基于语义分割神经网络UNet,利用GOCI(Geostationary Ocean Color Imager)卫星传感器数据,构建出能够有效提取大型漂浮藻类的深度学习模型,实现了对大型漂浮藻类信息端到端、像素到像素的分割识别。验证结果表明:所提出的深度学习模型对...基于语义分割神经网络UNet,利用GOCI(Geostationary Ocean Color Imager)卫星传感器数据,构建出能够有效提取大型漂浮藻类的深度学习模型,实现了对大型漂浮藻类信息端到端、像素到像素的分割识别。验证结果表明:所提出的深度学习模型对验证集中大型漂浮藻类的平均识别精度达到88.54%;通过与传统的归一化植被指数法和替代型漂浮藻类指数法进行对比,发现基于UNet构建的大型漂浮藻类监测模型具有更高的准确率且受云的影响较小。利用UNet大型漂浮藻类提取模型的识别结果对2017年东海藻类暴发过程进行了分析,模型显示出很好的实用性。展开更多
For space optical remote sensor (SORS) with either film or time delay and integrate charge coupled device (TDI-CCD) imaging, in order to achieve higher resolution it requires more accurate real-time image motion compe...For space optical remote sensor (SORS) with either film or time delay and integrate charge coupled device (TDI-CCD) imaging, in order to achieve higher resolution it requires more accurate real-time image motion compensation. This primarily depends on real-time computation of the image motion velocity vector (IMVV) and error budget and synthesis on related parameters. An effective modeling scheme is introduced and the derivation of IMVV equation, error budget and synthesis by Monte-Carlo method are presented in detail. This total solution was applied to SORS system test on orbit and has been confirmed to be very accurate based on the resolution, transfer function at Nyquist frequency, signal-to-noise ratio and average gray scale of the captured images.展开更多
文摘基于语义分割神经网络UNet,利用GOCI(Geostationary Ocean Color Imager)卫星传感器数据,构建出能够有效提取大型漂浮藻类的深度学习模型,实现了对大型漂浮藻类信息端到端、像素到像素的分割识别。验证结果表明:所提出的深度学习模型对验证集中大型漂浮藻类的平均识别精度达到88.54%;通过与传统的归一化植被指数法和替代型漂浮藻类指数法进行对比,发现基于UNet构建的大型漂浮藻类监测模型具有更高的准确率且受云的影响较小。利用UNet大型漂浮藻类提取模型的识别结果对2017年东海藻类暴发过程进行了分析,模型显示出很好的实用性。
文摘For space optical remote sensor (SORS) with either film or time delay and integrate charge coupled device (TDI-CCD) imaging, in order to achieve higher resolution it requires more accurate real-time image motion compensation. This primarily depends on real-time computation of the image motion velocity vector (IMVV) and error budget and synthesis on related parameters. An effective modeling scheme is introduced and the derivation of IMVV equation, error budget and synthesis by Monte-Carlo method are presented in detail. This total solution was applied to SORS system test on orbit and has been confirmed to be very accurate based on the resolution, transfer function at Nyquist frequency, signal-to-noise ratio and average gray scale of the captured images.
