373 results about "Semi-supervised learning" patented technology

The invention provides a semi-supervised image semantic segmentation method and system based on a visual basic model, and the method comprises the steps: constructing a multi-task model which comprises a visual basic model and a depth estimation basic model, and the visual basic model is connected with a task solution head, an adapter parameter efficient fine tuning module and a multi-modal cross fusion module; the task solution head comprises a semantic segmentation head and a depth estimation head; extracting semantic hierarchy features and a depth feature map of the RGB image, performing cross attention fusion on the semantic hierarchy features and the depth feature map, and inputting obtained fusion features into a semantic segmentation head and a depth estimation head respectively; semi-supervised learning is adopted to train a multi-task model, only parameters in the adapter parameter efficient fine tuning module and the multi-modal cross fusion module are trained, and a multi-task loss function is adopted. The image semantic segmentation model obtained through training can improve semantic segmentation performance, reduce training cost and is suitable for different tasks.

The invention provides a semi-supervised target detection method for a visible light-infrared multi-mode fusion scene. The method comprises the following steps: constructing a semi-supervised visible light-infrared multi-modal image data set based on an LLVIP data set; on the basis of a YOLOv11 model architecture, constructing a target detection model oriented to multi-modal image feature fusion, and training the target detection model by using a semi-supervised visible light-infrared multi-modal image data set in a deep learning end-to-end mode to obtain a trained target detection model; and inputting a to-be-detected multi-modal image into the trained target detection model, and outputting a target detection result of the to-be-detected multi-modal image by the trained target detection model. The method is based on a semi-supervised learning normal form, so that the precision and robustness of target detection in a multi-modal fusion scene are improved, and the requirements for high efficiency and reliability of target recognition in practical application scenes such as intelligent traffic and intelligent security and protection are met.

The invention discloses a winter wheat LAI and SPAD estimation method based on a lightweight semi-supervised model, and relates to the technical field of agricultural remote sensing monitoring, and the method comprises the steps: obtaining multispectral image data of a winter wheat key growth period, and carrying out the preprocessing of the multispectral image data to generate a standardized multichannel vegetation index image; the method comprises the following steps: constructing a lightweight semi-supervised model MCVI-SANet, and carrying out self-supervised training on the MCVI-SANet by adopting a semi-supervised training strategy driven by VICReg; and inputting the multi-channel vegetation index image into the trained MCVI-SANet, and outputting quantitative estimation results of the LAI and SPAD of the winter wheat. Through combination of a saturation perception mechanism and semi-supervised learning, estimation deviation caused by dense canopy vegetation index saturation and data noise is effectively eliminated, and the estimation precision and generalization ability in a complex agricultural scene are improved while the lightweight deployment characteristic of the model is ensured.

A dynamic adaptive learning method for mineral prediction includes: collecting a dataset including geological data and labels of the geological data; extracting features from the geological data, initializing parameters of a training model and optimizing the parameters to obtain training parameters; performing an associative training on the training model based on the training parameters and the labels in a dynamic adaptive learning framework to obtain a mineral prediction model, algorithms of the associative training including a variational expectation algorithm and a variational maximization algorithm, and the variational expectation algorithm including an unsupervised learning mode, a semi-supervised learning mode, and a fully supervised learning mode; and predicting, by using the mineral prediction model, a mineral to obtain a mineral prediction result. The method can break through limitations of the traditional machine learning technology, offering a more efficient, universal, and stable strategy for geophysical data analysis and mineral resource assessment.

