基于高斯混合模型的遥感影像云检测技术

【目的】高分辨率遥感影像云检测技术一直是遥感影像处理亟待解决的难题,尤其是边缘薄云和散云的检测。针对高分辨率遥感影像的成像特点,利用形态学运算、多边形简化技术,实现含云影像的准确提取。【方法】首先对影像进行高斯低通滤波平滑,取得一致均匀的明暗效果; 然后将影像分为多云、少云、无云3种情况,对多云影像采用Otsu阈值分割,对少云影像采用高斯混合模型进行阈值分割; 最后对云区进行形态学处理得到最终云区。【结果】高分辨遥感影像云检测方法目视效果较好,可有效提高云检测精度,该方法准确率为98.60%,查全率在90%左右,错误率约为2.58%,可以较为准确地检测出厚云、薄云、散云,同时还可有效地减少对房屋、道路、裸地的误判。【结论】基于Otsu阈值分割和高斯混合模型的高分辨率遥感影像云检测技术算法复杂度适中,计算量小,运算速度快,检测精度高,适用性广。
Abstract: 【Objective】 Developing cloud detection technology for high-resolution remote-sensing images is difficult, especially for thin edge and scattered clouds. We used morphological operations and polygon simplification techniques to accurately extract cloud-containing, high-resolution remote-sensing images.【Method】First, the Gaussian low-pass filter was used to smooth the image and create uniform shading. The image was then divided into three categories: cloudy, partly cloudy and cloudless. The Otsu threshold method was used on cloudy images, and Gaussian mixture model segmentation was used on partly cloudy images. Finally, the cloud area was morphologically processed to determine the final cloud area. 【Result】The high-resolution remote-sensing image cloud-detection algorithm based on Otsu threshold segmentation and Gaussian mixture model has good visual effect and can effectively improve the accuracy of cloud detection. The accuracy rate is 98.60%, recall rate of the method is approximately 90%, and the error rate is approximately 2.58%. It can accurately detect thick, thin and scattered clouds, and can also effectively reduce the misidentification of houses, roads and bare land. 【Conclusion】The high-resolution remote-sensing image cloud-detection algorithm based on Otsu threshold segmentation and Gaussian mixture model is moderately complex, and has a small computation size, fast operation speed, high-precision detection and wide applicability