Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
CONCRETE DAM DEFECT TIME SEQUENCE IMAGE INTELLIGENT IDENTIFICATION METHOD
Document Type and Number:
WIPO Patent Application WO/2023/216721
Kind Code:
A1
Abstract:
Disclosed is a concrete dam defect time sequence image intelligent identification method, comprising: using a double-flow network to extract a feature sequence of a time sequence image containing concrete dam defects, and adding a time dimension self-attention mechanism to obtain a global context feature relationship; in the training process of the model, using a target function based on a distance intersection-over-union ratio to match a positioning defect and a real defect, and calculating a time sequence position relationship of the defect to accelerate model converge; adding a loss item based on tight perception intersection-over-union ratio into the model loss function, so as to focus on the integrity of the defect sequence to improve accuracy; after defect positioning is completed, using a convolutional neural network based on 2D time sequence difference to extract defect features and identify a defect type. In the present invention, a dam defect time sequence image is effectively detected, the defect position in a long image sequence can be positioned, and the defect type can be accurately identified. The solution has high identification precision and good convergence performance in a task of identifying a dam defect time sequence image.

Inventors:
MA HONGQI (CN)
ZHOU HUA (CN)
MAO YINGCHI (CN)
CHI FUDONG (CN)
ZHOU XIAOFENG (CN)
CAO XUEXING (CN)
QI RONGZHI (CN)
CHEN HAO (CN)
TAN BIN (CN)
NIE BINGBING (CN)
Application Number:
PCT/CN2023/082484
Publication Date:
November 16, 2023
Filing Date:
March 20, 2023
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV HEHAI (CN)
HUANENG LANCANG RIVER HYDROPOWER INC (CN)
International Classes:
G06T7/00; G06N3/04; G06N3/08; G06T7/33; G06V10/22; G06V10/62; G06V10/764
Foreign References:
CN114913150A2022-08-16
CN113283298A2021-08-20
CN108921201A2018-11-30
CN112926396A2021-06-08
CN113239822A2021-08-10
US20200272823A12020-08-27
Attorney, Agent or Firm:
BEIJING RAINBOW XINGYUAN INTELLECTUAL PROPERTY LAW FIRM (CN)
Download PDF: