355 results about "Image evaluation" patented technology

The invention belongs to the technical field of image processing, and discloses a grape leaf AI grading processing method and system based on computer vision. Comprising the following steps: acquiring image data of grape leaves, and performing variety label marking and image optimization to obtain a clear leaf image data set; sequentially performing feature extraction and feature fusion on the clear leaf image data set to obtain fused disease condition features and fused growth condition features; constructing a leaf evaluation framework capable of being adjusted in real time based on the fused disease condition features and the fused growth condition features; performing image evaluation on the clear leaf image data set by using a leaf evaluation framework to obtain a disease degree evaluation index and a growth condition evaluation index, and performing disease degree grading and growth condition grading on the clear leaf image data set based on the disease degree evaluation index and the growth condition evaluation index; obtaining a grape leaf grading result; accurate grading of the grape leaves is achieved, and the industrial production efficiency is improved.

Disclosed herein are system, apparatus, device, method and / or computer program product embodiments for determining, in a remote deposit system, whether a deposit attempt is illegitimate (e.g. fraudulent). Whether the deposit attempt is illegitimate may be assessed based on one or more of the following processes: comparing location data to a location parameter determined from past deposits, comparing an image capture location with a deposit location, and analyzing image-of-image characteristics obtained through image processing to identify whether an image associated with the deposit attempt is an image of an image. In some embodiments, a remote deposit status related to acceptance of the deposit attempt may be provided in real-time

The invention provides a cloud-edge collaborative intelligent tool magazine tool life prediction method and system, and relates to the field of fault prediction and health management.The method comprises the steps that in a cloud data center, a current tool health index is obtained according to workpiece machining quality information and tool surface image evaluation; according to the current cutter health index, a cutter damage monitoring and early warning mechanism is configured and sent to an edge processing unit; according to the adaptive monitoring frequency, a sensor is controlled to conduct data monitoring, multi-source monitoring data are judged according to the adaptive early warning indexes, if not triggered, dynamic updating of a tool damage monitoring and early warning mechanism is conducted in an iteration mode, and if triggered, tool damage early warning is conducted. The invention aims to solve the technical problem that the accuracy and the reliability of tool damage early warning are insufficient due to the fact that a traditional early warning mechanism is difficult to adapt to the dynamic change of a tool machining state, and can improve the adaptation degree of a monitoring early warning mechanism and the tool machining state and remarkably improve the accuracy and the reliability of early warning by introducing a dynamic monitoring early warning mechanism.

The invention provides a self-adaptive accurate rust removal method and system based on unmanned aerial vehicle visual feedback, and relates to the technical field of large metal structure surface maintenance and anticorrosion treatment.The rust removal method comprises the steps that a global path is planned, and a global flight waypoint sequence is generated; a local image of the current to-be-derusted area is collected; the corrosion characteristics of the current to-be-derusted area are analyzed, and initial derusting parameters and initial derusting time are set; after each derusting operation is finished, a frame of intermediate result image is collected; based on the difference between the intermediate result image and the historical image, performing quantitative evaluation on the effect of the previous rust removal operation, and performing feature analysis on the residual rust region; self-adaptively adjusting and generating rust removal parameters of the next stage; and the steps of derusting operation, image collection, evaluation and parameter adjustment are repeatedly executed, and after the termination condition is met, the unmanned aerial vehicle is switched to the next to-be-derusted area.

The invention provides an image processing method, a training method of an image evaluation model and electronic equipment, and relates to the field of computer technology and artificial intelligence. The method comprises the steps that image data and inquiry text data are acquired, the image data are generated by an image generation model, and the inquiry text data are used for describing the requirement for quality evaluation of the image data by adopting a natural language; respectively inputting the image data and the inquiry text data into a plurality of image evaluation models, and respectively carrying out quality evaluation on the image data from corresponding evaluation dimensions by utilizing the plurality of image evaluation models to obtain a plurality of evaluation results; and integrating the plurality of evaluation results to obtain a quality evaluation result of the image data. According to the method and the device, the technical problems of relatively low accuracy and meticulousness of image quality evaluation in related technologies are solved.

