106results about How to "Improve fusion effect" patented technology

The invention discloses a medical diagnosis intelligent decision-making system based on multi-modal data fusion, and the system comprises the steps: obtaining a medical image, biological features and electronic medical record data of a patient through a patient information collection unit, and extracting feature vectors from different modal data through a feature processing unit; the core of the system is a modal fusion engine, and the module combines a deep neural network and a Bayesian inference engine to realize deep learning and probability-dependent modeling of an association relationship between features; carrying out weight optimization on different modal features by adopting an attention mechanism through a self-adaptive feature optimizer, and generating optimized feature representation; the intelligent diagnosis analysis unit calculates the diagnosis probability based on the optimized features, performs auxiliary diagnosis in combination with a case knowledge base when the confidence coefficient is insufficient, and outputs a diagnosis report containing diagnosis categories, basis and suggestions; according to the method, multi-source heterogeneous data are fused, deep learning and probabilistic reasoning are combined, intelligent medical auxiliary diagnosis is achieved, and diagnosis accuracy and interpretability are improved.

The invention relates to the technical field of medical systems, in particular to an intelligent gastrointestinal postoperative intestinal obstruction diagnosis and intervention decision-making system and method, and the system comprises a data preprocessing module, a multi-modal data fusion module, a gastrointestinal function recovery model construction module, a self-adaptive progressive reinforcement learning optimization module AHRL and a diagnosis and intervention recommendation module. The self-adaptive progressive reinforcement learning optimization module optimizes an intervention strategy layer by layer through a five-layer progressive algorithm from the steps of dynamic adjustment of feature weights, recursive state propagation, layered accumulation optimization, symptom similarity matrix generation, progressive weight reverse feedback and the like, deep fusion of key symptoms by multi-modal data is achieved, and the accuracy of the key symptoms is improved. And the intervention scheme is updated in real time according to the patient state change. In the intervention decision recommendation module, the system generates personalized diagnosis suggestions based on the optimized strategy, carries out real-time risk assessment, and provides continuous and dynamic intervention support for the patient.

The invention belongs to the technical field of marine surveying and mapping, and discloses an unmanned aerial vehicle laser / shipborne multi-beam data land and water topographic mapping method, which comprises the following steps of: coarsely extracting a data overlapping region through a virtual grid, finely extracting the overlapping region by using an alpha-shape algorithm, and obtaining boundary points; removing edge triangles in an irregular triangulation network constructed by the unmanned aerial vehicle laser sounding data and the shipborne multi-beam sounding data; calculating a vertex normal vector, taking a vertex tangent plane of the multi-beam triangular irregular network as a registration element, carrying out weight distribution according to the distance from a point to the vertex, and carrying out coarse registration through a point-to-plane ICP registration algorithm; and constructing a precise registration model based on a point-surface distance, and eliminating the elevation difference of the overlapping region by minimizing the normal distance from the unmanned aerial vehicle laser sounding point to the multi-beam sounding triangular surface element. By adopting the point cloud fusion model for processing, the phenomenon that different source point cloud elevations in the overlapping area are inconsistent is effectively improved, and underwater topographic data fusion is realized.

The invention discloses a defect identification method and system based on an efficient multi-path feature fusion network, relates to the technical field of cloth defect detection, and solves the problems that deep-layer, shallow-layer, large-scale and small-scale feature information cannot be effectively fused in the prior art, and an existing model is high in calculation complexity and large in model parameter quantity. The method comprises the steps that a cloth flaw data set is constructed, an improved YOLOv5s model is constructed, and model training is carried out, and the improvement content comprises the steps that GFPN is fused into a Neck network of YOLOv5s, cross-scale connection layers between interval levels and cross-layer jump connection layers between the same scales are introduced into a backbone network, in the Neck network, C3Ghost lightweight feature fusion nodes are adopted, and the cross-scale jump connection layers between the interval levels and the cross-scale jump connection layers between the interval levels are fused into the backbone network. According to the method, the deep feature map is subjected to up-sampling by adopting a Dysample method, deep and shallow feature information and large-scale and small-scale feature information are effectively fused, the performance of a YOLOv5s model in multi-scale and multi-proportion target detection is improved, and meanwhile, calculation parameters and complexity of the model are reduced.

