393 results about "Image prediction" patented technology

SOLUTION: A semi-supervised medical image segmentation method based on a visual language model includes the steps of: obtaining a medical image; inputting an unlabeled image and a text description into a visual language model, and obtaining a text-guided mask based on obtained dense image embedding and text embedding; inputting a labeled image into a student model, and calculating supervised loss by using obtained labeled image prediction; respectively inputting the unlabeled image into the student model and a teacher model to obtain unlabeled image prediction and a pseudo label, merging the text-guided mask with the pseudo label, and calculating semi-supervised loss by using the merged pseudo label and unlabeled image prediction; and performing medical image segmentation by using a trained student model on the basis of the supervised loss and the semi-supervised loss.EFFECT: A target segmentation region can be accurately identified by using advantages of text descriptions.SELECTED DRAWING: Figure 1

The invention discloses a soil heavy metal inversion method and system integrating satellite remote sensing and near-end sensing, and the method comprises the steps: collecting a soil sample, and measuring the soil heavy metal content and a soil visible light-near infrared spectrum; obtaining a time sequence multispectral image of a research area, calculating a spectral index, and selecting and screening bare soil pixels through a threshold value to obtain a bare soil image; performing spectrum correction on the bare soil image; obtaining a joint dictionary and a sparse coefficient through sparse representation and dictionary learning, and reconstructing a hyperspectral image of the bare soil image; converting the one-dimensional spectral data into a two-dimensional spectrogram by using continuous wavelet transform, extracting spectral features in combination with a 2D-CNN algorithm, and constructing a soil heavy metal inversion model; and using the trained inversion model to predict the soil heavy metal content of the research area based on the reconstructed hyperspectral image. According to the method, satellite remote sensing and near-end sensing are integrated to obtain a large-scale accurate soil heavy metal content distribution map, deep features are extracted in combination with a 2D-CNN algorithm, and the inversion model precision and model efficiency are improved.

The invention discloses a distributed photovoltaic cluster power prediction method and device based on multi-modal fusion. The method comprises the following steps: acquiring historical photovoltaic data and historical photovoltaic images of a photovoltaic region to be predicted; analyzing a time sequence relationship in the historical photovoltaic data, extracting photovoltaic time sequence characteristics, and giving a first prediction result in combination with the data time sequence prediction model; extracting spatial features in the historical photovoltaic image, reconstructing the historical photovoltaic image, and giving a second prediction result in combination with the image prediction model; and in combination with a preset fusion weight, fusing the first prediction result and the second prediction result to obtain a target prediction power, and completing power prediction of the distributed photovoltaic cluster, thereby effectively capturing the influence of sudden weather events on photovoltaic power generation, improving the accuracy of conventional cloud picture data when coping with complex and changeable cloud layer motion, and improving the prediction efficiency of the distributed photovoltaic cluster. Therefore, the accuracy of power prediction is improved.

The invention relates to a textile product defect identification method based on improved YOLOv11. The method comprises the following steps: acquiring a textile product defect image data set; performing pretreatment; dividing into a training set and a verification set; the method comprises the following steps: introducing MConv into a YOLOv11 backbone network, adding a CCIAP module behind a C2PSA module, and applying BiFPN in a path aggregation network; performing prediction through YOLO Head to obtain N prediction feature maps; the overall loss of the network is calculated, and network parameters are optimized through back propagation; predicting the verification set image through a network to output AP values of various categories; repeating the above steps to obtain a trained YOLOv11 network; and detecting the test image or video by using the trained detector to obtain a detection result. According to the method, the MConv is introduced into the YOLOv11 network to enlarge the receptive field, the CCIAP module is added behind the C2PSA to improve the feature extraction capability, and the BiFPN is applied to the Neck layer to enhance the feature fusion capability, so that the target detection precision is improved and the real-time detection of textile product flaws is realized under the condition that the reasoning speed is not influenced.