The invention discloses a photovoltaic panel surface pollution degree image evaluation system. The system comprises an image acquisition module, an image processing module, a multi-task evaluation module and a pollution degree determination module. Multi-spectral image data of the surface of a photovoltaic panel are collected, after radiation calibration and atmospheric correction processing, spectral reflectivity features and spatial texture features are extracted and fused, a multi-task deep learning model is input, a pollution type classification result and a power generation efficiency loss weight of each pixel region are output, and a pollution type classification result of each pixel region is obtained. And finally, generating a pollution type spatial distribution diagram, calculating the overall power generation efficiency loss percentage, and determining the pollution degree grade. According to the method, the problem that the mixed pollution type cannot be distinguished and the differentiation influence cannot be evaluated in the prior art is solved, accurate quantitative evaluation of the pollution degree is realized, and a reliable basis is provided for fine operation and maintenance of a photovoltaic power station.

A method for improved surroundings detection utilizing an optical sensor of a vehicle. The method includes detecting the surroundings of the vehicle and compiling a first surroundings model based on sensor data of the surroundings detecting sensor. The method also includes determining regions having low quality in the surroundings model by evaluating optical sensor data with an image evaluation device. The detectability of the regions having low quality are improved by taking a selected measure. The method also includes detecting the surroundings of the vehicle again with the optical sensor and comparing the sensor data of the optical sensor after taking the measure with the sensor data of the optical sensor prior to taking the measure. The sensor data following the measure is fused with the already existing sensor data in order to compile a second surroundings model.

The invention relates to the technical field of image recognition, in particular to a medical image intelligent evaluation system based on image recognition. The system comprises an image registration module, an image segmentation module, a preliminary fusion module, an image evaluation module, an optimization feedback module and an image output module. According to the method, the CT image and the MRI image are subjected to image registration, spatial alignment is ensured, then the region of interest is segmented and fused, namely, the skeleton contour in the CT image is superposed on the MRI image, and due to the fact that motion artifacts generated by movement of a patient in the scanning process possibly exist in the original CT image, the skeleton contour in the CT image is fused with the motion artifacts in the MRI image. If the skeleton contour does not exist in the MRI image, the overlapping degree and the blank degree of the skeleton contour and the anatomical structure edge of the MRI image are analyzed, and an optimized registration parameter or segmentation parameter is fed back, so that when the segmentation network is trained, the segmentation precision under the conditions of artifacts and low contrast is improved, spectrum and texture information of the two images is reserved to the maximum extent, and the fusion effect is guaranteed.

The invention relates to the technical field of image processing, discloses an image meaning analysis scene consistency evaluation system based on a visual model, and aims to solve the problem of insufficient recognition stability in complex environments such as illumination variation, angle deviation and local shielding in the prior art. The system comprises an image input module used for receiving and preprocessing a plurality of images; the visual large model analysis module is used for carrying out multi-dimensional semantic feature extraction on the preprocessed image; the structured description generation module is used for converting the semantic features into structured text description in a unified format; the scene consistency evaluation module is used for performing logic consistency analysis on the structured text descriptions of the plurality of images; and the result output module is used for generating and outputting a final consistency evaluation report. According to the technical scheme, the method can effectively improve the recognition stability of the system in a complex environment, achieves the multi-dimensional semantic understanding of the image content, and remarkably reduces the misjudgment rate of multi-view image evaluation.

The invention provides an image self-correction intelligent scanning platform and a sorting optimization method, and relates to the technical field of image processing, the platform comprises a multi-angle image acquisition module, a bar code restoration and identification module, an image evaluation module and an image classification module. The scanning device is used for obtaining image information of an article to be recognized and comprises a bar code image and an article sorting module. An AI repair algorithm with a fuzzy or damaged bar code is built in the bar code repair and recognition module, the fuzzy or damaged bar code image obtained through scanning can be repaired, the problems of adhesive tape coverage and recognition of stained and damaged bar codes are solved, and the code scanning recognition rate is increased. In the scanning process, goods types can be automatically marked, manual labeling classification is replaced, the labor cost of a sorting center is reduced, efficient and accurate logistics sorting optimization is achieved, and the problem of the sorting efficiency bottleneck caused by bar code failure in logistics storage is solved.

