4 results about "Adaptive value" patented technology

The application discloses a sardine optimization solving method and system and a storage mechanism, relates to the technical field of environmental monitoring, and comprises the following steps: acquiring a monitoring range of a sensor, obtaining a current position of each sardine according to the monitoring range and a preset random function; calculating an adaptive value of each sardine according to the current position of the sardine population, and obtaining a first optimal position of the sardine population; calculating according to a current iteration number and a preset iteration number, and obtaining a current physical strength of each sardine; determining a current stage of the sardine according to the current physical strength; calculating according to the current stage, the current physical strength, the current position and the first optimal position, and obtaining an updated position of each sardine; taking the updated position as an updated current position, recalculating the first optimal position according to the updated current position, and updating the updated position. The application can consider the influence of the iteration number on the updated position, and improve the accuracy of the algorithm.

The invention discloses a self-adaptive K value undersampling method based on multi-modal density perception, and belongs to the field of machine learning data preprocessing. The method comprises the following steps: constructing a shape sensing kernel function, and converting an Euclidean distance into a Mahalanobis distance through a k-nearest neighbor covariance matrix, so that a kernel shape adapts to a data cluster distribution form; calculating density estimation of each scale by adopting a multi-bandwidth shape sensing kernel, and realizing multi-scale fusion based on index weight distribution of global density deviation; constructing a noise suppression weight in combination with a local outlier factor to correct density estimation; and executing two-stage search: marking potential overlapping examples by taking majority classes as anchor points, and dynamically selecting klarge or ksmall values for secondary judgment according to density comparison of the majority classes and minority classes. The problem of insufficient precision caused by a fixed k value and a single Gaussian kernel is solved through a multi-modal density sensing framework, accurate positioning of overlapped instances under complex distribution and optimal reservation of majority class information are realized, and the classification performance of class imbalance data is remarkably improved.

This invention relates to the field of gear surface defect detection technology, and more particularly to a cross-domain gear surface defect detection method and system. To enrich the diversity of training samples and reduce background interference, an implicit information fusion strategy based on Fourier decoupling is proposed. By retaining invariant phase information and weakening amplitude information, the model can learn more robust semantic features. Addressing the difficulty in bridging inter-domain distribution differences, a domain adversarial adaptation framework is constructed to reduce global domain offset, and a domain-aware attention mechanism is designed to guide the model to focus on semantic regions with cross-domain adaptive value. Based on global alignment, the maximum mean difference of class prototype labels is designed, further achieving more refined class-level conditional alignment on top of global adversarial alignment. This invention effectively improves the detection accuracy of surface defect detection models for gear surface defects in unsupervised cross-domain scenarios.

The invention provides a model training parameter determination method and device, equipment, a product and a storage medium, and relates to the technical field of model training.The method specifically comprises the steps that a training parameter sequence of each first model in a first model set is disturbed, and a first disturbance training parameter sequence of each first model in the first model set is determined; based on the first disturbance training parameter sequence of each first model in the first model set, the training precision of each first model is determined; performing multiple disturbance operations on a first disturbance training parameter sequence of a first model with the maximum training precision in the first model set, and determining a second model set; determining an adaptive value of each second model in the second model set; and determining the training parameter of the second model with the maximum adaptive value in the second model set as a model training parameter. According to the method, the training parameters with excellent performance can be efficiently determined to improve the training precision of the large model.