Scholarlink

To achieve automatic dense target localization and detection, the following issues need to be addressed: firstly, due to the complex and constantly changing environmental information (Liu Y, Complex background ground based on artificial intelligence Research on multi-objective extraction technology. Xi’an Technological University, Xi’an) collected by mobile cameras and high-altitude drones, pre-processing methods such as filtering, Laplacian enhancement (Yang L, Hai NW, Yu QF, Chin J Aeronaut 36:353–365, 2023) and threshold segmentation are required on the original images to achieve accurate target extraction. Secondly, the shape, size, and color of targets in the image change over time and under changing lighting conditions. Achieving accurate recognition of targets under these conditions is a challenging problem. The team plans to use methods such as dense target tilt correction, overlapping segmentation, and target deduplication. Our team ultimately completed the classification and feature extraction of motion intensive targets under the camera, and based on deep learning, studied the recognition and perception of target behavior in complex environments relying on visual information, ultimately achieving more reliable and universal accurate detection of multiple targets in complex environments.

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

Dense Object Detection in Complex Backgrounds Based on Deep Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xiyuan Wan,
  • Qingdong Luo,
  • Pengfei Zheng,
  • Jingjing Lou,
  • Chaoqun Jin

摘要

To achieve automatic dense target localization and detection, the following issues need to be addressed: firstly, due to the complex and constantly changing environmental information (Liu Y, Complex background ground based on artificial intelligence Research on multi-objective extraction technology. Xi’an Technological University, Xi’an) collected by mobile cameras and high-altitude drones, pre-processing methods such as filtering, Laplacian enhancement (Yang L, Hai NW, Yu QF, Chin J Aeronaut 36:353–365, 2023) and threshold segmentation are required on the original images to achieve accurate target extraction. Secondly, the shape, size, and color of targets in the image change over time and under changing lighting conditions. Achieving accurate recognition of targets under these conditions is a challenging problem. The team plans to use methods such as dense target tilt correction, overlapping segmentation, and target deduplication. Our team ultimately completed the classification and feature extraction of motion intensive targets under the camera, and based on deep learning, studied the recognition and perception of target behavior in complex environments relying on visual information, ultimately achieving more reliable and universal accurate detection of multiple targets in complex environments.