The invention discloses a cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning, and the method specifically comprises the following steps: introducing a difficult sample feature mining method, dynamically adjusting the loss weight of a sample according to the change of similarity, and building a dynamic difficult sample triple loss model; the method comprises the following steps: dynamically adjusting a confidence threshold value of a sample by adopting an index moving average weighting method, iteratively training and screening an unlabeled sample, namely a pseudo label, establishing a pseudo label self-training mechanism of a dynamic threshold value, mining and utilizing non-paired data, and solving the problem of high manual labeling cost; a reference image most similar to a query image is found through image retrieval, and the offset of a query position is predicted. Experiments on CVUSA and CVACT data sets show that as the distance threshold increases, the accuracy of the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning presents a stable rising trend, and the cross-view image geographic positioning method based on dynamic threshold pseudo tag self-training learning is superior to other methods under the same threshold condition.

A text-to-image model training method, apparatus, and computer-readable storage medium for enhancing text-to-image generation through object-aware training. The method trains a text-to-image model using cyclic iterative training with sample image and text pairs. Training involves selecting image-text sample pairs containing multiple objects, obtaining corresponding mask images and object class names that distinguish location regions of the objects, and inputting both the sample image with description text and the mask images with object class names into the model. The method obtains image predicted noise and object predicted noises, constructs a loss function based on these predictions, and performs parameter adjustment accordingly. This approach enables improved object-level understanding in text-to-image generation models.

The invention discloses a multispectral satellite cloud picture prediction method based on motion stripe decoupling. The method comprises the following steps: carrying out normalization preprocessing on multi-channel satellite observation data; utilizing a motion branch model to extract motion features based on a displacement field, iteratively updating a prediction frame in an autoregressive distortion-correction pipeline, and keeping physical consistency in combination with atmospheric dynamics and smoothness constraint; a texture branch model is utilized to sequentially pass through a high-fidelity encoder, long memory state space modeling and a high-fidelity decoder, time sequence texture features are extracted, and cloud picture details are kept; the motion features output by the motion branches and the texture features output by the texture branches are input into a gating fusion module, adaptive weighting of the features is achieved through convolution and a gating mechanism, and fusion features are output; and carrying out reverse normalization processing on the fusion features to obtain satellite cloud picture prediction results at a plurality of moments in the future. According to the method, the spatial texture fidelity of cloud picture prediction can be improved while the physical interpretability is ensured, and high-precision satellite cloud picture prediction is realized.

The invention belongs to the technical field of medical image processing, and particularly relates to a glioma segmentation method based on a multimodal fusion network guided by anatomical symmetry, which comprises the following steps of: jointly inputting an FLAIR image, a T2 image, a T1 image and a T1c image of the same glioma into a trained image segmentation model, and outputting a predicted segmentation image by the trained image segmentation model, the prediction segmentation image is a glioma MRI image with three segmentation areas obtained through prediction, and the three segmentation areas are an edema area, an enhanced tumor area and a necrosis area respectively; the image segmentation model comprises an encoder, a jump connection part and a decoder; the encoder comprises an ASG module, the jump connection part comprises an IMP module, and the decoder comprises a CMF module. Through a three-module cooperation mechanism, the performance of tumor localization, cross-modal fusion, subregion segmentation and the like is improved, and a reliable image basis is provided for glioma operation plan formulation, prognosis evaluation and personalized treatment decision.

The invention discloses a satellite cloud picture sequence prediction method combining deformable convolution and pyramid pooling, and relates to the field of deep learning. The method comprises the following specific steps: (1) preprocessing satellite cloud picture sequence data and dividing into a training set and a test set; (2) building a satellite cloud picture sequence prediction model of a basic encoder-translator-decoder structure; (3) combining a motion perception loss function and an L2 loss function to supervise the model for training; and (4) predicting a satellite cloud picture sequence image. According to the method, a deformable volume operator based on dynamic sparseness is used in an encoder and a decoder to replace the traditional convolution operation, so that the network can adaptively select proper parameters according to the space structure of a cloud picture, softmax normalization in deformable convolution space aggregation is removed, and the robustness of the network is improved. The memory access is optimized to accelerate the running speed; a pyramid pooling structure is used in the translator, so that the calculation process can be simplified, and spatial-temporal characteristics of different scales in the cloud picture sequence can be captured; meanwhile, a motion perception loss function and a traditional L2 loss function are combined to supervise model training, and motion information between adjacent frames of a satellite cloud picture sequence can be further obtained. The method is not only suitable for all satellite cloud picture sequence images, but also can be applied to sequence image prediction in other complex scenes.