The invention discloses an intelligent drought risk management method and system based on disaster attribution analysis. The method comprises the steps of data acquisition, data enhancement, disaster attribution analysis, disaster dynamic risk assessment and intelligent drought risk management. The invention relates to the field of drought risk management, and adopts an intelligent drought risk management integrated method combining disaster attribution analysis and disaster dynamic risk assessment to perform drought disaster dynamic analysis from the angle of attribution, and perform multi-factor feature mining in combination with a plurality of feature optimization models. The complexity processing capability and the real-time precision of intelligent drought risk management are improved, and the fusion performance of multi-field data is also improved; performing disaster attribution analysis by adopting an improved support vector machine model combining an expert model and graph neural network feature optimization; and carrying out disaster dynamic risk assessment by adopting a gating circulation unit which combines the generative adversarial network and the graph convolution network and is deepened in feature.

The invention discloses a lightweight rice disease prediction method based on SCL-YOLOv8n, and belongs to the technical field of rice disease prediction. The method comprises the following steps: constructing a rice disease data set; performing enhancement, labeling and division processing on the rice disease data set; in a GSConv module of a Slim-Neck network structure, before Shuffle operation, lightweight convolution is introduced to be connected with an original path in parallel to form a GSLConv module; the method comprises the following steps: based on a CSP-RFA structure, dividing data, and distributing the data to two branches to form an RFCA-CSP module; a lightweight sharing enhanced detection head LSCSBD is adopted; an SCL-YOLOv8n model is constructed according to the model; taking the processed data set as the input of an improved model SCL-YOLOv8n, and training the improved model SCL-YOLOv8n by using the training set; in the aspect of rice disease prediction under the complex background, the parameter quantity and the calculation complexity are reduced, the precision is improved, and an effective reference is provided for rice disease prediction under the complex background.

The invention relates to the technical field of artificial intelligence multi-modal large models, and discloses a multi-modal human body large model training method and system considering multi-granularity semantic alignment, and the method comprises the steps: carrying out the prefix construction of a large-scale training set through improving a network structure, adding prefixes related to description to the data in the training set based on human body features; and training the multi-modal human body large model in batches, pairing each image and each text in each batch of training data, improving comparison loss through similarity calculation and semantic relaxation, and optimizing visual and text modal encoder parameters, so as to improve the accuracy of the multi-modal human body large model. According to the multi-modal human body large model training system considering multi-granularity semantic alignment, aiming at the multi-granularity semantic alignment problem encountered by a current multi-modal large model in human body cross-modal retrieval application, a general multi-modal large model is improved from two aspects of a network structure and a loss function; and the performance of cross-modal human body retrieval under multi-granularity semantics is enhanced.

The invention relates to an infrared and visible light image fusion method and system based on CRITIC weight. The method comprises the following steps: decomposing an input visible light image and an infrared image into a feature image and an original image based on a Gaussian filter; respectively calculating corresponding variable weight coefficients based on the visible light original image and the visible light characteristic image; fusing the visible light original image and the infrared original image based on the variable weight coefficient, and obtaining a visible light fusion image; fusing the visible feature map and the infrared feature map based on a variable weight coefficient, and obtaining an infrared fusion map; and fusing the visible light fusion image and the infrared fusion image to obtain a final fusion image. According to the method, when the light is too bright or too dark, the variable weight coefficient is matched with the weight which can best reflect the features of the fused image at present for fusion, so that the fusion effect of the fused image is improved.

