利用区域信息融合混合活动轮廓模型的河流遥感图像分割

发布时间：2018-05-14 22:14

本文选题：河流遥感图像 + 图像分割　；参考：《中国图象图形学报》2017年02期

【摘要】：目的河流遥感图像是背景复杂的非匀质图像,利用传统的活动轮廓模型进行分割往往不够准确。针对这一问题,提出了基于区域信息融合的混合活动轮廓模型来分割河流遥感图像。方法该混合模型将Chan-Vese(CV)模型和基于交叉熵的活动轮廓模型的外部能量约束项相结合,并赋予归一化调节比例系数。通过计算轮廓曲线内外区域像素灰度的方差和交叉熵,指导曲线逼近目标边缘。为了加速混合模型的演化,引入曲线内外区域像素灰度的类内绝对差,取代原有的内外区域能量权值,以提高混合模型的分割效率。结果大量实验结果表明,相较于CV模型、测地线模型、基于交叉熵的活动轮廓模型、CV模型和测地线模型的混合模型以及局部全局灰度拟合能量模型(LGIF),本文混合模型分割河流遥感图像的灵敏度和上述方法都接近于100%,准确率大幅提升,在90%以上,虚警率则下降了约50%,且所需迭代次数和运行时间更少。结论本文提出的混合模型主要适用于具有一定对比度的河流遥感图像,在分割性能和分割效率两个方面,都有明显的优势。
[Abstract]:Objective River remote sensing image is a heterogeneous image with complicated background. It is often not accurate to segment river remote sensing image by using traditional active contour model. To solve this problem, a hybrid active contour model based on regional information fusion is proposed to segment river remote sensing images. Methods the hybrid model combines the Chan-Vesewe CVV model with the external energy constraints of the active contour model based on cross-entropy, and gives a normalized adjusting scale coefficient. By calculating the variance and cross entropy of pixels in the contour curve, the curve is guided to approach the edge of the target. In order to accelerate the evolution of the hybrid model, the intra-class absolute difference of pixel grayscale in the inner and outer region of the curve is introduced to replace the original energy weight of the inner and outer region, so as to improve the segmentation efficiency of the hybrid model. Results A large number of experimental results show that compared with CV model, geodesic model, Based on cross entropy, the mixed model of CV model and geodesic model and the local global gray level fitting energy model are used to segment river remote sensing images. The sensitivity and the above methods are close to 100, and the accuracy is greatly improved. Above 90%, the false alarm rate decreases about 50%, and the number of iterations and the running time are less. Conclusion the hybrid model proposed in this paper is mainly suitable for river remote sensing images with certain contrast and has obvious advantages in segmentation performance and segmentation efficiency.
【作者单位】：南京航空航天大学电子信息工程学院;长江水利委员会长江科学院武汉市智慧流域工程技术研究中心;黄河水利委员会黄河水利科学研究院水利部黄河泥沙重点实验室;南京水利科学研究院港口航道泥沙工程交通行业重点实验室;哈尔滨工业大学城市水资源与水环境国家重点实验室;
【基金】：国家自然科学基金项目(61573183) 港口航道泥沙工程交通行业重点实验室开放基金长江科学院开放基金项目(CKWV2013225/KY) 水利部黄河泥沙重点实验室开放基金项目(2014006) 城市水资源与水环境国家重点实验室开放基金项目(LYPK201304) 江苏高校优势学科建设工程(2012)~~
【分类号】：TP751

【参考文献】