The invention discloses a human body tumble detection method and device, and belongs to the technical field of human body tumble detection, and the method comprises the steps: obtaining a radar point cloud time sequence which comprises continuous frame point clouds of all parts of a human body within a set time; performing relative coordinate coding on each frame of point cloud in the radar point cloud time sequence to obtain a spliced feature vector time sequence; and inputting the spliced feature vector time sequence into a trained human body tumble detection model to obtain a human body tumble detection result, and training is semi-supervised learning training based on a dynamic threshold. Relative coordinate coding is realized through construction of a local coordinate system and coordinate conversion of the point cloud, dynamic enhancement of four-limb micro-motion features in the falling process is realized, a better classification effect is shown, and thus the detection accuracy is improved; through semi-supervised learning training based on a dynamic threshold value, the labeling requirement is greatly reduced, and high detection precision in a small amount of training data scene is also realized.

The invention relates to a semi-supervised learning data anomaly intelligent identification and treatment system, which is applied to a hydrogen energy commercial vehicle, and comprises a data acquisition and preprocessing module, which is arranged on a vehicle-mounted terminal and is used for acquiring a hydrogen storage system signal, a hydrogen supply system signal, a fuel cell system signal and a whole vehicle system signal in real time, processing the acquired signal data; the semi-supervised anomaly recognition module is arranged on a cloud platform, is connected with the data acquisition and preprocessing module, and is used for fusing the processed signal data into a rule engine and semi-supervised learning, constructing a semi-supervised training data set, and performing deep auto-encoder model training through the data set so as to realize accurate anomaly recognition of few sample scene writing; and the exception treatment and feedback module is arranged at an edge node, is connected with the semi-supervised exception recognition module, carries out exception recognition through a trained model, carries out graded treatment according to the exception severity, and establishes a model evolution mechanism to realize continuous optimization of the system.

The invention provides a semi-supervised spine segmentation method based on a global-local semantic constraint visual language model. Comprising the following steps: respectively extracting spine MR slice image embedding and structure description text embedding by using an image encoder and a text encoder; learning the semantic relationship between the image blocks and the text in the local level to measure the uncertainty of the local view; the Wasserstein 2 distance between the cross-modal distribution representations is measured in a global level so as to standardize the global semantic similarity; generating a prompt guide mask based on a pre-trained GLsc model in combination with a text and an unmarked slice, and reinforcing semantic constraint through feature supervision and region alignment loss; predicting results of student and teacher models are fused to generate a high-quality pseudo label, and a segmentation network is optimized by means of supervised and consistent loss. According to the method, the prompt guide mask is introduced into spine segmentation semi-supervised learning, the VLM cross-modal uncertainty perception capability is enhanced through the two-dimensional semantic constraint, the pseudo tag quality and the segmentation precision are effectively improved, and an innovative technical normal form is provided for medical image semi-supervised segmentation.

The invention discloses a wall surface heat flow prediction method considering material ablation response under a semi-supervised learning framework, and relates to the field of aerospace, and the method comprises the following steps: constructing a data set; building a neural network model, and introducing the ablation physical information loss item into a loss function of the neural network model; pre-training the neural network model; predicting the label-free data through a pre-trained neural network model, and generating a pseudo label; mixing the labeled data and the unlabeled data, and performing semi-supervised self-training on the pre-trained neural network model to obtain a final neural network model; and wall surface heat flow prediction of material ablation response is carried out through the final neural network model. According to the method, the requirement of the model for a labeled data set is reduced, the model generalization ability under the small sample size training condition is enhanced, the physical interpretability of the network model is improved, and the wall surface heat flow prediction precision is improved.

Implementations are provided for leveraging training data that is less costly to collect than state-action sequences to perform semi-supervised training of robot control policies. In various implementations, a first input prompt may be assembled with representations of an observed initial state of a robot and a goal state of the robot. The first input prompt may be processed using a goal-conditioned trajectory model to generate first output indicative of a sequence of predicted states to be reached by the robot between the observed initial and goal states. A second input prompt may be assembled to include representations of the sequence of predicted states. The second input prompt may be processed using an action prediction model to generate second output indicative of a sequence of predicted actions to be performed by the robot to reach the sequence of predicted states.