The invention relates to the technical field of remote sensing image processing, in particular to a hyperspectral image and laser radar image fusion classification method and device. According to the method, a deep network model is constructed and comprises a multi-level modal assimilation encoder and a multi-level assimilation modal reconstruction decoder; and for each hyperspectral pixel in the normalized hyperspectral image, inputting a neighborhood block of the current hyperspectral pixel and a neighborhood block of a laser radar pixel with the same position as the hyperspectral pixel into a multi-level modal assimilation encoder, and performing multi-stage modal information fusion and feature extraction to obtain the hyperspectral image. Obtaining a remote sensing assimilation mode of the current hyperspectral pixel and the corresponding laser radar pixel; and inputting each hyperspectral pixel and the remote sensing assimilation mode of the corresponding laser radar pixel into a multi-level assimilation mode reconstruction decoder to obtain a pixel classification result after each hyperspectral pixel and the corresponding laser radar pixel are fused. According to the method, the hyperspectral image and the laser radar image can be effectively fused and classified.

The invention discloses a hyperspectral and panchromatic image fusion method based on feature enhancement channel shuffling, and the method comprises the steps: obtaining a hyperspectral image and a panchromatic image, carrying out the preprocessing of the hyperspectral image and the panchromatic image, and correspondingly obtaining a preprocessed hyperspectral image and a preprocessed panchromatic image; a hyperspectral image feature enhancement module based on matrix rank constraint performs feature extraction on the preprocessed hyperspectral image to obtain hyperspectral image features; a filtering-based panchromatic image feature extraction module performs feature extraction on the preprocessed panchromatic image to obtain panchromatic image features; splicing the hyperspectral image features and the panchromatic image features to obtain input features of an image feature fusion part; and carrying out feature fusion on the input features based on multi-scale convolution, extracting features of different scales, and carrying out deep fusion on the features of different scales based on a deep and shallow fusion module of channel shuffling to obtain a final fusion result. According to the invention, the image fusion effect is further improved.

The invention discloses a continuous fiber reinforced co-extrusion die, a forming line body and a fiber reinforced foaming floor, and relates to the technical field of foaming floors, the continuous fiber reinforced co-extrusion die comprises a die body, the front end face and the rear end face of the die body are a mixed extrusion face and a core material feeding face respectively, and one side face of the die body is provided with a shell material feeding port; a core material supply channel penetrating through the mixed extrusion face and the core material supply face is formed in the mold body, a shell material supply cavity surrounding the core material supply channel is formed in the mold body, two fiber channels are formed in the mold body, the two fiber channels are symmetrically arranged up and down, and the fiber channels are arc-shaped bent channels. The forming line body comprises a continuous fiber reinforced co-extrusion die, a core material extruder and a shell material extruder; a shaping module, a vacuum water cooling group, an open type water cooling group and a traction machine are sequentially arranged on the mixed extrusion surface of the mold body. According to the invention, the overall strength and performance of the composite material are optimized, and uniform distribution and orientation of reinforced fibers are ensured.

The invention relates to the field of waste gas treatment, in particular to a waste gas treatment device and method for adhesive tape production, the waste gas treatment device comprises a treatment tank, one side of the treatment tank is communicated with a gas inlet pipe, one side below the treatment tank is connected with a gas outlet pipe, and a liquid discharge pipe is mounted below the treatment tank. According to the waste gas treatment device, the connecting holes are annularly distributed and are also obliquely formed, so that waste gas is annularly conveyed into the treatment tank when entering the treatment tank, and the waste gas and conveyed air can be spirally downwards conveyed while the air conveying efficiency is improved through the rotation of the fan I, so that the waste gas and the conveyed air are uniformly sprayed when a spraying nozzle sprays; according to the device, the mixing effect with waste gas can be improved, particles obtained after spraying and filtering can fall on the connecting disc and are scraped through rotation of the connecting disc, and the spraying liquid falls and finally remains at the bottom of the treatment tank, so that waste gas particles left in the used spraying liquid are reduced, and the recycling and filtering cost of the waste spraying liquid is reduced; meanwhile, the waste gas filtering efficiency and quality can be improved.