The invention discloses a CBCT system bed board artifact correction method, device and equipment and a storage medium, and relates to the technical field of computed tomography. The method comprises the following steps: scanning first object projection data of an object by a bed board; performing three-dimensional reconstruction on the first object projection data to obtain a first CBCT image; constructing a three-dimensional digital image of the bed board according to the first CBCT image; forward projecting the three-dimensional digital image of the bed board to obtain an independent bed board projection; according to the independent bed board projection, separating the contribution of the bed board from the first object projection data to obtain second object projection data; performing three-dimensional reconstruction on the second object projection data to obtain a second CBCT image; and carrying out image evaluation index judgment until the evaluation index is greater than or equal to a preset threshold value, and obtaining an image without the bed board artifact. According to the method, the three-dimensional digital image of the bed board is constructed through one-time scanning to achieve artifact removal, related problems of secondary scanning are avoided, low-complexity forward projection is adopted to separate bed board contribution, and higher robustness and convenience are achieved.

The invention discloses a cutting control method and system for femtosecond laser machining equipment, and relates to the technical field of laser precision machining and intelligent control. The method comprises the steps of collecting a cut image sequence in real time, extracting and normalizing feature parameters representing cutting quality, and forming a standardized feature vector; performing quality evaluation based on the feature vector and generating a score; identifying anomalies and predicting a quality trend by threshold comparison, generating a parameter optimization request when deviation or predicted degradation is detected; inputting the feature vectors, the process parameters and the quality scores into a pre-trained AI model, outputting process parameter correction and adjusting equipment parameters; and after the equipment is updated, acquiring the image again to evaluate the quality so as to decide whether to maintain the adjustment result. Self-adaptive control and dynamic optimization of femtosecond laser cutting are achieved, parameter matching and path adjustment can be automatically completed under the conditions of multiple materials and multiple procedures, and manual intervention and debugging are reduced.

The invention discloses an image transmission method, device and equipment of a remote operation system and a medium, the method is applied to a local operation end, the remote operation system comprises the local operation end and a remote operation end, and the method comprises the following steps: acquiring an initial ultrasonic image corresponding to a target object at the local operation end, performing feature extraction on the initial ultrasonic image to obtain a target ultrasonic image; determining an ultrasonic sub-image corresponding to the target ultrasonic image based on the knowledge graph, and determining an image value of the ultrasonic sub-image according to the ultrasonic image evaluation dimension; and transmitting the ultrasonic sub-image to a remote operation end according to the image value. Based on the technical scheme, the image is divided into the sub-images of different areas, the image values of the sub-images are determined, and the transmission methods corresponding to the sub-images are determined according to the image values, so that the preferential transmission of the high-value image is ensured, the overall bandwidth occupation and the system load are reduced, and the transmission efficiency is improved. And a key technical support is provided for the accuracy and safety of a remote operation.

The invention discloses a model generation image evaluation method and device and a storage medium, and relates to the technical field of image processing. According to the method, the image generated based on the text cue word is obtained, wherein the image comprises the image element corresponding to the text cue word; filtering the image to obtain a filtering response value of each image pixel in the image, and generating a visual attention map of the image based on the filtering response values, the visual attention map representing frequency domain energy distribution of the image; and determining a visual saliency value of the image element according to the visual attention map, and determining a layout score of the image according to the visual saliency value of the image element. According to the method, spatial relations such as distances, alignment modes and hierarchical structures among elements are quantitatively evaluated through visual saliency values, and the defect that a previous model cannot accurately evaluate and adjust image layout is overcome.

The embodiment of the invention provides a multi-view consistent image generation method and device based on geometric information guidance, a storage medium and electronic equipment, and relates to the technical field of image generation, and the method comprises the steps: obtaining a multi-view geometric image; extracting a cross-modal condition feature, an image condition feature and a geometric condition feature based on the multi-view geometric image to serve as condition input of a diffusion model; constructing and initializing a multi-view diffusion generation model containing a decoupling geometric attention enhancement mechanism based on the condition input; optimizing parameters of the multi-view diffusion generation model by adopting a geometric information intensity adjustment mechanism; and generating a multi-view image based on the optimized multi-view diffusion generation model, and performing image evaluation based on perception similarity, distribution difference and diversity indexes. According to the method, the problems of structure mismatching, shielding disorder, detail drift and the like in the existing single-view-to-multi-view-angle generation are solved, and the multi-view-angle consistency and the detail fidelity are improved.