The invention discloses a wheat seedling missing detection method for performing semi-supervised learning by fusing spatio-temporal information, and belongs to the technical field of seedling missing detection methods. The invention discloses a wheat seedling missing detection method for performing semi-supervised learning by fusing spatio-temporal information. According to the method, a set of semi-supervised learning framework is constructed; an initial model is trained by using a small number of precisely-labeled frames, and pseudo labels are generated for unlabeled frames; and'time sequence consistency 'and confidence coefficient are introduced to form a joint screening mechanism, and high-quality spatio-temporal data are obtained at low cost. In order to break through the limitation of single-frame detection, a space-time attention module is integrated in a detection network, and multi-frame information is aggregated to improve the robustness to motion blur and the like. The invention also provides a new normal form of space anomaly detection, which does not directly identify the missing seedlings, but learns a normal seedling space distribution rule, and identifies a low-density area as a seedling missing area by using a K-nearest neighbor (KNN) algorithm. According to the method, an unstable classification task is converted into a robust statistical problem, and the method has the advantages of low cost, high robustness and strong generalization.

The invention discloses an oral cavity tooth CBCT image segmentation method based on semi-supervised deep learning, and belongs to the technical field of oral diagnosis, and the method comprises the steps: collecting oral cavity CBCT image data, carrying out the preprocessing, and building an oral cavity CBCT image data set containing labels and no labels; a semi-supervised segmentation model is constructed based on a progressive average teacher framework, the semi-supervised segmentation model comprises double groups of teacher-student network structures, and each group comprises two student networks with different structures and a double-teacher network; iterative training is carried out on the semi-supervised segmentation model, the training process comprises a supervised learning stage and a semi-supervised learning stage, and a trained semi-supervised segmentation model is obtained; and taking any oral cavity CBCT image data as input, and performing tooth segmentation based on the trained semi-supervised segmentation model. According to the method, the accuracy of tooth segmentation can be improved, the segmentation effect at the tooth boundary position is enhanced, the tooth segmentation steps are simplified, and the diagnosis efficiency of doctors is improved.

The invention discloses a telecommunication fraud identification method and device, equipment, a storage medium and a product, and relates to the technical field of artificial intelligence, and the method comprises the steps: constructing a first communication behavior heterogeneous graph of telecommunication users; mapping the first Euclidean feature of each node to a first hyperbolic space of a multilayer hyperbolic graph convolution layer through a multi-curvature parameterized index to obtain a first hyperbolic node feature, and allocating different curvature parameters to different types of nodes; in a multilayer hyperbolic graph convolution layer, updating the first hyperbolic node features to obtain an embedded matrix; optimizing model parameters of a multi-curvature hyperbolic heterogeneous graph neural network model containing multiple hyperbolic graph convolutional layers through a semi-supervised learning mode in combination with the embedded matrix to obtain a trained multi-curvature hyperbolic heterogeneous graph neural network model; and outputting a node fraud probability based on the trained multi-curvature hyperbolic heterogeneous graph neural network model to identify telecommunication fraud users, thereby improving the accuracy and efficiency of fraud identification in a telecommunication scene.

The invention discloses an electric power-traffic coupling network vulnerability identification method and system. The method comprises the following steps: S1, coupling an electric power network and a traffic network to form a graph model; the graph model comprises a node set and an edge set; constructing an adjacent matrix and a node feature matrix based on the node set and the edge set; s2, inputting the adjacent matrix and the node feature matrix into a graph convolutional neural network model, performing feature extraction model training through a semi-supervised learning mechanism, and outputting an updated node feature matrix; and S3, learning an optimal strategy by using a deep Q network, and identifying a fragile node sequence through interaction with the environment. According to the method, a traditional single network evaluation mode is broken through, accurate quantification of the cross-network coupling effect is achieved, complex topological information can be captured, intelligent dynamic recognition is achieved, dependence on artificial experience is reduced, decision visualization can be supported, the system practicability is high, and the method can be applied to power-traffic coupling network catastrophe scenes.

