Power transformation project progress monitoring multi-source data acquisition and labeling method

Abstract

The invention relates to a power transformation project progress monitoring multi-source data acquisition and labeling method, which specifically comprises the following steps of: 1) acquiring samples, generating samples from different identified targets, performing geometric transformation on images by using a data amplification method, and further expanding training samples; 2) sample data labeling: according to a recognition algorithm task, labeling work is divided into frame labeling and region labeling; the frame annotation comprises a main control building, a transformer and a capacitor, and objects to be detected are selected from the image in a frame mode: object names and target bounding box information; and 3) inputting two-stage images of a fixed point area by constructing a change detection data set, and learning and extracting related feature information according to image data. The power transformation project progress monitoring multi-source data acquisition and labeling method has the advantages that the scheme is simple and effective, multi-source data acquisition is convenient, and the labeling effect is good.

Description

technical field [0001] The invention relates to a multi-source data collection and labeling method for substation project progress monitoring. Background technique [0002] Data preprocessing is mainly used to expand the number of samples, improve the challenges of insufficient data and too few data types, so as to improve model robustness and generalization performance. Specifically, proper preprocessing can effectively highlight the overall and local characteristics of the image, focus on the change area of ​​building facilities, increase the difference between different building features in the sample data set, enrich feature information, improve the recognition performance of the change detection model, and meet the needs of change detection building facility recognition. need. Therefore, data preprocessing is a key step in the implementation of the scheme. [0003] The deep learning model needs to use a large number of training samples to optimize and generate model f...

AI Technical Summary

Problems solved by technology

Samples are scarce and data is difficult to obtain targets. When training models, there is not enough data for sample training, which often leads to model performance degradation and insufficient model generalization performance

Images