discloses an unmanned aerial vehicle visual positioning navigation method based on a deep twin network and multi-modal fusion, and the method comprises the steps: obtaining a real-time aerial image of an unmanned aerial vehicle and a pre-stored satellite reference image, carrying out the self-adaptive scale zooming of a satellite image according to the output height of a barometer, and obtaining the real-time aerial image of the unmanned aerial vehicle; eliminating scale difference; the images after scale unification are input into a deep twin network for feature extraction and matching, and two-dimensional plane transformation parameters between the two images are obtained; and calculating the parameters as initial absolute geographic coordinates, adopting a tight coupling factor graph optimization frame, fusing a re-projection factor formed by the visual coordinates, an IMU pre-integration factor, a barometer height factor and a magnetometer course factor, and estimating and outputting six-degree-of-freedom pose information of the unmanned aerial vehicle through nonlinear optimization. According to the invention, the problems of poor matching robustness of different-source images and incomplete pose output in a GNSS denial environment are solved, and high-precision autonomous positioning navigation is realized.

The invention discloses a multi-platform SAR (Synthetic Aperture Radar) data and ground monitoring image fused earth surface deformation identification method, and belongs to the technical field of remote sensing monitoring and geological disaster early warning. According to the method, SAR images before and after deformation of a target area are obtained through a spaceborne, airborne or foundation SAR platform, and radar sight direction and azimuth direction deformation fields are extracted through preprocessing and offset tracking or interference processing; synchronously utilizing an unmanned aerial vehicle or ground photography equipment to obtain a high-resolution optical image, and calculating an east-west horizontal deformation field and a south-north horizontal deformation field through an image matching technology; after unifying the multi-source deformation fields to the same spatial resolution, constructing a deformation observation equation set based on a three-dimensional deformation decomposition model by combining radar geometric parameters, and inverting three-dimensional deformation in the vertical, east-west and north-south directions of the earth surface by adopting a weighted fusion algorithm (such as Helmert variance component estimation to determine a weight matrix); and finally, performing time sequence optimization on the deformation field by using Kalman filtering or a stress-strain model to generate a space-time continuous three-dimensional deformation product.

The invention relates to the technical field of image enhancement processing, in particular to an image enhancement method of a chest X-ray film for evaluating cardiac insufficiency. The method comprises the following steps: firstly, obtaining a shielding region of a normal X-ray film lung region, and projecting the shielding region to a target image to obtain a shielding part; further, according to similar characteristics of a skeleton line of the foreground connected domain and an edge line of a shielding part, combining an edge fuzzy characteristic and a local gray scale difference of the connected domain, obtaining a pulmonary blood vessel information amount of a pixel point and obtaining a pulmonary blood vessel region; further matching pixel points of the pulmonary vessel areas in the X-ray films which are inclined left and right according to the gray level change similarity in the vertical direction; and finally, carrying out translation superposition on the X-ray films which are inclined leftwards and rightwards on the basis of the overlapped pixel point group and the pulmonary blood vessel information amount, fitting a fused pulmonary blood vessel part to a shielding part, obtaining a pulmonary blood vessel X-ray image, matching translation superposition to repair a shielding region, overcoming oblique distortion and improving a pulmonary blood vessel display effect.

The invention discloses a high-precision satellite attitude correction method, system and equipment for a maximum initial positioning error, and belongs to the technical field of machine vision image matching. A proper reference region is selected to prepare a first reference image and a second reference image, and feature descriptors of the reference images are extracted offline; performing down-sampling on a remote sensing image acquired by a satellite remote sensing camera, and performing image matching on a real-time image after down-sampling and the first reference image; calculating the optimal window position of the ground calibration point of the reference area in the real-time image according to the matching result; cutting the real-time image according to the window position to obtain a high-resolution calibration point real-time image; performing image matching on the high-resolution calibration point real-time image and the second reference image; and calculating latitude and longitude coordinates of the center of the real-time image according to a matching result for satellite attitude correction. The position error of the target area is gradually reduced in a two-stage remote sensing image matching mode, the longitude and latitude coordinates of the center of the real-time image are accurately positioned, and high-precision satellite attitude correction is realized.

A device for training a contour extraction model that detects the contour of a semiconductor lithography pattern, according to one embodiment of the present invention, comprises: a memory having a contour extraction learning program stored therein; and a processor for executing the program stored in the memory, wherein the contour extraction learning program inputs an SEM image of a new pattern to a contour extraction unit so as to extract a first contour image, inputs the first contour image into a style transfer model so as to generate a virtual SEM image, and trains the contour extraction model on the basis of a learning data set in which the first contour image and the virtual SEM image are matched.

The invention discloses a pre-operation CT and intraoperative X-ray image registration method based on meta-learning, and relates to the field of operation image registration, and the method comprises the steps: preprocessing historical pre-operation CT data; constructing a feature extraction sub-model containing a double attention mechanism to mark the region of interest, a multi-branch joint learning multi-task sub-model, and a dynamic evolution sub-model for capturing a dynamic law by means of a neural memory ordinary differential equation; training a fine registration model of the network in a coarse registration stage and a fine registration stage; and subsequently, establishing a verification set by using untrained data, comparing errors and adjusting a parameter strategy by combining three indexes of attitude estimation error, image matching accuracy and model reasoning speed, and outputting a final model for CT (Computed Tomography) of a current patient and real X-ray registration in an operation. According to the method, CT and intraoperative real X-ray image registration can be carried out on different cases through one neural network model, the generalization performance and registration precision of the model are improved, and guarantee is provided for subsequent operation of surgical navigation.

