【摘要】：衛(wèi)星云圖云量計算是衛(wèi)星氣象應(yīng)用的基礎(chǔ),現(xiàn)階段對其的研究未能充分利用衛(wèi)星云圖的特征,導(dǎo)致云檢測及云量計算的效果不好。針對該問題,利用多層神經(jīng)網(wǎng)絡(luò)進行衛(wèi)星云圖的特征提取,并通過大量實驗尋找到最優(yōu)的深度學(xué)習(xí)的網(wǎng)絡(luò)結(jié)構(gòu)�；诙葮O限學(xué)習(xí)機對衛(wèi)星云圖的云進行檢測和分類,再利用"空間相關(guān)法"計算云圖中的總云量。實驗結(jié)果表明,基于傳統(tǒng)極限學(xué)習(xí)機的深度極限學(xué)習(xí)機能夠充分提取云圖的特征,在進行云分類時能夠較清晰地區(qū)分厚云和薄云間的界限。相比于傳統(tǒng)閾值法、極限學(xué)習(xí)機模型以及卷積神經(jīng)網(wǎng)絡(luò),深度極限學(xué)習(xí)機的云識別率以及云量計算準確率更高,且所提方法比卷積神經(jīng)網(wǎng)絡(luò)的效率更高。
[Abstract]:The cloud volume calculation of satellite cloud image is the basis of satellite meteorological application. At present, the research on it fails to make full use of the characteristics of satellite cloud image, which leads to the poor effect of cloud detection and cloud amount calculation. To solve this problem, the multi-layer neural network is used to extract the features of satellite cloud images, and the optimal network structure of depth learning is found through a large number of experiments. Based on the degree limit learning machine, the cloud of satellite cloud image is detected and classified, and the total cloud amount in the cloud image is calculated by "spatial correlation method". The experimental results show that the depth extreme learning machine based on the traditional extreme learning machine can fully extract the features of cloud images and can clearly distinguish the boundary between thick cloud and thin cloud in cloud classification. Compared with the traditional threshold method, the cloud recognition rate and cloud volume calculation accuracy of the depth ultimate learning machine are higher than that of the traditional threshold method, and the proposed method is more efficient than the convolution neural network.
【作者單位】： 南京信息工程大學(xué)江蘇省大氣環(huán)境與裝備技術(shù)協(xié)同創(chuàng)新中心 南京信息工程大學(xué)江蘇省大數(shù)據(jù)分析重點實驗室
【基金】：國家自然科學(xué)基金(61503192) 江蘇省自然科學(xué)基金(BK20161533) 江蘇省六大人才高峰高層次人才資助計劃(2014-XXRJ-007)資助
【分類號】：TP181;TP391.41
，

本文編號：2308216