10 results about "Instance selection" patented technology

The invention discloses a single-stage full-sparse three-dimensional target detection method based on point cloud completion, and the method mainly comprises the steps: firstly, processing a point cloud through a sparse backbone network to extract the features of each point, and then inputting the features into an instance selection module to generate clustering instances; processing each clustering instance by an alignment-based point cloud completion module, and finally generating a complete point cloud; further, point features are input to a center point voting network, where each point predicts its perceived object center. The predicted center point and complemented point cloud features are fused at a feature level, and finally the generated features are sent to a detection head to obtain a 3D bounding box and a corresponding confidence score. The method is suitable for a single-stage full-sparse three-dimensional target detection network, and can improve the detection performance of a single-stage full-sparse three-dimensional target detector for severely shielded objects.

The invention discloses a power distribution network fault rapid positioning method and system. The method comprises the steps of data acquisition, triple undersampling processing, secondary decomposition, dynamic similar distance matrix calculation and condensed hierarchical clustering. The invention belongs to the technical field of power distribution networks, and particularly relates to a power distribution network fault rapid positioning method and system.According to the scheme, triple undersampling processing is carried out on original data based on unmarked data injection, instance selection and neighborhood cleaning; a residual term is secondarily decomposed into an intrinsic mode function and a residual, so that data can be more finely disassembled, and more signal components and features are extracted; a dynamic similarity distance calculation method is adopted to more accurately describe the similarity between the two time sequences; the dynamic similarity distance matrix is adopted for hierarchical clustering, so that dynamic similarity information between data can be fully utilized; fault positioning is carried out based on double tags, and the positioning speed and accuracy are improved.

Process knowledge creation, development, and management techniques allow for and enable the creation of a universal process definition (UPD) of an industrial process, the automatic conversion or transformation of the UPD into different site-specific process definitions, and the implementation of the site-specific process definitions at different manufacturing, production, and / or automation sites. Typically, the UPD is site- and equipment-agnostic, and the transformation may generate and provide a set of site-specific process definition implementation files or routines to configure and / or govern the behavior of various site-specific execution systems, e.g., as site-specific operational instances of the UPD. The techniques may utilize feedback and information generated by site-specific operational instances to generate learned knowledge and update the UPD accordingly so that subsequent instantiations of the UPD may incorporate (and reap the benefits of) the learned knowledge. The techniques may automatically select a most suitable site for a particular instantiation of the UPD.

The invention discloses a low-illumination target detection method based on class consistency screening and generative completion. The method comprises the steps of 1, performing feature extraction on a low-illumination input image, and outputting classification, positioning and centrality results; 2, class consistency screening is executed according to the instance-level basic features, and high-quality representative instance samples are obtained; step 3, constructing a category orthogonal space by using the instance sample and a vision-language model to form enhanced category-level semantic features; 4, generating corresponding phantom features based on the memory bank, and providing the phantom features and the instance samples to the step 5 for use; 5, multi-layer domain alignment is executed at the image level, the instance level and the category level, and unified optimization is completed; and 6, performing open set rejection judgment on the low-confidence and low-affinity targets in a reasoning stage, and outputting a final detection result. According to the method, three key problems of high-quality instance selection, cross-domain category discrimination enhancement and missing category completion are cooperatively solved on the premise of not depending on target domain labeling and not increasing the overhead of a reasoning stage.

The invention discloses a face image detection method based on distributed optimization instance selection. The face image detection method is applied to a work and coordination node network. A work node carries out multi-objective evolutionary optimization with classification precision and compression rate as targets on a local sample; the coordination node aggregates the population by adopting direct splicing or based on a conflict instance strategy according to an iteration process and executes global evolution; variation parameters are adjusted by counting the instance frequency of the Pareto optimal solution set, and feedback information is generated to guide subsequent iteration of the working nodes; and finally, selecting an optimal SVM model for detection. Through a distributed coevolution and feedback mechanism, redundant samples can be effectively eliminated under the condition that data are not concentrated, and the recognition efficiency and classification accuracy of face detection are improved.

This invention discloses a fault diagnosis method and apparatus that integrates alarm and trend event data. The method includes the following steps: preprocessing the raw time-series data of an industrial system to obtain preprocessed data; extracting multi-value alarm events and multi-class trend events, and fusing the two types of feature events to form a hybrid sequence; using the Word2Vec core model to vectorize and encode the events in the hybrid sequence; constructing an LSTM model; and employing an instance transfer learning strategy to update the model trained using source domain data to meet the needs of target domain diagnosis, thus achieving cross-domain fault diagnosis. This invention, by simultaneously extracting alarm and trend events, can more comprehensively capture the evolution of faults; through instance selection and weight adjustment, it can effectively solve the problem of scarce target domain data and improve diagnostic accuracy; the method can be extended to complex systems such as shipbuilding, petroleum, and chemical industries, and is compatible with different industrial scenarios. It can be adapted to new systems by adjusting the event definitions, making it widely applicable.

The application discloses an instance selection algorithm based on local Coulomb force, and relates to the technical field of data preprocessing, which comprises the following steps: calculating the local density of each data point in a given data set; mapping the calculated local density to the amount of electric charge; calculating the local resultant force of each data point on its reverse neighbors based on a Coulomb force model; defining and calculating the change rate difference of the local resultant force; detecting boundary points based on the change rate difference; selecting the detected boundary points and performing random sampling; and using the selected boundary points and the randomly sampled data to jointly construct a compressed training set. The application quantifies the change rate difference of the force of the data points by using the Coulomb force model, accurately identifies the boundary points, and fuses the random sampling to construct a compressed training set with high quality and high robustness, thereby significantly improving the data efficiency.

The invention discloses a model reasoning service judgment system and method based on an intention engine, which are used for sequentially completing service mode selection, model instance selection and routing execution, result output and closed-loop updating of strategy parameters by taking request characteristics and a system resource state as input for each reasoning request. According to the method, mechanisms such as request-level dual-mode switching, semantic and resource joint judgment, multi-dimensional result judgment and metadata-based closed-loop learning are integrated in the same edge inference system, so that the edge multi-model inference system can consider inference quality and resource utilization efficiency while ensuring a service level target.

In a computing-aware traffic steering (CATS) network, such as those conforming to an IETF proposal, a service contact instance for a requested service is selected based on at least cache metrics associated with data cached at one or more egress nodes of the CATS network. In some embodiments, the selection is also based on compute and network metrics. In one implementation, the selection is based only on cache metrics unless that selection is too costly in terms of compute and / or network load. In that case, the selection is based on the compute and network metrics. In this way, cache metrics are prioritized over compute and network metrics as long as the cache-based selection is not too costly.

This invention discloses a noisy data classification method based on random sampling clustering, comprising: acquiring noisy data to be classified; performing random sampling processing on the acquired noisy data, selecting several processed data to construct a partial centroid set; performing iterative processing on the acquired noisy data, selecting several processed data to add to the constructed partial centroid set, constructing a centroid set; constructing weighted instances for the constructed centroid set; and selecting a weighted noise method to complete the classification of the noisy data. This invention also discloses a user classification method including the aforementioned noisy data classification method based on random sampling clustering. By processing the noisy data to be classified through random sampling, selecting an iterative processing method, and employing stratified sampling, the noisy data classification is achieved. Moreover, this invention has high classification accuracy, strong reliability, good practicality, and low algorithm complexity.