The invention provides a pipeline leakage positioning method, device and equipment and a storage medium, and belongs to the field of intelligent detection.The pipeline leakage positioning method comprises the steps that concentration data and a pipeline image are synchronously obtained when a detection probe conducts detection along a to-be-detected pipeline; performing image matching on the pipeline image and a reference image to determine spatial position data of the detection probe in a three-dimensional space, the reference image being an image obtained by photographing the pipeline by using known spatial position data in the three-dimensional space; determining a concentration distribution model of the tracer gas on the to-be-detected pipeline based on the concentration data and the spatial position data at different moments; and determining one or more leakage point positions on the pipeline based on the area with the concentration value reaching a preset condition in the concentration distribution model. In the process, by establishing a concentration distribution model, the problem that in the prior art, due to gas diffusion, a pain point is difficult to accurately position is greatly solved, and the repair efficiency and the detection reliability are remarkably improved.

The invention provides a three-dimensional image modeling method and device based on plane data, and the method comprises the steps: carrying out the full-color / multispectral band fusion, bit depth adjustment, thin cloud removal, image enhancement, stripe removal and light and color uniformity of a satellite surveying and mapping image, and achieving the real color recovery of the image; performing splicing preprocessing on the plurality of small orthographic satellite surveying and mapping images after real color recovery, performing image registration on the to-be-registered image and a reference image, performing image geometric correction on the registered image, and performing image mosaic processing on the plurality of images after image geometric correction to obtain a single large-scene image; and three-dimensional image matching, building mask extraction, three-dimensional building modeling and three-dimensional building post-processing are carried out on a single large scene image to obtain a three-dimensional digital surface model, and ground three-dimensional real image modeling is completed. By applying the technical scheme of the invention, the technical problem that the existing two-dimensional surveying and mapping data is difficult to meet the increasing application requirements of three-dimensional scenes is solved.

The invention relates to the technical field of bar code identification, and discloses a rapid image matching incomplete bar code accurate identification method, which comprises the following steps: carrying out image contrast enhancement processing based on bar code candidate area response on an incomplete bar code image; positioning a bar code area of the incomplete bar code image after contrast enhancement, and detecting and marking an incomplete area in the bar code area; carrying out inference and reconstruction processing on the incomplete area based on a bar code standard coding rule and a statistical redundancy mode; and carrying out identification decoding and redundancy check on the reconstructed bar code image. According to the method, through a multi-step processing mode of image contrast enhancement, incomplete area positioning and labeling, reconstruction based on coding rules and redundancy modes and identification verification, rapid and accurate identification of the incomplete bar code is realized, and under the condition that partial information of the bar code is missing, fuzzy and interfered, the bar code structure can still be effectively recovered, and the identification accuracy of the bar code is improved. And the accuracy, the stability and the robustness of bar code identification are obviously improved.

The invention discloses a recording method and device and electronic equipment, and belongs to the technical field of electronic equipment. The method comprises the following steps: receiving a first input for triggering a target shooting mode; under the condition that the target camera shooting mode is a delay camera shooting mode or a slow shot mode, in response to the first input, recording an audio and an image respectively in the target camera shooting mode; based on the recorded audio, determining the audio matched with the recorded image; and packaging the recorded image and the audio matched with the image to obtain a target file.

The invention discloses an underwater target measurement method, system and device for salient region sparse matching in the technical field of underwater optical imaging and computer vision, and the method comprises the steps: calculating an intensity image and a linear polarization degree image based on a Stokes vector according to an obtained underwater binocular polarization image; according to the intensity image and the linear polarization degree image, performing image fusion by using a polarization image fusion network to obtain a fused image; according to the linear polarization degree image and the fusion image, performing target detection by using a saliency target detection network to obtain a target area mask; extracting feature points in the target area of the underwater binocular polarization image by using a SuperPoint network according to the target area mask, and generating a feature descriptor; according to the feature points and the feature descriptors, binocular image matching is carried out through a SuperGlue network, and matching point pairs are output; and calculating the distance from the target to the polarization binocular camera according to the matching point pairs and the parameters of the polarization binocular camera.

The invention discloses a molecular classification prediction method based on molecular biomarkers and images, and relates to the technical field of medical image processing. Comprising the following steps: constructing a molecule and image matching correlation analysis model; training by using the training data set; the channel number of the target preoperative medical image data is increased through the feature extraction convolution block; a space matching relation between an image and a molecular biomarker in the processed target preoperative medical image data is established through an image molecule fine granularity insight module, and a feature map after the space relation is established is obtained through a space molecule mapping map output module; adjusting the feature map into a corresponding molecular channel through a space molecular mapping map output module to obtain a space molecular mapping map; and pooling the spatial molecular mapping map through a global average pooling module, and activating and calculating a molecular category prediction probability through Sigmoid to obtain a molecular classification prediction result. According to the invention, the spatial relation between molecules and images is established.