The invention relates to the technical field of geological disaster hidden danger recognition, in particular to a geological disaster hidden danger recognition method based on InSAR and semi-supervised learning, which comprises the steps of dividing hydrological sensitivity of a monitoring area according to a regional underground water unit supply schematic diagram, and setting an InSAR data sampling frequency; acquiring earth surface time sequence displacement data, including accumulated displacement, displacement rate and displacement acceleration, by using a set sampling frequency for preliminarily identifying a potential hidden danger area; geological radar detection is carried out in the potential hidden danger area, underground structure features are evaluated through a geological radar profile map, and a multi-source feature vector is constructed; according to the method, the hydrological sensitivity of the monitoring area is divided according to the regional groundwater unit supply schematic diagram, and the InSAR data sampling frequency is set, so that the problem that the high-risk area is not monitored in time due to the fact that the hydrological sensitivity change is not considered because the conventional monitoring frequency mostly adopts a fixed sampling interval is solved.

The invention discloses a dynamic early warning method and system for soft rock tunnel extrusion large deformation catastrophe, and the method comprises the steps: building a soft rock tunnel surface reference model, carrying out the point cloud registration, updating the soft rock tunnel surface model, recognizing a deformation mode, adjusting a deformation mode threshold value, carrying out the apparent dynamic early warning, and carrying out the multi-scale compensation. Constructing a Bayesian catastrophe prediction network based on a control-state coupling mechanism to obtain a large-deformation dynamic catastrophe index, adjusting a threshold value of the large-deformation dynamic catastrophe index to perform internal dynamic early warning, performing multiple regression to obtain a damage mapping relation set, constructing a deformation feature vector, and performing dynamic early warning; and constructing a dangerous event prediction model based on semi-supervised learning, obtaining a dangerous event prediction result in combination with the destruction mapping relation set, and performing event dynamic early warning. The method not only can improve the precision of dynamic early warning of the soft rock tunnel extrusion large deformation catastrophe, but also has good interpretability, and can be directly applied to a dynamic early warning system of the soft rock tunnel extrusion large deformation catastrophe.

The invention provides a fault detection semi-supervised learning method for a three-dimensional seismic data sparse labeling scene. The fault detection semi-supervised learning method is used for realizing stable and continuous three-dimensional fault identification under the condition that only a small number of two-dimensional slices are labeled and a large number of voxels are not labeled. According to the method, firstly, a supervision mask is generated in a three-dimensional space based on two-dimensional marking, and supervision learning is executed only in an effective area to adapt to marking scarcity characteristics; and then, constructing a double-student model, keeping prediction coordination in an unlabeled area through mutual generation of pseudo labels, probability consistency and confidence constraint, and applying consistency constraint to overlapped areas of adjacent three-dimensional sub-blocks in combination with a teacher model so as to maintain spatial continuity of a fault structure. Through joint optimization of synthetic data pre-training and field data migration training, the model can still obtain reliable fault structure expression under the field conditions of strong noise, weak reflection and obvious cross-work-area difference.

The invention belongs to the field of semi-supervised learning and semantic segmentation, and discloses a pseudo tag generation method based on conformal prediction and application thereof in semantic segmentation. Calibrating and predicting the visual basic model by using conformal prediction, and generating a corresponding category prediction set for a semi-supervised semantic segmentation unlabeled image; constructing a teacher model, training on the labeled data, and outputting a category label in the category prediction set by using the teacher model as a pseudo label of non-labeled data for subsequent training; performing parameter initialization on a student model to be trained by using a teacher network, and adopting a double-stage training strategy: training the model by using labeled data and non-labeled data with a pseudo label in a full supervision mode; and carrying out fine tuning on the student model by adopting a self-training strategy of false label removal. According to the method, extraction and utilization of effective information in unlabeled data by the model are enhanced, meanwhile, the problem of domain deviation existing in a basic model and a downstream task is relieved, and strong semantic representation capacity is injected into the model.

