您好，欢迎访问安徽省农业科学院 机构知识库！

ESR-DETR: Explicit spatial relationship-based framework for end-to-end aphid detection under field environment

文献类型： 外文期刊

作者： Chen, Hongbo 1 ; Chen, Tianjiao 1 ; Liu, Haiyun 1 ; Du, Jianming 2 ; Wang, Rujing 1 ; Dong, Wei 4 ;

作者机构： 1.Univ Sci & Technol China, Hefei 230026, Peoples R China

2.Chinese Acad Sci, Hefei Inst Phys Sci, Inst Intelligent Machines, Hefei 230031, Peoples R China

3.Intelligent Agr Engn Lab Anhui Prov, Hefei, Peoples R China

4.Anhui Acad Agr Sci, Agr Econ & Informat Res Inst, Hefei 230001, Peoples R China

关键词： Aphid; Explicit spatial priors relationship; Query feature enhancement; End-to-End

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 237 卷

页码：

收录情况： SCI

摘要： Aphids are among the most destructive pests affecting wheat crops, posing severe threats to both the quality and yield of wheat. Therefore, accurate detection of wheat aphids is crucial for improving wheat quality and promoting the development of the agricultural economy. However, current DETR-like detection methods have not fully accounted for the characteristics of aphids, such as their small size and spatially clustered distribution in natural environments. These methods neglect the spatial relationships between object queries, which restricts detection accuracy. To address this problem, an end-to-end detection framework based on explicit spatial priors relationship among pests is proposed in this paper. Firstly, a self-attention mechanism based on receptive field distance is designed to explicitly construct spatial position relationships among queries, enhancing the spatial perception capability between object queries. Secondly, a spatial relationship extractor and query updater are introduced to enhance the query features within local regions, thereby improving the feature representation of pests in complex environments. Finally, a wheat aphid dataset (WA2024) with complex background in natural environment is constructed. We evaluate the effectiveness of the proposed method through extensive experiments. Experimental results show that the proposed method achieves 51.1% mAP and 94.6% AP on the WA2024 dataset, outperforming state-of-the-art object detectors. Notably, experiments on the public VisDrone dataset also demonstrate the superior performance of the proposed method, further validating its generalization capability.

  • 相关文献
作者其他论文 更多>>
置顶
购物车
反馈

© 京ICP备09089781号-29 主办单位：安徽省农业科学院

学术资源： 中国农业科学院 农业专业知识服务系统 农科联盟资源共建共享平台 中国农业科技文献与信息服务平台 中国农业期刊集成服务平台