17results about How to "Fully trained" patented technology

The invention discloses a cigarette micropore target detection method and system, and belongs to the technical field of cigarette quality detection, and the method comprises the steps: carrying out the three-dimensional point cloud data collection of cigarettes of a plurality of brands, carrying out the filtering processing of collected cigarette punching original point cloud images, and removing useless background information; fusing feature information of different levels of cigarette punching original point cloud images, introducing a data adjustment function to fully train hole data in the feature information, and obtaining an optimized PointNet + + deep network model; and extracting point cloud data of the cigarette punching area, segmenting the whole predicted point cloud image according to the feature information of the distance between points, mapping the connection and edge contour of all points of the single hole into a two-dimensional image, and obtaining the maximum depth value of the to-be-detected hole. According to the method, the perforation segmentation and perforation depth calculation of cigarettes of various brands are realized, and a more efficient, convenient and reliable detection scheme is provided for cigarette production.

An adaptive virtual synchronous machine control method based on deep reinforcement learning belongs to the field of power grid operation control. It establishes a photovoltaic storage grid-connected model containing photovoltaic systems, battery energy storage and variable loads; uses virtual synchronous generator control to achieve complementary photovoltaic output power for the battery energy storage system; in order to solve the problem of microgrid frequency deviation caused by microgrid load power changes, a deep reinforcement learning algorithm is used to control and adjust the virtual inertia and damping coefficient of the virtual synchronous generator; the double deep Q learning algorithm in deep reinforcement learning is used to control the virtual inertia and damping coefficient; a double-layer evaluation network is constructed, combined with a reward function, and after multiple trainings, the adaptive adjustment of the virtual inertia and damping coefficient is achieved. It uses deep reinforcement learning technology to adaptively control the virtual inertia and damping coefficient, suppress the frequency deviation, accelerate the recovery of the system operating frequency, and improve the stability of the system.

The invention relates to a multi-mode-based dynamic facial expression recognition method, which comprises the following steps of: acquiring an initial video sample, and acquiring an image sample sequence according to the initial video sample; inputting the image sample sequence into a dynamic facial expression recognition model, extracting a facial expression feature sequence from the image sample sequence, obtaining a feature vector embedded in a facial expression class, and matching text description according to the embedded feature vector of the facial expression class to obtain a text feature vector; similarity calculation is carried out on the facial expression feature sequence and the text feature vector, a facial expression recognition result of the initial video sample is obtained, a dynamic facial expression recognition model is constructed based on a CLIP model and is obtained based on training of a training set, and the training set comprises a historical image sample sequence and corresponding facial expressions. According to the method, the context relation between the visual class and the text description is analyzed, and finally, the dynamic facial expression recognition is completed through similarity calculation, so that the training speed and the facial expression prediction precision are improved.

The present invention discloses a method for mining difficult samples of remote sensing images. The method pre-trains a semantic segmentation module including two semantic segmentation models, inputs the image to be interpreted into the semantic segmentation module, obtains a first semantic segmentation result and a second semantic segmentation result, and then obtains a common semantic segmentation result, a full semantic segmentation result and an intersection-and-union ratio through intersection calculation, union calculation and intersection-and-union ratio calculation. Finally, by comparing the intersection-and-union ratio and a preset IOU threshold, the samples with an intersection-and-union ratio less than the preset IOU threshold are recorded as difficult samples, and the samples with an intersection-and-union ratio greater than the preset IOU threshold are recorded as simple samples. The difficult samples are manually labeled, the simple samples are manually quality-checked, and added to a sample library. The method obtains different semantic segmentation results for the same image through two different semantic segmentation models, and distinguishes simple samples from difficult samples by calculating the intersection-and-union ratio of the segmentation results. Simple samples and manually labeled difficult samples can expand the sample library.

The present invention discloses a method for reconstructing simultaneous multi-layer imaging signals. The reconstruction method comprises: generating virtual automatic calibration signals of multiple layers according to the acquired automatic calibration signals of multiple layers, wherein the automatic calibration signals of multiple layers and the virtual automatic calibration signals are both complex data of k-space; generating a training data set according to the automatic calibration signals of multiple layers and the virtual automatic calibration signals; training a neural network model using the training data set; inputting the acquired original aliased data of multiple layers into the trained neural network model to obtain corresponding network output data; obtaining complete reconstruction data of k-space according to the fusion of the original aliased data and the network output data; and generating a reconstructed image according to the complete reconstruction data. The phase information of the virtual automatic calibration signal is fully utilized to more fully train the neural network model, and the virtual signal is used as another training data set to obtain a higher quality reconstructed image without increasing the extra scanning time.

