Scholarlink

<p>The rapid growth of visual data in the artificial intelligences demands more efficient sensing and processing architectures. Conventional approaches, including sensing–computing fusion and dynamic vision sensors, often suffer from redundant data acquisition and limited intrinsic selectivity. Here, inspired by the polarization-based mate recognition mechanism of <i>Heliconius cydno</i> butterflies, we present an attention-driven in-sensor selective computing paradigm. We demonstrate an attention vision sensor that uses anisotropic optoelectronic synapses with in-sensor memory to achieve polarization-phase sensitivity and polarization-degree-based contrast filtering. The attention vision sensor array exhibits robust and repeatable polarization-dependent synaptic responses, enabling feature-specific contrast imaging and selective object recognition in cluttered, unpolarized environments. Integrated with a neural network for semantic segmentation, the system achieves 97.57% accuracy for human-made objects. Our bioinspired strategy provides a route towards energy-efficient and task-aware neuromorphic vision hardware.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Attention-driven in-sensor selective computing

  • 期刊论文

  • 发表日期：

    2026年7月27日

查看全文
  • Decai Ouyang,
  • Yu Wu,
  • Yingying Li,
  • Hanbing Li,
  • Qifan Yang,
  • Yihua Huang,
  • Yang Guo,
  • Meiyan Wang,
  • Pengfei Guan,
  • Na Zhang,
  • Lanhao Qin,
  • Xingyu Song,
  • Hua Xu,
  • Jinsong Wu,
  • Xiong Xiong,
  • Yanqing Wu,
  • Huiqiao Li,
  • Yuan Li,
  • Tianyou Zhai

摘要

The rapid growth of visual data in the artificial intelligences demands more efficient sensing and processing architectures. Conventional approaches, including sensing–computing fusion and dynamic vision sensors, often suffer from redundant data acquisition and limited intrinsic selectivity. Here, inspired by the polarization-based mate recognition mechanism of Heliconius cydno butterflies, we present an attention-driven in-sensor selective computing paradigm. We demonstrate an attention vision sensor that uses anisotropic optoelectronic synapses with in-sensor memory to achieve polarization-phase sensitivity and polarization-degree-based contrast filtering. The attention vision sensor array exhibits robust and repeatable polarization-dependent synaptic responses, enabling feature-specific contrast imaging and selective object recognition in cluttered, unpolarized environments. Integrated with a neural network for semantic segmentation, the system achieves 97.57% accuracy for human-made objects. Our bioinspired strategy provides a route towards energy-efficient and task-aware neuromorphic vision hardware.