The invention discloses a data fusion method and system based on deep learning, and relates to the technical field of deep learning, and the method comprises the steps: carrying out the classification of various types of sensor data in an unmanned system according to a set classification rule, and obtaining the classification data; performing preprocessing based on the basic attributes of the classified data to obtain preprocessed classified data; performing data-level fusion on the preprocessed classification data to obtain a plurality of classification fusion data; based on a deep recurrent neural network DRNN, in combination with a bidirectional long-short term memory unit Bi-LSTM and a sparse attention mechanism, constructing a data fusion model, and optimizing the data fusion model through model pruning to obtain an optimized data fusion model; and inputting the plurality of classified fusion data into the optimized data fusion model for secondary fusion to obtain a final fusion feature. The potential of multi-source information is fully mined, and the accuracy and effectiveness of decision making in a complex environment are improved.

The application discloses a kind of millimeter wave radar and the fusion method and system of visual identification target, the method comprises: obtaining the data information of multiple first target points and multiple second target points;The first target point represents the target object detected by millimeter wave radar sensor, and the second target point represents the target object detected by camera;With the first target point as the center, the absolute threshold set is expanded longitudinally and transversely, to form the first confidence interval;With the second target point as the center, the relative threshold is expanded longitudinally and transversely to form the second confidence interval;The size of the relative threshold is linearly related to the distance and speed of the second target point;KM algorithm is executed to fuse, as many as possible millimeter wave radar detected first target points and camera detected second target points are one-to-one associated as the same target object.The method can improve the target fusion effect and reduce the generation of virtual target.

The invention provides a crop phenotype data fusion method and device, electronic equipment and a storage medium, and relates to the technical field of electrical digital data processing, and the method comprises the steps: inputting target data corresponding to each piece of three-dimensional first to-be-fused crop phenotype data in each piece of to-be-fused crop phenotype data into a weight adjustment model, and obtaining a weight value corresponding to each piece of first to-be-fused crop phenotype data output by the weight adjustment model, and performing data fusion on each piece of to-be-fused crop phenotype data based on the weight value corresponding to each piece of first to-be-fused crop phenotype data and internal and external parameter data of a crop phenotype sensor for collecting each piece of to-be-fused crop phenotype data, the target data comprises environmental data when the crop phenotype data is collected, relative position information between the crop phenotype sensor and the crop, and a target parameter value of the crop phenotype data. According to the method, the dynamic weight is given to the crop phenotype data to be fused, so that the fusion effect, relevance, fusion efficiency and timeliness of data fusion can be improved.

The invention discloses a hyperspectral and multispectral image fusion method and system based on dictionary extension, and the method comprises the steps: obtaining a high-resolution multispectral image Ym and a low-resolution hyperspectral image Yh in an incomplete overlapping scene, dividing Ym into an overlapped high-resolution multispectral image # imgabs0 # and a non-overlapped high-resolution multispectral image # imgabs1 # to solve an overlapped region spectrum dictionary Ds and an overlapped region sparse coefficient As, and fusing and reconstructing an overlapped region high-resolution hyperspectral image Xs; the common spectrum # imgabs4 # of the # imgabs2 # and the # imgabs3 # is extracted, the common spectrum # imgabs5 # is used for carrying out spectrum dictionary expansion, an expanded non-overlapped region spectrum dictionary Dr and a non-overlapped sparse coefficient Ar are obtained, and a non-overlapped region high-resolution hyperspectral image Xr is reconstructed; and synthesizing Xs and Xr to obtain a high-resolution hyperspectral image X of a full coverage area of the high-resolution multispectral image Ym. According to the method, the sparse coefficient matrix of the non-overlapping area is solved by utilizing the spectral similarity and the spatial information of the multispectral image, and the image fusion of the full coverage area of the multispectral image in the incomplete overlapping scene is realized.