A sleep staging method based on semi-supervised learning belongs to the field of artificial intelligence and health monitoring, and comprises the following steps: preprocessing original PPG signals, extracting heart rate and respiration signals, calculating morphological and frequency domain features, extracting sleep staging related representation features, modeling short-time features and long-time changes of sleep, and outputting a classification result of each sleep stage; calculating supervised loss by using the labeled PPG data, predicting unlabeled PPG data to obtain a soft label, screening a high-confidence sample, and converting the soft label into a hard label to calculate unsupervised loss; calculating a confidence coefficient, calculating category center features of each sleep stage category, and performing feature adjustment; performing confidence weighting on the sample features and the category center features; and constructing positive and negative sample pairs to carry out semi-supervised contrast learning. According to the method, the generalization ability of the sleep staging model is enhanced, the sleep staging accuracy is improved, and the method can be applied to scenes such as family health management, auxiliary diagnosis and sleep monitoring.

The invention belongs to the technical field of image processing, and discloses a multi-scale heterogeneous rock scanning image segmentation model construction method, which is based on structure perception and semi-supervised learning, and comprises the following steps: constructing an annotated image; generating a pseudo label image; dividing the labeled images and the pseudo label images into a plurality of batches of training subsets; respectively inputting the annotation image and the pseudo-tag image of any batch of training subsets into a rock-soil scanning image segmentation model, outputting corresponding segmentation images, and obtaining corresponding structure perception indexes at the same time; based on the structure perception index, respectively constructing loss functions corresponding to the annotated image and the pseudo label image, and adding the loss functions to obtain total loss; optimizing the parameters of the rock-soil scanning image segmentation model by using the total loss; and a trained rock-soil scanning image segmentation model is obtained through a plurality of iterations. According to the rock-soil scanning image segmentation model constructed by the invention, the segmentation precision and the structure fidelity are remarkably improved.

The invention discloses a cataract fundus image automatic grading method based on semi-supervised learning, and belongs to the field of medical artificial intelligence. Characteristic parameters are extracted through multi-modal feature fusion of the eye fundus image of a patient, a semi-supervised learning neural network model is constructed, a pseudo-label training model is formed in combination with a small amount of labeled data and a large amount of unlabeled data, automatic classification of the eye fundus image of cataract is realized, and the problems of low manual classification efficiency and high labeled data acquisition cost are solved. In the preprocessing stage, texture and color multi-dimensional feature parameters of an image are extracted, and image data are obtained through dimension reduction. In the model construction stage, network training data of double convolution activation layers and double maximum pooling layers are adopted, and the model outputs all levels of probabilities to realize grading. The semi-supervised iterative training adopts a pseudo tag generation and model optimization alternating strategy, and a high-confidence sample expansion training set is screened. In the evaluation stage, the performance of the model is comprehensively measured by using multiple indexes and a visualization technology, and finally the model is deployed and applied to provide reliable support for clinical diagnosis.

The invention provides an image semantic segmentation active domain adaptation method and system based on a segmentation cutting model. The method comprises the following steps: generating a full-image mask for an image of a target domain by using a pre-trained segmentation cutting model; the images of the target domain are sampled in proportion, initial labeling is carried out on the full-image mask of each sampled image, an initial target domain labeling data set is obtained, and labeling is that a semantic category is given to a full-image mask area generated by each image; training the semantic segmentation model through semi-supervised learning; wherein in the training iteration of the semi-supervised learning, active learning is triggered according to a preset period, and an updated target domain labeling data set is obtained and used for continuous training of the semi-supervised learning. According to the method, the technical purpose of obtaining initial annotation data with complete semantics and clear structure at relatively low labor cost is achieved, and the technical problems that existing semantic segmentation training depends on large-scale pixel-level manual annotation, the annotation efficiency is low and the cost is high are solved.

