Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
IMAGE STICHING POSITIONING DETECTION METHOD BASED ON ATTENTION MECHANISM
Document Type and Number:
WIPO Patent Application WO/2023/109709
Kind Code:
A1
Abstract:
Disclosed is an image stitching positioning detection method based on an attention mechanism. 1: Preparing an image stitching data set, and dividing the image stitching data set into three parts - a training set, a verification set and a test set; 2: Designing a two-stream multi-task learning neural network structure; 3: Designing a multi-task loss function; 4: Performing optimization training to obtain a stitching area positioning model; 5: Inputting the to-be-detected image into the trained model from step 4 to obtain a stitching positioning result. Compared with the prior art, the advantages of this method lie in: the low-level features of a shallow layer are introduced, more detailed information can be provided, the feature representation capability of the network is improved; the edge of the image and the edge of the stitching area are introduced as supervision information, and a multi-task loss function is designed, so the stitching area can be positioned more accurately; a squeeze-and-exitation attention mechanism is introduced, and the fused features are recalibrated, enabling the model to pay more attention to the features with a relatively large positioning contribution, and obtaining a more accurate stitching positioning result.

Inventors:
ZHANG YULAN (CN)
ZHU GUOPU (CN)
YANG JIANQUAN (CN)
LIU ZUQUAN (CN)
Application Number:
PCT/CN2022/138200
Publication Date:
June 22, 2023
Filing Date:
December 09, 2022
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
SHENZHEN INST ADV TECH (CN)
SHENZHEN INST OF ADV TECH CAS (CN)
International Classes:
G06T3/40; G06N3/04; G06T7/13
Foreign References:
CN114418840A2022-04-29
CN110414670A2019-11-05
CN101493927A2009-07-29
CN111080629A2020-04-28
CN112465700A2021-03-09
EP3171297A12017-05-24
Attorney, Agent or Firm:
BEIJING ZHONG XUN TONG DA INTELLECTUAL PROPERTY AGENCY CO., LTD. (CN)
Download PDF: