Unlock instant, AI-driven research and patent intelligence for your innovation.

Model migration training method for semantic segmentation of remote sensing image

A technology of semantic segmentation and remote sensing images, applied in the field of remote sensing images, can solve problems such as poor prediction results, waste of public resources, and difficulty in discovery, and achieve the effects of improving capabilities, reducing waste, and expanding data sets

Pending Publication Date: 2021-11-16
江苏天汇空间信息研究院有限公司
View PDF3 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Most of the existing technologies re-label the data set to make a training set for retraining from the beginning, which wastes available public resources; or use multi-model prediction results to fuse and make a training set for retraining, when dirty data or wrong samples appear , it will be difficult to find the problem and lead to poor prediction

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Model migration training method for semantic segmentation of remote sensing image
  • Model migration training method for semantic segmentation of remote sensing image
  • Model migration training method for semantic segmentation of remote sensing image

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0081] Embodiment 1: as Figure 2-4 As shown, image data is trained through the following steps:

[0082] D01: Obtain public image datasets;

[0083] D02: Preprocess the data in the public image data set, extract part of the data in the public image data set, and form the first training data set;

[0084] D03: Train on the first training data set to obtain a pre-training model;

[0085] D04: Preprocessing the image data in the first training data set to form a prediction data set;

[0086] D05: Predict the effect of the data in the prediction data set through the pre-training model, and obtain the prediction result;

[0087] D06: According to the prediction effect, judge the classification accuracy, process and output the prediction results in the prediction data set.

[0088] by putting Figure 7-Figure 10 By comparison, it can be determined that Figure 10 The classification of the blue water body is accurate, and the rose red is the wrong classification of the unused ...

Embodiment 2

[0090] Example 2: In Figure 4 Among them, W01 and W02 refer to the green cultivated land, and W03 refers to the rose-red unused land;

[0091] like Image 6 , the sample marker 1 refers to the marked red area, 2 refers to the marked blue area, and 3 refers to the marked yellow area.

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a model migration training method for remote sensing image semantic segmentation. The method comprises the following steps: D01, obtaining a public image data set; D02, extracting and preprocessing part of image data in the public image data set to form a first training data set; d03, training the first training data set to obtain a pre-training model; D04, preprocessing image data in the to-be-predicted image data set to form a prediction data set; D05, performing effect prediction on the image data in the prediction data set through the pre-training model to obtain a prediction result; and D06, according to the prediction effect, the classification accuracy is judged, and the prediction result in the prediction data set is processed and output. Data are extracted from a public image data set, public data are fully utilized, and waste of resources is avoided. An image data set is expanded in a targeted mode, and the model transfer learning speed can be increased; through the training model, a data set with a poor effect in the remote sensing image can be observed, and the capability of identifying the data set with the poor effect is improved.

Description

technical field [0001] The invention relates to the technical field of remote sensing images, in particular to a model migration training method for semantic segmentation of remote sensing images. Background technique [0002] Remote sensing images refer to films or photos that record the size of electromagnetic waves of various ground objects. They are mainly divided into aerial photos and satellite photos. Remote sensing images are multi-source remote sensing data based on a unified geographic coordinate system to generate new information or synthetic images using algorithms. The process of remote sensing image fusion matches the information between multiple remote sensing platforms, multiple remote sensing data and non-remote sensing data, and the fused data is conducive to comprehensive analysis; [0003] Remote sensing data refers to the analysis of the spectrum, space, texture and other characteristics of various objects based on remote sensing images, and according to...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
IPC IPC(8): G06K9/00G06K9/34G06K9/62G06N3/04G06N3/08G06T3/40G06T7/10G06T11/40
CPCG06N3/08G06N3/04G06T3/4038G06T7/10G06T11/40G06T2207/10032G06T2207/20132G06T2200/32G06T2207/20081G06T2207/20084G06F18/2431G06F18/214
Inventor 刘芳顾行发黄祥志王珂
Owner 江苏天汇空间信息研究院有限公司