The invention belongs to the technical field of optical module assembly, and particularly relates to an AR equipment intelligent active alignment method based on artificial intelligence, and the method comprises the steps: enabling a system to obtain the real-time motion posture data and real-time image definition data of a six-axis robot in an alignment period in real time, the two sequences are marked as a motion posture sequence and an image definition sequence after being aligned and denoised; extracting an initial evaluation value from the image definition sequence; acquiring an image evaluation value of the detection position; constructing a weighted instantaneous gradient index; obtaining an out-of-focus curve correction factor; outputting a nonlinear correction distance factor representing the residual physical movement amount; obtaining a self-adaptive confidence weakening factor; calculating coordinates of a dynamic fine adjustment starting point; and the six-axis robot is driven to move to the dynamic fine adjustment starting point coordinates, and final optimization is executed. The technical problem that in the prior art, the fixed fine adjustment starting point is not matched with the batch tolerance of optical components, and consequently the optimal imaging point dynamically shifts is effectively solved.

The invention discloses an endoscope pneumoperitoneum pressure system based on AI image evaluation and an intelligent adjusting method, and the method comprises the steps: 1, obtaining an original image of the abdominal cavity of a patient through CT equipment, carrying out the standardization processing of the original image, generating a standardized CT image data set, and constructing a three-dimensional dynamic model of the abdominal cavity of the patient based on an AI algorithm; in the second stage, quantitative analysis of subcutaneous fat thickness, visceral organ volume and operable space volume is carried out on the three-dimensional dynamic model by adopting an AI segmentation algorithm, and a personalized pneumoperitoneum pressure target value Ptarget and an adjustment threshold range are generated through a machine learning model in combination with basic illness state data of the patient; in the third stage, working data are collected in real time through a built-in sensor of the endoscope system, and the working data are input into the reinforcement learning model to dynamically calculate a pressure adjusting instruction delta P; through deep fusion of the AI technology and system closed-loop control, the intelligent level of the endoscope system is remarkably optimized, good hardware support is provided, and the function of optimizing the operation environment and effect is achieved.

A method (200) for focusing an industrial camera (102) fixed on a mobile robot (101) includes: capturing a first target image of a target object (105) with a first step length mobile robot (101) (S201); determining a first ROI image (S202); evaluating the first ROI image to generate a first plurality of sharpness values ​​(S203); determining a single-peak search direction based on the first plurality of sharpness values ​​(S204); capturing a second target image of the target object (105) with a second step length mobile robot (101) according to the single-peak search direction (S205); determining a second ROI image (S206); evaluating the second ROI image to generate a second plurality of sharpness values ​​(S207); and estimating the sharpest focus position based on a portion of the first plurality of sharpness values ​​and a portion of the second plurality of sharpness values ​​(S208). It enables fast and accurate autofocus of industrial cameras (102) without human intervention, effectively improving on-site work efficiency. It is also flexible and applicable to various industrial cameras (102) and lenses, thereby effectively reducing on-site costs.

The present invention relates to a self-propelled forage harvester (1) comprising a height-adjustable header (4) for receiving crop (2), working units (20) for processing the received crop (2), an unloading device (15) for discharging the processed crop (2), a camera system (16) for capturing images (41) of a crop flow (21) passing through the forage harvester (1), an image evaluation device (27) for evaluating the images (41), and a driver assistance system (17) for controlling the header (4), the working units (20), and the unloading device (15), wherein the driver assistance system (17) has a memory (40) for storing data and a computing device (39) for processing the data stored in the memory (40), wherein the image evaluation device (27) together with the driver assistance system (17),The front attachment (4) and the working units (20) form an automatic setting system, in that the image evaluation device (27) is configured to continuously analyze the images (41) of the crop flow (21) for the proportion of, in particular inorganic, impurities contained in the crop flow (21) by means of a machine learning algorithm and to transmit a derived degree of contamination (DC) to the driver assistance system (17), which autonomously and continuously adapts a setting of the front attachment (4) and / or at least one of the working units (20) depending on the degree of contamination (DC).