The invention provides a distributed photovoltaic generating capacity prediction method, which comprises the following steps of: acquiring original data, and acquiring historical photovoltaic output data, meteorological data and node labels of distributed photovoltaic nodes in a region; preprocessing the collected historical photovoltaic output data and meteorological data; constructing a prediction model based on depth separable convolution and an inter-group cross attention mechanism, and training the prediction model based on the preprocessed historical photovoltaic output data and meteorological data; historical photovoltaic output data, meteorological data and node labels of all distributed photovoltaic nodes in the region are input into the prediction model, and the total distributed photovoltaic power generation capacity in the future time region is predicted. According to the method, the prediction model based on the depth separable convolution and the inter-group cross attention mechanism is constructed, the fusion strategy is dynamically adjusted according to the importance of different modals and time sequence characteristics, and the accuracy and robustness of prediction of the photovoltaic generating capacity in the region are further improved.

The present application relates to computer vision and underwater detection technology, provide a kind of underwater small biological target detection method and system based on deep learning, the WLDA-YOLOv8 deep learning model constructed includes main network, neck network and detection head, main and neck embed light multi-scale module, detection head introduces dynamic detection head and adaptive sample selection strategy;Training stage uses improved WIoUv3s loss to carry out boundary box regression, the model is trained using underwater biological data set, finally realizes the accurate detection of small target in underwater image, enhances the capture and fusion ability to small target subtle features.

The utility model discloses a stirring kettle for preparing base number compound lubricating oil, which relates to the technical field of lubricating oil processing, and comprises a stirring mechanism, the stirring mechanism comprises a stirring kettle body, the top of the stirring kettle body is provided with a top cover, the top of the top cover is provided with a motor, the top of the stirring kettle body is provided with a limiting groove, and the limiting groove is communicated with the stirring kettle body. A plurality of positioning holes are formed in the bottom side of the inner wall of the limiting groove in an annular array mode, a bamboo hat type filter screen is arranged at the top end in the stirring kettle body, and a mounting sleeve is arranged on the outer surface of the bamboo hat type filter screen. When the rotating rod rotates, the two groups of stirring plates and stirring blades are synchronously driven to rotate, so that lubricating oil and alkaline additives are preliminarily mixed when the stirring blades and the stirring plates rotate, in addition, in the rotating process of the stirring plates, the mixed lubricating oil can pass through the through holes, the molecular adsorbability of the mixed oil is dispersed, and the lubricating oil and the alkaline additives are uniformly mixed. The fusion effect of the lubricating oil and the alkaline additive is further enhanced.

The utility model discloses an auxiliary heating structure of an injection mold, which comprises a heating rod arranged in a fixed mold core, the heating rod penetrates through a cavity of the fixed mold core, and a space between a bulge structure in the cavity and the inner side wall closest to the bulge structure in the cavity is set as a heating section needing to be heated. The projection of the heating section in the depth direction of the cavity of the fixed mold core is partially or completely overlapped with the projection of the heating rod in the depth direction of the cavity of the fixed mold core, the cavity penetrates through the top wall of the fixed mold core to form a discharging port for taking out a product, the heating rod is located on the side, away from the discharging port, of the fixed mold core, and the heating rod is detachably connected with the fixed mold core. According to the utility model, the overall temperature is high when the injection molding material flows to the fusion position, so that the fusion position is not easy to form a fusion line.

