Multi-angle rotation remote sensing image small target detection method

The invention provides a multi-angle rotation remote sensing image small target detection method. A network structure and a loss function are respectively improved. A feature pyramid network structureis adopted, and a global attention structure is fused into the structure to serve as a feature extra...

Full description

Saved in:
Bibliographic Details
Main Author CAO LIANYU
Format Patent
LanguageChinese
English
Published 10.11.2020
Subjects
CALCULATING
COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS
COMPUTING
COUNTING
HANDLING RECORD CARRIERS
PHYSICS
PRESENTATION OF DATA
RECOGNITION OF DATA
RECORD CARRIERS
Online AccessGet full text

Cover

More Information
Summary:The invention provides a multi-angle rotation remote sensing image small target detection method. A network structure and a loss function are respectively improved. A feature pyramid network structureis adopted, and a global attention structure is fused into the structure to serve as a feature extraction network. The GAFPN network generates features of different scales, the features of multiple scales are unified to a single scale through up-sampling, then the features are further fused and extracted through Merge Networks, the features are more semantically associated and can represent smalltargets, and then the expression ability of the network to the features is improved through an attention mechanism. Then, a classification score and a frame position of the target are given through an RPN target detection module, and a final detection result under any rotation is obtained by adopting rotation non-maximum suppression (RNMS). A multi-task loss function is introduced for end-to-endtraining, so that the model
Bibliography:Application Number: CN202010829461