The invention discloses a remote sensing big data processing method and system based on partitioning and parallel machine learning. The processing method comprises the steps of obtaining remote sensing big data; based on a preset rule, partitioning the remote sensing big data to obtain a plurality of sub-region data; inputting the plurality of sub-region data into a pre-constructed remote sensing image prediction model in parallel to obtain each piece of sub-remote sensing image information corresponding to each piece of sub-region data, the remote sensing image prediction model comprises a plurality of sub-models, and each piece of sub-region data is selected based on a preset screening rule to obtain a plurality of pieces of sub-remote sensing image information corresponding to each piece of sub-region data; selecting one sub-model from the plurality of sub-models for independent training and prediction; and integrating each piece of sub remote sensing image information corresponding to each piece of sub region data to obtain a target remote sensing image. Multiple sub-region data are input into the sub-model in parallel, a large amount of data can be processed, the overall operation efficiency is improved, spatial heterogeneity can be better captured, and the remote sensing image prediction precision is improved.

The invention relates to a dynamic fine tuning method and device for a visual detection model, a storage medium and equipment, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining an original image, inputting the original image and question prompt information corresponding to the original image into a preset large visual language model, and obtaining multiple pieces of candidate response information; according to an image prediction result in each piece of candidate response information, constructing a dynamic derivable reward function corresponding to each piece of candidate response information; determining a current reward value of each piece of candidate response information according to the dynamic derivable reward function, and generating a model optimization strategy according to a preset strategy optimization function and the current reward value; and performing dynamic fine adjustment on the preset large visual language model based on the model optimization strategy to obtain the visual detection model. The training efficiency of the visual detection model is improved.

The invention provides a crack image segmentation method, system and device for a large building surface. The method comprises the following steps: acquiring a building surface original image of a to-be-detected target area; extracting a multi-dimensional feature vector of the original building surface image through a heterogeneous network formed by a convolutional neural network and a Transform decoder branch, and generating an image prediction result; carrying out dynamic adjustment based on an image prediction result, and carrying out constraint by adopting a consistency supervision mechanism to obtain an optimized prediction label; pixel-level contrast learning is carried out on the multi-dimensional feature vector and the optimized prediction label, and a final crack segmentation result is obtained; and reasoning based on the final crack segmentation result to obtain a crack segmentation probability graph. According to the method, automatic identification and accurate segmentation are carried out on the surface cracks of key infrastructures such as a dam through an artificial intelligence algorithm, the crack detection precision is improved under the condition of limited labeled data, the model robustness is enhanced, the facility operation and maintenance efficiency is optimized, and the structure safety is improved.

The invention relates to the technical field of image processing, and discloses a stomach tumor image segmentation method and system based on a hybrid model, a terminal and a storage medium, and the method comprises the steps: carrying out the resampling of a stomach tumor image, carrying out the normalization processing of voxels, and obtaining a compressed image set of each target object; performing iterative optimization on the compressed image set by using a generative adversarial network to obtain a target false image set, inputting the target false image set into an encoder of a target segmentation network, fusing the target false image set to obtain dimension fusion information, screening multiple pieces of expert data through a gating mechanism, and generating a corresponding fusion weight; and training the tumor segmentation model by utilizing expert data so as to predict the compressed image set and output an image prediction result. According to the method, iterative optimization is carried out on the image, finally, a sample with better quality is input into the model to participate in training, finally, the generalization ability and robustness of the model can be improved on the premise of not enhancing the labeling cost, and the accuracy of a prediction result is improved.

A method performed by an electronic apparatus, includes: obtaining position information of an object in an input image; predicting trailing blur state information of the input image, based on the position information and the input image; and obtaining an output image by performing processing on the input image, based on the predicted trailing blur state information.

The invention provides an image authenticity determination method and device, equipment, a medium and a program product, and relates to the technical field of image processing. The method comprises the following steps: acquiring an original image; performing feature extraction processing on the original image from different dimensions to obtain image features of corresponding dimensions; carrying out feature fusion processing on the image features extracted from different dimensions based on a self-adaptive weight fusion mechanism to obtain multi-modal features with dynamically adjusted weights; and inputting the multi-modal features into a trained image prediction model for prediction processing, and outputting a tampering probability value for tampering the original image. The embodiment of the invention is used for solving the defect of low image authenticity judgment accuracy in the prior art, effectively integrating the information of three dimensions of the spatial domain, the frequency domain and the statistical features of the original image, and predicting the tampering probability value of the original image on the basis of feature fusion, so as to determine the authenticity of the original image according to the probability value, thereby improving the authenticity judgment accuracy of the original image. And the authenticity discrimination accuracy is greatly improved.

