6 results about "Projection space" patented technology

This invention proposes a structurally enhanced discriminative width learning system for image classification. First, it introduces graph regularization based on class relationships to ensure that similar samples remain closely distributed in the projection space. Second, it adds intra-class divergence regularization to further enhance the aggregation of samples within the same class. Furthermore, it introduces inter-class divergence regularization to ensure that the centers of different classes are far apart in the feature space. Finally, in the output layer regression stage, it replaces the Frobenius norm regularization in ridge regression with sparse norm regularization to effectively suppress noise and redundant information. This image classification scheme, while maintaining the simplicity and efficiency of the optimization process, organically integrates three different types of structured information—intra-class structure, inter-class structure, and local manifold information—into a unified model, enabling the learned feature representations to possess stronger structure preservation and discriminative capabilities.

The invention discloses a multi-modal BEV perception consistency alignment checking method based on language structure prior, and relates to the technical field of automatic driving, and the method comprises the steps: obtaining multi-modal sensor data, and mapping the multi-modal sensor data to a virtual projection space; and based on the large language model, generating language structure priori in a structured text form, and mapping the language structure priori to the virtual projection space to generate a structure constraint field. In the training stage, based on a structure constraint field, weak supervision consistency loss is constructed, an evidence gating mechanism is introduced, loss weight is dynamically adjusted according to an evidence intensity score, and joint optimization is carried out on the network. In the reasoning stage, consistency checking is carried out on BEV sensing results by utilizing a structure constraint field, and confidence reestimation, hallucination suppression and open set rejection are carried out in combination with a structure conflict score and an evidence intensity score. The problems of structure default and high-confidence illusion under the weak evidence working condition are solved, and the physical consistency and reliability of the sensing result are improved.

A computer-implemented method for profile detection in a discrete 3D model. The discrete 3D model represents a mechanical part. The method includes, for each 3D surface of at least one 3D surface of the 3D model, providing a candidate parameterization of the 3D surface and a candidate projection of the 3D surface into . The method further includes computing a function. The function penalizes, for each couple of points of the 3D surface having neighboring parameter values, a distortion. The distortion is between a disparity between the projection of the points, and a disparity between the parameterization of the points. The method further includes determining that the candidate parameterization and the candidate projection form a valid profile of the 3D surface if the computed function is smaller than a predefined threshold.

The application discloses a small sample detection method based on an effective receptive field, comprising the following steps: 1) based on a query image and a support set of each category, a pre-trained visual basic model is used to extract visual features and construct a category projection space; 2) based on the query image, a pre-trained region generation network is used to generate a candidate box set in combination with the category projection space; 3) based on the candidate box, an effective receptive field support region is constructed, a category prediction result and a bounding box regression result are obtained, and a support effective receptive field region prediction model is trained; 4) in an inference stage, the effective receptive field prediction model is used to screen an optimal support region, and a final detection result is output. According to the application, a more suitable context receptive region can be adaptively screened according to a candidate target, the accuracy and stability of target classification and positioning in a small sample scene are improved, and thus the generalization ability of new category target detection and the practical application reliability are improved.

The application discloses a kind of adaptive radiotherapy dose consistency fast verification methods for TOMO system, it is related to medical information technology field, especially the field of TOMO radiotherapy information processing.The method comprises the following steps: obtaining the flux distribution data in original radiotherapy plan and original region of interest;According to the beam geometry parameter of original radiotherapy plan, original region of interest is projected to flux projection space, and mapping relationship is established as reference index;Obtain the target region of interest in adaptive radiotherapy stage;Based on flux distribution data and target region of interest, update mapping relationship, and mapping relationship is used as target index;Detect the data consistency between target index and reference index, determine the dose distribution change information of target object.The present application solves the technical problems that the adaptive radiotherapy dose consistency verification in the prior art must rely on high-load dose calculation engine to carry out data reprocessing, resulting in large amount of calculation data, high system response delay, and unable to meet the timeliness requirements of online verification.

The application provides a kind of spaceborne SAR and electronic reconnaissance signal data fusion method, comprising: obtaining SAR image using spaceborne SAR system, detecting SAR image, obtaining target rough detection result;Obtain spaceborne electronic reconnaissance radiation source detection result;Complete SAR data and electronic reconnaissance data time and space alignment;Using load multi-attribute to construct multi-attribute feature extraction network, the feature of SAR detection data and electronic reconnaissance detection data is modeled to obtain feature fuzzy vector, and the feature that can be used is extracted;According to the space relationship of load, determine the projection space of SAR data and electronic reconnaissance data feature matrix, carry out feature mapping, extract effective information in feature, remove redundant information, and strengthen target multi-attribute feature;The strengthened feature is clustered, false alarm is eliminated, detection is completed, and target feature information after target fusion is obtained, and spaceborne SAR and electronic reconnaissance data fusion detection are completed.