Training method and device of survey model of remote sensing image target, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a training method and device for a survey model of a remote sensing image target, equipment and a medium, and the method comprises the steps: employing a preset number, obtaining training samples from a training sample set, and obtaining a target training sample subset; adopting a semi-precision training strategy, and training an initial model according to the target training sample subset, wherein the initial model is a model obtained based on a Resnet network; and repeatedly executing the steps of acquiring the training samples from the training sample set by adopting a preset number to obtain a target training sample subset until a preset training ending condition is met, and determining the initial model meeting the preset training ending condition as the survey model of the remote sensing image target. The semi-precision training strategy is applied to the training stage of the survey model of the remote sensing image target, the semi-precision floating-point number is used for accelerating training under the condition that the precision loss is reduced as much as possible, the time consumed by calculation is shortened, and occupied resources are reduced.

Description

technical field [0001] This application relates to the technical field of artificial intelligence, in particular to a training method, device, equipment and medium for a remote sensing image target survey model. Background technique [0002] In order to avoid the loss of neural network calculation accuracy, the traditional remote sensing image target survey model is constructed with a parameter accuracy of FP32 (that is, a 32-bit floating point number), and its value ranges from 1.4x 10-45 to 3.4x 1038. Because the FP32 value occupies a large number of bytes, it takes significantly longer to calculate the neural network than a floating-point number with a lower precision (for example, a 16-bit floating-point number), and the resource usage is also significantly larger than a floating-point number with a lower precision. Contents of the invention [0003] The main purpose of this application is to provide a training method, device, equipment and medium for the survey model ...

AI Technical Summary

Problems solved by technology

[0003] The main purpose of this application is to provide a training method, device, equipment and medium for the survey model of remote sensing image targets, aiming to solve the problem of parameter accuracy used in order to avoid the loss of calculation accuracy when constructing the survey model of remote sensing image targets in the prior art It is a 32-bit floating-point number, which leads to a technical problem that the calculation time is significantly longer than that of a floating-point number with a lower precision, and the resources occupied are also significantly larger than that of a floating-point number with a lower precision.

Images