The invention relates to the technical field of artificial intelligence and computer vision, in particular to a time sequence remote sensing image prediction method and related equipment, and the method comprises the steps: firstly obtaining historical multi-frame remote sensing images of a target area, and constructing a time-space sequence data set; the model is a ViT-Informer model, and a ViT spatio-temporal feature extractor captures spatio-temporal correlation features among images through a self-attention mechanism to generate a spatio-temporal joint coding sequence; the Informer time sequence predictor adopts a probability sparse attention mechanism to process long sequence dependence, and multi-step recursive prediction is achieved. The model can effectively model a remote sensing image spatio-temporal evolution law, and a prediction image at a specified moment in the future is generated through iterative reasoning.

The invention belongs to the technical field of medical image analysis, and relates to a brain tumor segmentation method based on a boundary perception mechanism, and the method comprises the steps: inputting T1, T1c, T2 and Flair images of a brain tumor into a trained image segmentation model, and outputting a prediction segmentation image through the trained image segmentation model, the prediction segmentation image is a brain tumor MRI image which is obtained through prediction and has a complete tumor area, a tumor core area and an enhanced tumor area; according to the brain tumor segmentation method based on the boundary perception mechanism provided by the invention, the boundary perception mechanism is introduced, and the boundary information is fused into the image segmentation model, so that the discriminability of the model to features is improved, and accurate segmentation of tumor subregions is realized; a multi-modal fusion method is adopted, different MRI sequence complementary information is integrated, and tumor features are comprehensively understood; in combination with uncertainty quantification and a loss function based on uncertainty, confidence measurement is provided for a segmentation result, the accuracy and reliability of segmentation are enhanced, and a clinician is assisted in evaluating a prediction result.

The invention discloses a wavefront detection-free adaptive optical correction method and system, and the method comprises the steps: obtaining a real in-focus image and a real out-of-focus image of a to-be-corrected light beam, and outputting an updated deformable mirror control voltage through a physical information neural network; generating a predicted wavefront phase according to the deformable mirror control voltage signal and a deformable mirror voltage-phase mapping model; generating a predicted in-focus image and a predicted out-of-focus image through a derivable Fourier optical propagation model; constructing a label-free composite loss function according to the real in-focus image, the real out-of-focus image, the predicted in-focus image and the predicted out-of-focus image; performing unsupervised training updating on the physical information neural network according to the composite loss function, and performing repeated iteration to obtain an optimized deformable mirror control voltage; and driving the deformable mirror to carry out self-adaptive optical correction in real time according to the optimized deformable mirror control voltage to obtain a correction result. According to the invention, the correction efficiency and the system stability are obviously improved.

The invention provides a kidney cancer recurrence risk prediction method based on a deep learning model, and relates to the technical field of deep learning, and the method comprises the steps: collecting an image data set for kidney cancer high recurrence risk prediction; carrying out registration on the collected multi-stage enhanced CT image; constructing and training a kidney tumor automatic detection and segmentation model; carrying out ROI positioning cutting and quality control; and constructing a deep learning model for renal cancer recurrence risk prediction based on the multi-modal convolutional neural network, and realizing renal cancer recurrence risk prediction through the constructed prediction network model. According to the method, the multi-phase enhanced CT image of the kidney cancer patient is analyzed through the deep learning model, the tumor postoperative recurrence risk is predicted, an objective basis is provided for a clinician to make an individualized follow-up visit scheme and an auxiliary treatment decision, and excessive treatment of a low-risk patient and insufficient treatment of a high-risk patient are avoided.