A high-strength refractory material and its preparation method and application, belonging to the technical field of refractory materials, the raw materials of the refractory material include fused corundum, high-alumina bauxite clinker, metal silicon powder, anti-cracking agent, titanium aluminum carbide and titanium carbide, etc., wherein the anti-cracking agent is prepared by modifying boron phenolic resin, mesophase asphalt, artificial graphite powder and silicon carbide fiber. The refractory material of the present invention adopts metal silicon powder to improve wear resistance and oxidation resistance, titanium aluminum carbide to improve oxidation resistance, titanium carbide to improve strength, and anti-cracking agent after surface modification and film formation to greatly improve thermal shock resistance and reduce thermal expansion rate and slag erosion rate. In addition, adding titanium aluminum carbide can improve strength and thermal shock resistance to a certain extent, has good oxidation resistance, and improves erosion resistance. The refractory material can be used for high carbon steel and low carbon steel, and is particularly suitable for casting high oxygen steel, calcium treated steel and high manganese steel.

The invention discloses a PHSM-YOLO-based multi-target wearing identification method. The PHSM-YOLO-based multi-target wearing identification method comprises the following steps: constructing a PHSM-YOLO target detection model used for identifying whether a constructor correctly wears a safety appliance; collecting a target image containing constructors in the construction environment; and inputting the target image into a PHSM-YOLO target detection model, and outputting a wearing identification result of the safety wearing tool of each construction worker in the target image by the PHSM-YOLO target detection model. According to the multi-target wearing identification method, the wearing of the safety wearing tool can be identified on the target image containing the construction personnel through the constructed PHSM-YOLO target detection model, so that the personal safety of the construction personnel is guaranteed.

The invention relates to the field of artificial intelligence, and provides an agricultural product risk control method and system based on production and fusion combination, and the method comprises the steps: carrying out the block chain storage of financial risk assessment data, agricultural product risk assessment data and real-time agricultural data, and obtaining public production and fusion data; utilizing a pre-trained agricultural product risk assessment model to perform risk assessment operation based on yield, market and natural disasters on agricultural product risk assessment data in the public production and fusion data to obtain an agricultural product risk assessment result; utilizing a pre-trained agricultural product risk control model to perform risk control policy prediction operation on the agricultural product risk assessment result and financial risk assessment data in the public production and fusion data to obtain a policy scheme; and according to the insurance policy scheme, constructing an insurance relationship between the target agricultural area and the financial institution. According to the method, the integration and transparency of agricultural data and financial data are improved, and a risk management and control scheme for benefits of an agricultural party and a financial party is constructed.

The application discloses a smart community police management system and method based on Internet of Things security, and belongs to the technical field of smart communities, and solves the problem that existing system data processing cannot utilize edge computing nodes for low-delay real-time analysis, and cannot meet timely early warning requirements, the method comprising preprocessing of multi-source heterogeneous community monitoring information, analysis and processing of the community monitoring information based on edge computing, and generation of a risk propagation path in combination with front-end analysis results and heterogeneity characteristics; in the application, analysis and processing of the community monitoring information based on edge computing avoids the delay of centralized processing, can quickly trigger early warning in the early stage of a risk event, thereby gaining golden disposal time for risk event disposal, and a risk analysis model realizes a leap from single-node anomaly detection to composite risk quantification through multi-dimensional feature fusion and dynamic space-time reasoning, and realizes multi-dimensional feature fusion and dynamic reasoning capability of the risk event.

The present invention discloses a 3D printed bionic anti-dislocation movable artificial cervical vertebra and intervertebral connection complex, including: an upper end plate, a lower end plate, a bionic vertebral body and an anti-dislocation structure, the bionic vertebral body includes a vertebral body body hollowed out in the middle, an upper acetabulum is arranged on the upper part of the vertebral body body, and a lower acetabulum is arranged on the lower part; the upper acetabulum is connected to the upper end plate through an upper end plate joint structure, and the lower part is connected to the lower end plate through a lower joint structure; the present invention has sufficient support structure, and can provide stable support after anterior cervical vertebral resection; the present invention replaces the movement function of the normal cervical vertebra through the movement of its own ball joint, can retain the original range of motion of the cervical vertebra, and prevent the occurrence of adjacent segment diseases of the cervical vertebra caused by fusion surgery, and the present invention has a matching anti-dislocation structure, which is placed in the occurrence of dislocation during joint activity. The implantation operation of the present invention is less difficult, less traumatic, and easy to promote.

