遥感视频实时云检测的FPGA高效实现方法

针对目前遥感图像云检测算法及算法运行所需硬件平台复杂度高,无法进行在轨实时检测的问题,提出了一种基于FPGA的面向卫星在轨实时运行的遥感视频云检测方法.首先根据不同的遥感视频输入格式对其自适应降采样处理;其次对顺序流入的图像自适应阈值分割,然后对分割后的图像进行聚类获取云区域,进而提取每一块云区域的特征向量;最后计算整幅图像的云覆盖率和可用度,以此判断是否将图像下传.实验结果表明,在60 MHz的时钟下,且Camera Link接口每个时钟周期同时输入两个像素时,822×1096大小的遥感视频云检测速度可达132 fps,相对于传统的嵌入式双核CPU,速度提升了6～7倍.该方法可实现卫星在轨实时云检测,极大地缓解了有限的星地数传带宽和巨大的遥感数据量之间的矛盾,大幅提升遥感卫星系统应用效能,具有很强的实用价值.

Efficient implementation method of real-time cloud detection in remote sensing video based on FPGA

At present, cloud detection algorithms for remote sensing image and hardware platform required by the algorithms are very complicated. In order to solve the real-time problem of cloud detection on orbit, this paper proposed a remote sensing video cloud detection method based on FPGA for satellite. Firstly, it down-sampled the remote sensing video adaptively according to the different input formats. Secondly, it segmented the image by the adaptive threshold. Then, it clustered the segmented image to obtain the cloud regions, and extracted the characteristic of each cloud region. Finally, it calculated the cloud coverage ratio and image availability of the image to determine whether the image was downloaded or not. Experimental results show that, at 60 MHz clock, the speed of cloud detection for image of 822×1096 can reach 132 fps when the Camera Link interface inputs two pixels in each clock cycle, which is 6~7 times faster than the traditional embedded dual core CPU. This method can realize real-time cloud detection on satellite, greatly alleviate the contradiction between data transmission bandwidth and the volume of remote sensing data, and greatly improve the application efficiency of remote sensing satellite system, which has strong practical value.