Model training method, multi-source multi-target data association method and device

Abstract

The present invention provides an end-to-end deep neural network association model training method and device, a multi-source and multi-target data association method, device, electronic equipment and storage medium. The specific implementation method includes: acquiring the historical sample data sequence of the first radiation source, the current sample data of the first radiation source, and the label of the current sample data of the first radiation source; using an end-to-end deep neural network association model based on the historical sample data of the first radiation source sequence and the current sample data of the first radiation source, obtain the correlation probability of the current sample data of the first radiation source and the historical sample data sequence of the first radiation source; calculate the correlation loss function according to the correlation probability and the label; and adjust according to the correlation loss function End-to-end deep neural network correlation model parameters.

Description

technical field [0001] The invention relates to the technical field of passive detection, in particular to an end-to-end deep neural network association model training method and device, a multi-source and multi-target data association method, device, electronic equipment, storage medium and computer program product. Background technique [0002] In the field of passive detection, target association of multiple radiation sources can be achieved based on traditional clustering algorithms or target association models. However, the research on the correlation of passive detection radiation sources based on the traditional clustering algorithm is usually suitable for the scene where the electromagnetic characteristic parameters do not change or change little, which leads to the problem of incorrect correlation and leakage of radiation sources in complex electromagnetic environments. The traditional target correlation deep neural network model mainly uses target motion parameters...

AI Technical Summary

Problems solved by technology

However, the research on the correlation of passive detection radiation sources based on traditional clustering algorithms is usually applicable to scenarios where the electromagnetic characteristic parameters do not change or change little, resulting in the problem of wrong correlation and missing correlation of radiation sources in complex electromagnetic environments.

The traditional target association deep neural network model mainly uses the target motion parameters to realize the association, and does not apply the electromagnetic characteristic parameters, and usually separates the state prediction process from the target association process, and there are problems of feature details being ignored and error accumulation, and the two processes The training objectives are not consistent, resulting in low model association accuracy

Images