The invention relates to the technical field of computer vision and artificial intelligence, and provides a remote sensing image target detection method based on improved YOLOv9s, and the method comprises the steps: obtaining a remote sensing image, and carrying out the preprocessing operation; a kernel selection feature fusion (KSFF) module is introduced into a neck network of the YOLOv9s model to replace at least one Concat structure in the original model; adding a cross-space multi-scale attention (CSMA) module in the neck network; replacing at least one SPPELAN structure in the original model with a parallel pooling feature modulation (PPFM) module in the neck network; and training, testing and evaluating the improved YOLOv9s network. Through comprehensive application of network structure improvement, data preprocessing optimization, a semi-supervised learning strategy and an advanced training method, the improved YOLOv9s model has higher detection precision in a remote sensing image target detection task.

The invention discloses a satellite image target detection method based on semi-supervised learning, and relates to the field of semi-supervised target detection methods. The satellite image target detection method based on semi-supervised learning comprises the steps of 1, constructing a data set; 2, constructing a model training framework; step 3, screening and optimizing partition density; step 4, self-adaptive loss weighting is carried out, double branches comprise supervised branches and unsupervised branches, pseudo label noise influences are obviously reduced, a partition density screening module (PDS) carries out dynamic space density analysis, gridding region division and local target statistics are adopted, low-quality pseudo labels are filtered in a self-adaptive mode, false detection noise of a sparse region is reduced, and the false detection noise of the sparse region is reduced; a target adaptive weight module (OAW) performs differentiated weighting on pseudo labels based on target geometric feature consistency, such as direction, aspect ratio difference and prediction quality multi-dimensional evaluation, and improves training contribution of reliable samples.

The invention provides a catenary dropper defect detection method based on semi-supervised learning. The catenary dropper defect detection method comprises the following steps: collecting a first catenary dropper image marked with defects and a second catenary dropper image not marked with defects; carrying out image preprocessing to obtain a processed image; constructing an unbiased teacher semi-supervised learning model, and performing model training by using the processed first overhead line system dropper image to obtain an initial model; inputting the processed second catenary dropper image into the initial model to obtain soft label data; calculating supervised loss by using the pseudo label data, and calculating unsupervised loss by using the soft label data so as to obtain a trained semi-supervised learning model; and performing defect detection on a to-be-detected overhead line system dropper image by using the trained semi-supervised learning model to obtain a detection result. According to the invention, the problems of high labeling cost and low reliability and detection precision in the prior art can be solved.

The invention discloses a semi-supervised medical image segmentation method based on momentum prototype alignment and adaptive uncertainty estimation, and aims to solve the problems of unstable feature alignment and false label noise accumulation caused by a batch effect in an existing semi-supervised learning method in a scene of scarcity of medical image annotation data. The method comprises the following steps: firstly, constructing a dual-network model based on a mean teacher architecture; secondly, designing a mixed strong and weak disturbance strategy, respectively applying weak data disturbance and strong data disturbance to the unlabeled image, respectively inputting the weak data disturbance and the strong data disturbance into a teacher and student segmentation network, and forcing the model to learn local and global robust features; secondly, a self-adaptive uncertainty estimation mechanism is introduced to dynamically screen false labels, and isolated noise is eliminated in combination with maximum connected domain analysis; meanwhile, the prototype of the current batch is calculated through uncertainty weighting, and a global category prototype library is updated and maintained through momentum. According to the method, the segmentation precision and the boundary integrity are remarkably improved under a small amount of annotated data.

