| 摘 要: 针对从深度学习角度探究杂乱背景下多小目标运动检测与追踪易于导致小目标误检和漏检的问题,探讨基于蜻蜓视觉神经响应特性的前馈视觉神经网络及多小目标检测与追踪模型。模型设计中:基于蜻蜓视觉信息处理机制建立反映视觉场景变化状态的前馈视觉神经网络;借助输出的运动能量矩阵提取小目标的运动特征;利用多特征融合方法获取小目标的细节特征信息;使用匈牙利算法获取小目标及其运动轨迹。由此,获得计算成本低、无需模型训练、小目标追踪准确率高、结构简单的前馈蜻蜓视觉多小目标运动追踪模型与算法。对比实验结果表明,相较于相关视觉神经网络及深度学习,所提模型能降低复杂场景下小目标易误检和漏检的问题。 |
| 关键词: 多小目标 目标检测 目标追踪 蜻蜓视觉信息处理机制 特征融合 匈牙利算法 前馈蜻蜓视觉神经网络 |
|
中图分类号: TP389.1
文献标识码: A
|
|
| Detection and Tracking of Multi-Small Targets Inspired by Dragon fly Visual Systems |
|
XIA Di1，2, ZHANG Zhuhong1，2
|
(1.College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China;
2.Guizhou Provincial Characteristic Key Laboratory of System Optimization and Scientific Computation, Guiyang 550025, China)
2353124373@qq.com; zhzhang@gzu.edu.cn
|
| Abstract: In petroleum exploration and development, oilfield stations often collect anomalous data due to equipment limitations and operational techniques, leading to deviations in data interpretation. To address this issue, an anomaly detection method based on an improved Isolation Forest using local density is proposed. By calculating the local density of data, the structure and algorithm of Isolation Forest are enhanced, incorporating local density information to improve the identification of local anomalies in low-density regions. Performance and robustness comparison experiments validate the superiority of the improved Isolation Forest method over other machine learning approaches, offering a new perspective for anomaly detection in oilfield stations. |
| Keywords: mult-i small targets target detection target tracking visual information processing mechanism of dragonflies feature fusion Hungarian algorithm feedforward dragonfly visual neural network |
