摘要
蝇类昆虫–果蝇复眼视觉系统的高适应性和高可靠性是一种自然特性,对于视觉杂乱场景中感兴趣目标/区域(统称“图形”)的分辨和飞行追踪过程,在本质上却是良态和适定的。本文针对遥感影像水体信息提取所涉及的图像分割和“图形–背景(figure-backgrounds, FB)”分辨这一逆问题求解存在的病态的(不适定的)本质性困难,基于昆虫生理学研究的新发现,分析果蝇复眼视觉信息加工的神经过程,模拟其无需背景建模、先验信息以及不依赖于样本数据训练隐式模型,所具有的视觉杂乱背景且噪声干扰下“图形–背景”分辨的功能优势,提出一种仿蝇视觉“图形–背景”分辨的遥感影像水体提取方法,通过多组仿真实验,并与标准的归一化差异水体指数NDWI、改进的NDWI (MNDWI)、决策树模型以及SVM分类等方法做了分析对比,验证了新方法的优越性。
The high adaptability and reliability of the compound visual system of flies and drosophila is a natural characteristic, and the identification and flight tracking process of the target/region of interest (general called “graphics”) in the visual clutter scene is essentially well-conditioned and well-adapted. This paper focuses on the ill-posed (not well-posed) inherent difficulties of image segmentation and inverse problem of “figure-backgrounds (FB)” resolution in water extraction from remote sensing images. Based on the new findings of insect physiology, the neural processing of compound visual information in drosophila is analyzed, and the implicit model is trained by simulating modeling without background and prior information, or relying on the sample data, which has the advantages of visual clutter and “figure-backgrounds” resolution under noise interference. A method of water extraction from remote sensing image based on simulating fly’s vision “figure-backgrounds” resolution is proposed. Compared with the standard normalized differential water body index (NDWI), improved NDWI (MNDWI), decision tree model and SVM classification method, the superiority of the new method is verified.
出处
《图像与信号处理》
2021年第4期155-165,共11页
Journal of Image and Signal Processing