The invention relates to the technical field of agricultural automatic intellectualization, in particular to a citrus picking and obstacle detection method and system, and the method comprises the following steps: building a detection image evaluation model, and carrying out the evaluation and analysis of original citrus detection image data, and obtaining citrus target detection image information; determining analysis characteristic parameters of the citrus detection image according to citrus picking and obstacle detection requirements, and further constructing a citrus detection image multi-level analysis system; setting calculation prediction models of different feature analysis layers in the orange detection image multi-level analysis system according to the analysis feature parameters, and analyzing the target detection image information by combining the multi-level analysis system and the calculation prediction models to obtain a multi-level analysis result of the orange detection image; on the basis of a multi-level analysis result and citrus target detection image information, citrus fruit types and obstacle conditions are analyzed, and different citrus picking schemes are formulated. According to the invention, automation of citrus picking and obstacle detection is realized, and agricultural intelligent development is facilitated.

The invention relates to the technical field of image data processing, and discloses an AIGC image rapid generation method and system, and the method comprises the steps: obtaining an original image source, dividing the original image source into image blocks, constructing an image block set based on a multilayer pyramid rule, extracting the frequency domain sparseness, the space gradient density, the wavelet energy distribution and other texture features, and carrying out the rapid generation of an AIGC image. Generating, sorting and compressing a texture index map; analyzing the semantic structure of the image block, constructing a semantic background image, calculating a semantic residual error, generating a semantic residual error mapping image, and inputting the semantic residual error mapping image into an AIGC model to generate the content of the image block; fusing the plurality of image blocks to generate a preliminary image, evaluating image quality and extracting texture, color and structure indexes; and optimizing a generation path based on an evaluation result, realizing iterative updating of image reconstruction, and outputting a final image. According to the method, by constructing a dual guide mechanism of the compressed texture index map and the semantic residual mapping map, the control of the AIGC generated image on the structure and semantic level is improved.

The invention belongs to the field of medical image processing, and provides a thrombus composition visual navigation method based on a CT image, which comprises the following steps: step 1, evaluating the CT image, and sketching thrombus according to the CT image of a stroke patient; 2, feature selection and model construction, wherein thrombus is divided into different areas; 3, performing histological evaluation, and associating different regions with tissue components; 4, outputting to a user: extracting the characteristics of each region to predict the prognosis of the patient; and fusing the CTA image and the DSA image, performing thrombus extraction navigation, and automatically and visually presenting different regions of thrombus components. The invention provides a visualization method for helping doctors to evaluate thrombus components of ischemic stroke patients.

The invention relates to the technical field of intelligent robot vision, and discloses a light environment image evaluation system based on a bionic robot, which comprises an image acquisition module, a light environment feature extraction module, a self-adaptive image processing pipeline, a quality evaluation feedback loop and a low-delay hardware acceleration architecture, the image acquisition module consists of a visible light camera, a local brightness sensor and a spectral analysis sensor, and adopts a partition photosensitive design; the light environment feature extraction module is used for establishing a three-dimensional feature space of illumination intensity-color temperature-time; the adaptive image processing pipeline is used for dynamically selecting an optimal image processing mode according to light environment characteristics and reconstructing the image; the quality evaluation feedback loop is used for evaluating the quality of the processed image and dynamically adjusting evaluation parameters; the low-latency hardware acceleration architecture is used for accelerating the data processing speed of a specific function. The adaptive capacity of the system in a dynamic light environment is improved, the image quality stability is improved, and effective processing of a high dynamic range scene is realized.

The invention discloses a self-adaptive contrast enhancement and noise suppression method and system for a pavement crack image. The method comprises the following steps: performing pavement image region segmentation on a to-be-processed pavement crack original image to obtain a pavement region; performing pavement area texture characteristic analysis on the segmented pavement area, performing contrast enhancement processing on the pavement image to obtain a contrast enhanced image, performing multi-stage noise suppression processing on the obtained contrast enhanced image to obtain a noise suppression image, and performing pavement crack enhancement evaluation based on the noise suppression image to obtain a pavement crack enhancement result. According to the method, a corresponding crack image evaluation result is obtained, a contrast enhancement coefficient graph is constructed by analyzing texture characteristics and local contrast distribution of a pavement region, and differential processing of different regions is realized. Compared with a traditional global enhancement method, the method has the advantages that higher enhancement can be provided for a low-contrast area, the enhancement strength can be properly weakened for a high-contrast area, and therefore noise is prevented from being excessively enhanced while the crack visibility is improved.

