Open AccessHarbor Ship Detection Based on Channel Weighting and Spatial Information Fusion
Author(s) -
Quancheng Zhou,
Fei Song,
Zhenghao Chen,
Rui Zhang,
Ping Jiang,
Tao Leí
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1738/1/012057
Subject(s) - weighting , channel (broadcasting) , computer science , pyramid (geometry) , feature (linguistics) , artificial intelligence , dimension (graph theory) , pattern recognition (psychology) , stage (stratigraphy) , data mining , spatial analysis , computer vision , remote sensing , geography , mathematics , telecommunications , geology , linguistics , philosophy , medicine , paleontology , geometry , pure mathematics , radiology
Aiming at the problems of false alarms and missed detection caused by large differences among ship types and complex background in harbor remote sensing images, a robust single-stage ship detection method is proposed. First, a Channel Weighting Mechanism is devised, which self-learns the weights of different channel to enhance valid features in channel dimension. Second, a Spatial Information Fusion Module is designed to enhance features in spatial dimension, which extracts more information of ship appearance from shallow feature maps, then excavates potential contextual information in deep semantic features. Finally, to promote the detection ability of ships of various scales, a Multi-stage Weighted Fusion Pyramid is applied to optimize the fusion of high-stage features and low-stage features. Extensive experiments conducted on self-established dataset for harbor ships show that the proposed method provides the performance by 3.08%mAP compared to RetinaNet.