文献类型：期刊文献

英文题名：PSTNet: object detection in remote sensing images with point supervision and object templates

作者：Liu, Peng[1];Miao, Jun[1];Qiao, Yuanhua[2];Zou, Baixian[3]

第一作者：Liu, Peng

通讯作者：Miao, J[1];Zou, BX[2]

机构：[1]Beijing Informat Sci & Technol Univ, Coll Comp Sci, Beijing 102206, Peoples R China;[2]Beijing Univ Technol, Coll Appl Sci, Beijing 100124, Peoples R China;[3]Beijing Union Univ, Coll Appl Arts & Sci, Beijing 100191, Peoples R China

第一机构：Beijing Informat Sci & Technol Univ, Coll Comp Sci, Beijing 102206, Peoples R China

通讯机构：[1]corresponding author), Beijing Informat Sci & Technol Univ, Coll Comp Sci, Beijing 102206, Peoples R China;[2]corresponding author), Beijing Union Univ, Coll Appl Arts & Sci, Beijing 100191, Peoples R China.|[114172]北京联合大学应用文理学院;[11417]北京联合大学;

年份：2026

卷号：29

期号：2

外文期刊名：PATTERN ANALYSIS AND APPLICATIONS

收录：;WOS:【SCI-EXPANDED(收录号:WOS:001757563300002)】；

基金：This research is partially sponsored by Natural Science Foundation of China (No. 61650201) and Beijing Natural Science Foundation (No. 4202025).

语种：英文

外文关键词：Object detection; Remote sensing images; Point annotation; Template

摘要：Object detection in remote sensing images has become increasingly critical with the rapid development of fields such as unmanned aerial vehicles. However, while rotated bounding box annotations for such images are costly, the use of low-cost point annotations holds great promise. Nevertheless, the inability of point annotations to provide object size and orientation information poses a significant challenge for precise object localization in models. To address this, we propose PSTNet, a framework integrating Template Overlay Learning, Multi-scale Attention Dilated Block (MADB), and Dynamic K-value Sample Assignment. First, category-specific templates are randomly flipped, scaled, and overlaid as pseudo-labels to teach models size and orientation. Second, MADB replaces FPN with dilated convolutions and attention mechanisms to enhance multi-scale feature fusion for small objects. Third, a dynamic K-value strategy leverages classification scores to adaptively assign positive samples, bypassing IoU dependency. Extensive experiments on four datasets show substantial improvements to the baseline.