This application relates to the field of children's playground technology and discloses a children's activity area based on a suspended walkway. The activity area includes a ground surface, a suspended walkway on the ground surface, and a movable house on the ground surface, located on one side of the suspended walkway. The movable house includes a single-story building with a second-story building on top. This children's activity area based on a suspended walkway, through the movable house, the first climbing frame inside the movable house, and the slide on one side of the suspended walkway, fully utilizes vertical space. The three-dimensional design of the activity area based on the suspended walkway allows for more diverse playground facilities within a limited ground area. The activity area features various playground facilities, providing children with more challenging and fun play experiences. The suspended walkway within the park enhances the interaction and integration with the natural environment.

The invention relates to the technical field of deep mineral detection, and discloses a deep mineral rapid detection method based on multiband spectrum fusion, which comprises the following steps: step 1, building a detection system; 2, carrying the detection system on a detection platform, and synchronously collecting three types of spectral data and course angle, pitch angle, roll angle and speed parameters; step 3, data preprocessing; and step 4, a cross-modal fusion module realizes alignment of three types of data through cross-modal feature matching, and then weighted fusion is carried out. According to the deep mineral rapid detection method based on multi-band spectrum fusion, an attitude sensing module collects a course angle, a pitch angle, a roll angle and speed parameters in real time, a data preprocessing module cooperates with a cross-modal fusion module, and an attitude correction matrix is constructed based on Euler angle transformation to carry out geometric correction on multi-band spectrum data; the detection platform advances according to a preset height and speed working condition, a high-precision inertial navigation system is not needed, and the cost and the detection precision are both considered.

The invention relates to a BERT and LSTM fused double-tower recommendation method, which combines the advantages of BERT and LSTM, effectively fuses user comments and article information, processes the user comments through a BERT model, extracts deep features from the user comments, effectively captures user interest changes, obtains article features through an embedded layer, and then fuses the features in the LSTM, so that the user comments and the article information can be recommended. Capturing the time sequence dependence of the user behavior, and converting the prediction score; according to the recommendation method, the score prediction accuracy is improved, the information fusion capability is enhanced, a more accurate, more transparent and more personalized recommendation process is provided for the user, and the user satisfaction and the user stickiness are improved.

The invention discloses an image-text data rapid processing method and system, and relates to the technical field of data processing.The method comprises the steps that image-text data uploaded by a user are received, and the image-text data are stored through a double-cache queue; classifying and marking the image-text data in the double-cache queue, separating the text data from the image data, and establishing a feature index for the text data; performing feature extraction on the image data to generate an image feature vector, and associating the image feature vector with a corresponding feature index; synchronously processing the feature index and the image feature vector to generate fused feature data; and storing the fused feature data into a distributed database, and organizing and storing the fused feature data according to a preset data structure. According to the method, the image-text data processing efficiency and accuracy are remarkably improved by adopting the technologies of storing the image-text data by adopting the double cache queues, extracting features by adopting the deep learning model, performing parallel computing processing, performing feature fusion and the like.

The invention discloses a multi-modal recommendation method for cross-modal semantic alignment and graph relation enhancement. The multi-modal recommendation method comprises the following steps: S1, initializing a model and setting parameters; s2, constructing an article heterogeneous relation graph fusing semantic information and behavior information; s3, executing cross-modal local-global multi-granularity semantic alignment; and S4, performing model optimization and preference prediction based on the aligned multi-modal representation. Through the integrated design of heterogeneous relation fusion and cross-level alignment, the problem that in an existing method, article relation mining is one-sided and cross-modal feature alignment is insufficient is effectively solved, and recommendation accuracy and robustness are remarkably improved.