The invention provides a clinostatic head and neck muscle strength training device. The clinostatic head and neck muscle strength training device comprises a training bed, the supporting mechanism is arranged outside the bedside end of the bed board; the head and neck training mechanism is arranged at the upper end of the supporting mechanism and comprises a helmet and a fixing belt, and the end, away from the bed board, of the helmet is connected with the supporting mechanism through a universal ball joint; the four-limb exercising mechanism comprises handles arranged on the two sides of the upper surface of the bed board in a sliding mode and two pedal plates arranged on the upper surface of the tail end of the bed board in a sliding mode respectively, the handles and the pedal plates are connected with the outer edge of the helmet through pull ropes, and the pull ropes are connected with four different positions on the outer circumference of the helmet respectively. The head of a person can be driven to do actions such as head raising, head lowering and lateral bending through four-limb exercise, so that neck muscle exercise is achieved, and the device is suitable for exercise of patients or sub-healthy people in all stages and particularly suitable for patients needing to do clinostatism exercise in the initial training recovery stage; the device has the advantages of whole-body exercise, blood circulation promotion, recovery acceleration, high interestingness, easy persistence of patients and good training effect.

The application discloses a training method and system for improving the honesty of a large model with a small amount of labeled data, and a lightweight network is newly added to assist the LLM in self-labeling data, thereby significantly expanding the data scale available for model training on the basis of limited labeled data, so that the model can still be fully trained and optimized under the condition of limited data resources. A close cooperative working mechanism is formed between the lightweight network and the LLM, the lightweight network generates reliable labeling information for unlabeled data by performing honesty evaluation on the answers generated by the LLM, this process not only fully utilizes the probability evaluation ability of the LLM itself, but also combines the quantitative evaluation of the lightweight network on the honesty, so that the model can learn a more honest answer strategy under the condition of limited data, and the purpose of enhancing the honesty of the LLM by using a small amount of data is achieved.

The present invention discloses a federated learning method and system based on image privacy protection of a large model computing network, obtains node information of an image computing network and node processing image data information, divides nodes into passive nodes and active nodes according to the node information and node processing image data information, and prevents active nodes from inferring feature information of passive nodes through intermediate results of passive nodes and global module gradients of passive nodes by encrypting and protecting the intermediate results of passive nodes and global module gradients of passive nodes; and prevents passive nodes from inferring labels of active nodes through the gradients of total outputs of active nodes by encrypting and protecting the gradients of total outputs of active nodes. The present invention can solve the technical problems of malicious passive nodes stealing labels of active nodes through global modules or gradients and active nodes stealing features of passive nodes through the outputs of global modules of passive nodes.

The invention relates to the technical field of network security, and discloses an unknown attack-based network defense method, which comprises the following steps of: dividing according to tuple information of a network data packet to obtain a network data stream; aggregating communication nodes of the network data packet to obtain a data communication channel; performing feature statistics and original byte extraction on the data stream to obtain a preprocessed network stream, and enhancing the preprocessed network stream to obtain an enhanced network stream; optimizing network parameters through a joint loss function to obtain an optimized feature vector; extracting feature vectors by freezing parameters of the triple network, and training a single classifier to obtain a trained single classifier; and based on the feature vector, judging a preset test sample to be benign or malicious, and when the test passes, obtaining a trained attack distinguishing model, and performing unknown attack detection. The invention further provides a network defense device and equipment based on the unknown attack and a storage medium. According to the method, the detection capability of unknown network attacks can be coped, and the security of network protection is improved.

The application discloses a network acquisition method and device, electronic equipment and a computer readable medium, and is applied to the electronic equipment. The method comprises the following steps: acquiring an initial super network, wherein the initial super network comprises at least one sub network; searching for a target sub network meeting a specified operation requirement from the at least one sub network; and training the target sub network based on pre-acquired training data to update the initial super network. The application searches for and trains the target sub network meeting the specified operation requirement, so that the target sub network can be fully trained, and the trained sub network is gathered under the given constraint condition as much as possible, thereby avoiding excessive dispersion of the sub network.

The present application provides a method, training method, device, equipment and medium for identifying abnormal tooth areas. The method for identifying abnormal tooth areas includes: obtaining the texture in the first tooth model to be identified and the geometric information of the three-dimensional points in the first tooth model; projecting the texture in the first tooth model and the geometric information of the three-dimensional points in the first tooth model respectively to obtain the texture map and geometric information map of the first tooth model; superimposing the texture map and geometric information map of the first tooth model into a first target image; using a pre-trained abnormal tooth area recognition model to perform abnormal tooth area recognition processing on the first target image to obtain an abnormal area recognition result corresponding to the first target image. The efficiency and accuracy of abnormal tooth area recognition are improved.