The invention relates to a method and a device for generating a report under the assistance of a label, and the method comprises the following steps: 1) extracting a structured label set from a text report of a sample based on a large language model, wherein the structured label set comprises a multi-classification group consisting of dichotomous labels and mutual exclusion options; 2) aggregating the labels in batches, after a threshold value is reached, merging and de-weighting, performing specification and mutual exclusion group merging on synonymous, near-synonymous and redundant labels, and converging into a unified label library; 3) based on the text report and the tag library, outputting a tag subset of each sample through a large language model; 4) multi-modal multi-label classification model training: extracting each visual modal feature, performing weighted aggregation and splicing, and outputting each label group logits through a classification head to perform weighted group loss optimization; (5) carrying out joint training on the multi-modal large language model by using samples of'only images-reports' and'images + labels-reports', and (6) carrying out label prediction and screening on the images by using the classification model, and inputting'images + prediction labels' into the multi-modal large language model to obtain a final report.

The invention discloses a night near-shore surface layer suspended sediment prediction method fusing a dynamic model and remote sensing, and relates to the technical field of night suspended sediment prediction, and the method comprises the following steps: obtaining a real-time blue-green ratio based on a real-time remote sensing image; acquiring a first type of historical data; obtaining an image prediction function based on the first type of historical data; acquiring the flow velocity of to-be-detected offshore seawater, and marking the flow velocity as a real-time flow velocity; acquiring a second type of historical data; obtaining a dynamic prediction function based on the second type of historical data; obtaining an image prediction concentration based on the real-time blue-green ratio and an image prediction function; obtaining a power prediction concentration based on the real-time flow velocity and a power prediction function; obtaining a final suspended sediment prediction concentration based on the image prediction concentration and the power prediction concentration; the method is used for solving the problem that the concentration of suspended sediment cannot be accurately obtained through remote sensing at night due to the fact that an appropriate suspended sediment prediction method cannot be set at night based on remote sensing in an existing suspended sediment prediction technology.

The system generates a three-dimensional model with layered depth images based on an input two-dimensional image. For training, layered depth images are derived from existing three-dimensional models. The system trains a machine learning model to predict multiple layered depth images from an input image of an object. The system compares the generated, multiple layered depth images to the derived layered depth images for the object to update the machine learning model during training. At inference time, the system receives an input image for an object. The system applies the machine learning model to the input image to output predicted layered depth images. The system generates a three-dimensional model from the predicted layered depth images.

The invention discloses a video frame image coding method and device, a video frame image decoding method and device and a storage medium. The coding method of the video frame image comprises the following steps: acquiring an inter-frame image matching block corresponding to a target image block in a reference frame adjacent to the video frame according to the target image block in the video frame; obtaining an intra-frame image matching block corresponding to the target image block in the video frame according to the target image block; determining a first weight of the inter-frame image matching block and a second weight of the intra-frame image matching block according to matching errors between the target image block and the inter-frame image matching block and between the target image block and the intra-frame image matching block; determining a matching image prediction result according to the inter-frame image matching block, the intra-frame image matching block, the first weight and the second weight; and coding the target image block based on the inter-frame image matching block and the matched image prediction result to obtain an image coding result. According to the embodiment of the invention, the prediction performance of the to-be-coded target image block can be improved, so that the coding performance of the to-be-coded target image block can be improved.

The invention belongs to the technical field of image recognition, and provides a classified dumping management method and system for a garbage compression station based on computer vision. The method comprises the following steps: predicting a dumping time window of a garbage transport vehicle based on a first video image of the garbage transport vehicle in a preset area and a second video image of another vehicle in a garbage dumping area; shooting a plurality of high-definition close-range images of the garbage transport vehicle at the dumping time window, extracting first image features of garbage carried by the garbage transport vehicle based on the high-definition close-range images, and fusing the first image features to obtain second image features; processing a historical dumping record of the garbage transport vehicle by adopting an attention mechanism, and classifying the second image features based on a processing result to obtain a garbage type; and generating a dumping guide instruction. According to the invention, invalid snapshots can be significantly reduced, the feature extraction efficiency and classification accuracy are improved, the labor cost and the error investment rate are reduced, and intelligent classified dumping management is realized.

Provided herein are systems and methods for optimizing a 3D model of an individual's teeth. A 3D dental model may be reconstructed from 3D parameters. A differentiable renderer may be used to derive a 2D rendering of the individual's dentition. 2D image(s) of an individual's dentition may be obtained, and features may be extracted from the 2D image(s). Image loss between the 2D rendering and the 2D image(s) can be derived, and back-propagation from the image loss can be used to calculate gradients of the loss to optimize the 3D parameters. A machine learning model can also be trained to predict a 3D dental model from 2D images of an individual's dentition.