The invention discloses a double-dynamic alternating consensus teacher network semi-supervised medical image segmentation method, and relates to the field of computer vision and medical image processing. The method comprises the following steps: firstly, constructing a system comprising a student network and two differentiated initialized teacher networks, predicting a label-free sample by using the double teacher networks at the same time in a training stage, obtaining a consensus degree value by calculating the pixel-level consistency of a prediction result in a foreground category, and dynamically adjusting the learning rate of the student network according to the consensus degree value, and the training intensity is adaptively controlled. Meanwhile, the activation and freezing states of the two teacher networks are controlled by adopting a periodic alternating strategy, and only the teacher network in the activation state is subjected to index moving average updating at any phase. Through a time dimension parameter decoupling mechanism and a consensus degree-based reliability feedback mechanism, confirmation bias errors in semi-supervised learning are effectively suppressed, and the segmentation precision and generalization ability of the model under the condition of low label data are improved.

The invention discloses a sandstone slice image component identification method based on a GL-SLIC algorithm and semi-supervised learning. The method comprises the following steps: firstly, acquiring a sandstone slice image amplified by 200 times under single polarization, converting an RGB image into a CIE-Lab color space, and extracting a GLBP texture feature vector; then a GL-SLIC superpixel segmentation algorithm is constructed, and adaptive segmentation based on texture and color features is realized in combination with a Gabor filter and a local binary pattern; then implementing a region merging algorithm to generate complete mineral particles and pore fragments; finally, a classifier based on VGG16 and a discriminator based on ResNet18 are constructed, a semi-supervised self-training framework is adopted, a model is initialized by using about 6% of manual labeling samples, a high-confidence-coefficient pseudo-label extension training data set is generated through iteration, and automatic recognition of sandstone components such as quartz, pores, kaolinite, rock debris and a matrix is achieved. According to the method, the recognition accuracy of 96.3% on a test set is achieved, compared with a traditional method, the segmentation precision and the recognition accuracy are remarkably improved, the data labeling cost is greatly reduced, and an efficient technical scheme is provided for automatic analysis of geological images.

The invention belongs to the technical field of medical image processing, and particularly relates to a bone layer recognition method and system based on semi-supervised learning. The method specifically comprises the steps of network model construction: constructing a semi-supervised learning network model which comprises a student-teacher branch model, a student model being an encoder-decoder structure, a teacher model being an encoder-decoder structure, and the teacher model being used for performing exponential moving average EMA weight updating on the student model; a spine CT image X1 with a label and a spine CT image Xu without a label are collected, double-flow disturbance is carried out on the spine CT image Xu without the label, and weak disturbance label-free data Xw and strong disturbance label-free data Xs are obtained; network model training: inputting the spine CT image X1 with the label, the weak disturbance non-label data Xw and the strong disturbance non-label data Xs into a network model, and performing model training; and bone layer identification: segmenting the spine CT image data by using the trained network model.

The disclosure provides an adaptive self-learning method and an adaptive self-learning system. The adaptive self-learning method includes steps of inputting a first complex model and unlabeled data to an adaptive semi-supervised learning module and performing a pre-semi-supervised learning module to generate an average precision variation. If the average precision variation does not satisfy a condition value at any one time out of an inference count, the semi-supervised learning module is performed. After performing the semi-supervised learning module, the steps include performing a self-learning module for refining the target model, and then the trained target model is disposed to a site device. The site device deploys the trained target model to perform an object detection procedure.

The invention discloses a model training method and device, a target detection method and device, equipment and a medium, which are used for improving the performance of a detection model. According to the method, an attribute enhancement graph and a structure enhancement graph are constructed based on multiple views; based on the graph features with the label nodes in the attribute enhancement graph and the graph features with the label nodes in the structure enhancement graph, determining the graph features after weighted fusion of the different view features; inputting the graph features into a classifier, and determining classification loss based on a classification result; based on the image features after weighted fusion, attribute feature reconstruction of label nodes of the attribute enhancement image and structural feature reconstruction of label nodes of the structure enhancement image are carried out, and feature reconstruction loss is determined; regularizing the model based on the graph features of the label-free nodes of the attribute enhancement graph and the graph features of the label-free nodes of the structure enhancement graph, and determining semi-supervised learning loss of the label-free nodes based on a regularization result; and based on each loss, determining the joint training target loss of the model.