The present disclosure relates to an image assessment method and apparatus, and a device, a storage medium and a program product. The method comprises: acquiring an image to be assessed; and inputting said image to be assessed into an image assessment model, so as to obtain a quality assessment result corresponding to said image to be assessed, wherein the image assessment model comprises: a multilevel transformation network, a fusion network and a fully connected layer; the multilevel transformation network is used for processing said image to be assessed to obtain image features, which are output by each layer of transformation network; the fusion network is used for fusing the image features, which are output by the each layer of transformation network, so as to obtain a fused image feature; and the fully connected layer is used for processing the fused image feature to obtain the quality assessment result.

The invention relates to the technical field of ship maintenance, and provides a method and system for analyzing the cleaning effect of a ship decontamination robot, and the method comprises the steps: employing a plurality of independently controlled illumination units, and activating all illumination units according to a preset time sequence within a preset time; when each lighting unit is activated, synchronously acquiring a frame of image corresponding to the surface of the ship body to obtain a plurality of frames of images, and packaging into an activated acquired image corresponding to the surface of the ship body; performing optical interference area identification on the activated acquisition image to obtain an optical interference area identification result; based on an optical interference area identification result, performing information fusion on a plurality of frames of images in the activated acquisition image to reconstruct an optical interference-free image corresponding to the ship body indication, and obtaining an interference-free surface image; based on the non-interference surface image, the cleaning quality of the ship surface is evaluated, and a ship decontamination robot cleaning effect analysis result is obtained. The automatic cleaning device has the effect of improving the efficiency and thoroughness of automatic cleaning operation.

The present application relates to the target ISAR image evaluation method based on signal processing and image change, belongs to ISAR image processing field. In view of the ISAR image imaging fuzzy under the condition of target posture change on ISAR image, feature extraction is difficult, leading to the problem of target threat assessment difficulty, a kind of target ISAR image evaluation method based on signal processing and image change is presented. Including: based on the rearrangement and elimination method of micro-motion to realize the de-micro-motion imaging of space micro-motion target;The ISAR image is carried out morphological processing open operation, using Canny operator carries out edge detection processing, carries out Hough transformation and carries out straight line detection, obtains the size and boundary of target;The echo model of compound micro-motion target of combined micro-motion is established, and the micro-motion period is estimated using the method based on the phase difference signal aggregation degree, realizes the de-micro-motion imaging of space target and micro-motion target feature extraction and solves the problem of target threat assessment difficulty.

The present invention relates to the field of crop prediction and management technology, and discloses an intelligent agricultural Internet of Things (IoT) management system and method. The system includes a pest monitoring module that uses an image acquisition device to capture images of crop leaves within a greenhouse, assesses the extent of leaf damage, and marks the infestation start date when the leaf damage reaches an initial infestation threshold; and a primary prediction module that calculates the average daily growth rate of leaf damage from the infestation start date to the date the infestation risk threshold is reached. By monitoring leaf damage, the system can provide timely warnings of potential pest problems, enabling agricultural managers to take preventive measures before a large-scale outbreak. Furthermore, based on the pest risk assessment, agricultural managers can more accurately decide when and how to apply pesticides, thereby reducing unnecessary chemical use.

The embodiment of the invention provides an image processing method and device.The image processing method comprises the steps that in response to an editing instruction for an original image, the original image is updated into an edited image through an image editing model; inputting the original image and the edited image into an image evaluation model for processing to obtain image evaluation information corresponding to a plurality of image evaluation dimensions; dividing the editing image into an editing area and a non-editing area, and determining editing evaluation information corresponding to the editing area and non-editing evaluation information corresponding to the non-editing area; and according to the region information of the editing region, fusing the image evaluation information, the editing evaluation information and the non-editing evaluation information to obtain target evaluation information corresponding to the editing image.

This application discloses an unsupervised adaptive method and system for evaluating insurance claims images. The method includes: normalizing multimodal claims images; extracting feature vectors from a Siamese network encoder, bringing features of different modalities under the same case ID closer together and pushing features of images under different case IDs further apart, achieving case-level cross-modal semantic alignment; constructing a structured evaluation database storing historical image feature vectors and business dimension labels; generating evaluation results for new images through a similarity retrieval database; fusing contrast loss and reconstruction loss, and automatically adjusting the dynamic balancing parameter γ based on the retrieval matching degree or reconstruction error to achieve adaptive evolution of the encoder incremental updates. This application eliminates the need for manual annotation, solves the problems of semantic fragmentation in multimodal data and model update lag, and improves evaluation accuracy and efficiency.