The application provides an image deblurring processing method, relates to the field of Internet of Vehicles and the field of artificial intelligence technology, and comprises the following steps: performing cascade coding processing on a first image based on M scales, and sequentially obtaining M scale coding images; performing cascade refining processing on the second scale coding image to the M scale coding image based on N scales, and sequentially obtaining N scale refining images; performing cascade decoding processing on the second scale coding image to the M scale coding image and the N scale refining image based on N scales, and sequentially obtaining N scale decoding images; and performing image prediction processing on the first scale coding image, the N scale decoding image in the N scale decoding image of the N scale decoding image, and the N scale refining image in the N scale refining image of the N scale refining image, so as to obtain a second image with higher definition than the first image. Through the application, the deblurring effect of multimedia images can be significantly improved.

The present invention discloses a large-scale greenhouse interpretation method based on drone images, which belongs to the field of image processing technology. The present invention first constructs and trains a multi-task greenhouse extraction network, which includes an image feature extraction network and a multi-task prediction network, and obtains a greenhouse interpreter based on the trained image feature extraction network and the greenhouse segmentation task prediction branch. A first multispectral image and a first visible light image are obtained by aligning multispectral images of different channels; and the first multispectral image is interpreted by the greenhouse interpreter to obtain a small image prediction result; the first visible light image is spliced ​​to obtain a large visible light image, and the small image prediction result is used as the spliced ​​input image, and the corresponding camera posture information and point cloud information of the first visible light image are used as the camera posture of the spliced ​​input image and the point cloud information between images, and image splicing processing is performed to obtain a greenhouse interpretation result of the large visible light image. The present invention can have high-precision registration performance in a variety of scenarios.

This invention belongs to the field of medical image analysis technology and relates to a brain tumor segmentation method based on a boundary-aware mechanism. The method involves inputting T1, T1c, T2, and FLAIR images of the brain tumor into a trained image segmentation model, which outputs a predicted segmented image. The predicted segmented image is a brain tumor MRI image with a complete tumor region, a tumor core region, and an enhanced tumor region, obtained through prediction. The proposed brain tumor segmentation method based on a boundary-aware mechanism incorporates boundary information into the image segmentation model, improving the model's ability to distinguish features and achieving accurate segmentation of tumor subregions. A multimodal fusion method is used to integrate complementary information from different MRI sequences, providing a comprehensive understanding of tumor characteristics. Furthermore, uncertainty quantification and an uncertainty-based loss function are combined to provide confidence measurements for the segmentation results, enhancing the accuracy and reliability of the segmentation and assisting clinicians in evaluating the prediction results.

The application discloses a light field image quality evaluation method based on rich features. The steps of the application are as follows: 1: randomly split the data set into a training set and a test set. Through fixing two dimensions of angle information, the light field image is converted into a 9*9 sub-aperture image array. 2: convert each sub-aperture image from an RGB space to an HSV coordinate, and extract brightness, hue and saturation as color information features. 3: extract disparity structure features. 4: obtain angle texture features of the image. 5: standardize all the features, splice and fuse, and obtain a light field image prediction score by using a support vector model (GA-SVR) based on a genetic algorithm. The application uses high-dimensional singular value decomposition, reduces information redundancy between multiple sub-aperture images, improves calculation efficiency, combines color information features, disparity structure features and angle texture features, better simulates a human visual system, and makes a more scientific evaluation on the light field image.

The invention discloses an intensive agricultural crop detection method and system based on deep learning. The method comprises the following steps: extracting different branch features of a dense agricultural image through a MobileNet pre-training model; obtaining a dense agricultural image rough prediction map, a dense agricultural image detail prediction map and a dense agricultural image prediction map model; and by sampling the rough prediction map, the detail prediction map and the prediction map to the size of an original image, constructing a loss function calculation error, and using the obtained error to reversely update the dense agricultural segmentation model. And inputting the dense agricultural image into the updated model, and obtaining and outputting a corresponding dense agricultural image segmentation prediction map. According to the method, the capturing capability of hidden crops in dense agriculture is enhanced, and the light weight of the model is kept while the segmentation precision is remarkably improved.