The invention discloses a remote sensing image water body extraction method and system based on an RGB + X data view angle, and relates to the technical field of computer vision and remote sensing. The remote sensing image water body extraction method and system based on the RGB + X data view angle mainly comprises the following steps: extracting an X modal feature image according to red, green, blue and near-infrared bands of a remote sensing image; a dual-complexity backbone network and a hybrid fusion module are utilized to extract multi-modal, multi-scale and global features, a multi-scale MLP decoder is utilized to obtain a water body segmentation result, and up-sampling is performed to obtain a water body extraction result of original image resolution. According to the remote sensing image water body extraction method and system based on the RGB + X data view angle, the precision and generalization ability of water body extraction can be improved.

The present application discloses a method, device, equipment and medium for assigning initial values ​​to the supply side of a gas network, which relates to the field of smart city operation technology, including: obtaining the current predicted gas consumption and the accumulated duration of historical data on the demand side of the gas network; if the accumulated duration belongs to the first preset duration interval, the current supply data of the supply side is filtered out from the historical data using comprehensive similarity; if the accumulated duration belongs to the second preset duration interval, the current supply data of the supply side is obtained based on the predicted characterization information of the current predicted gas consumption obtained using the target characterization extractor; if the accumulated duration belongs to the third preset duration interval, the current supply data of the supply side is obtained using the target supply data prediction model, the current predicted gas consumption and the predicted characterization information; the duration of the first preset duration interval is less than the duration of the second preset duration interval, which is less than the duration of the third preset duration interval. The above scheme is used to reasonably determine the data on the supply side of the gas network.

The invention relates to the field of data processing and artificial intelligence, in particular to an agricultural product purchase behavior prediction method and system based on a graph neural network. According to the technical scheme, the agricultural product purchase behavior prediction method and system based on the graph neural network comprise the agricultural product purchase behavior prediction method based on the graph neural network; compared with a traditional agricultural product purchase behavior prediction method based on a graph neural network, the method adopts a traditional graph neural network model to carry out feature extraction and relation modeling, and generally, training is directly carried out on a large-scale graph, so that the calculation and storage requirements are too high, the training efficiency is low, and the prediction efficiency is high. According to the agricultural product purchase behavior prediction method based on the graph neural network, a multi-scale variational graph auto-encoder is introduced, local and global structure features of the graph can be captured more comprehensively, node representation richness and diversity are improved, prediction accuracy is improved, calculation and storage requirements are reduced through a static sub-graph segmentation method, and the prediction efficiency is improved. And the training process is accelerated.

The invention provides a self-training learning method and system for medical image segmentation and electronic equipment, and relates to the technical field of image processing. The method comprises the following steps: firstly, acquiring original training including benign lesion segmentation training data and malignant lesion segmentation training data, and a test data set; then, equally dividing the training data set, and combining the training data set with the test data set for iterative training to form a plurality of groups of data sets; then, executing a training process by utilizing each group of data set: taking benign data as a main part in the initial stage, gradually increasing the proportion of malignant focus segmentation training data, and training until convergence to obtain a first segmentation model; and predicting a test set by using the model to obtain a focus prediction image. And by comparing the prediction image with the annotation image, generating a false label image of the leak detection area and the false detection area. And adding the pseudo label images into a training set, and training the first segmentation model again until convergence. According to the technical scheme, the segmentation precision and generalization ability of the medical image segmentation model can be improved.

The invention discloses a multi-sight-distance deep learning track line detection method based on global features, and the method comprises the steps: 1, installing a far-focus camera and a near-focus camera at the head or side surface of a train, obtaining track line images in different scenes in the train operation process, and constructing a data set; 2, in image preprocessing, a Sobel operator is used for calculating gradient energy to generate a focusing degree image, a decision diagram is generated through multi-scale morphological optimization operation, and smooth transition of fused image boundaries is achieved through Gaussian weighted fusion; step 3, constructing an improved ResNet18 backbone network, fusing multi-scale features, combining row selection and structure perception loss, and performing Adam optimization training; and 4, real-time track line detection is carried out. According to the invention, through multi-sight-distance image fusion and global feature extraction, technical support is provided for realizing high-precision and real-time detection of a train track line, automatic train driving, track state monitoring and other key tasks.

The embodiment of the present disclosure relates to a network search method and device, electronic equipment and storage medium, and relates to the technical field of network search. The network search method comprises the following steps: obtaining an intermediate network by training in a super network, and training the intermediate network based on a kernel size to obtain a trained intermediate network; performing depth search training on the trained intermediate network by determining a learning rate and an adaptive search module for adjusting a depth selection space to obtain a depth network; and performing width search training on the depth network by determining a learning rate and an adaptive search module for adjusting a width selection space to perform network search. The technical solution in the embodiment of the present disclosure can improve the accuracy of the network.