收集2015年1月至2018年12月共计4年的MODIS光学遥感图像，利用神经网络算法建立适用于南海北部深海与浅海的内孤立波振幅反演模型.模型的计算结果与现有的南海实测数据对比，无论对于浅海（文昌海域）还是深海（东沙岛附近海域和深海盆地），反演精度都高于15%.利用模型得到了南海北部内孤立波的振幅分布，发现振幅分布与水深呈正相关，相关系数为0.81.计算了南海北部文昌海域、东沙岛附近海域以及深海盆地的内孤立波振幅分布，发现同一条内孤立波，振幅不同.南海北部文昌海域、东沙岛附近海域以及深海盆地的单条内孤立波振幅与水深呈正相关，相关系数分别为0.79，0.91，0.81.本文为南海北部海域深海与浅海的内孤立波振幅反演提供借鉴.

The internal solitary waves are studied by analysis of optical remote sensing images in the whole sea area of the Northern South China Sea. This paper mainly collects MODIS optical remote-sensing images from Jan.2015 to Dec.2018, and the amplitude inversion model of internal solitary waves applicable to deep sea and shallow sea in the Northern South China Sea are established by using neural network. The calculation results of the model are compared with the situ data in the Northern South China Sea, and the accuracy is higher than 15% for both shallow sea and deep sea. Using the inversion model, the amplitude distribution of internal solitary waves in the Northern South China Sea is obtained. It is found that the amplitude distribution is positively correlated with water depth, with a correlation coefficient of 0.81. The amplitude distribution of internal solitary waves in Wenchang sea area, Dongsha atoll area and deep sea basin was calculated. It is found that the amplitude of single internal solitary waves is different. The amplitude of internal solitary waves in Wenchang sea area, Dongsha atoll area and deep sea basin is positively correlated with the water depth, with correlation coefficients of 0.79, 0.91 and 0.81 respectively. This study provides a new method for inversion of internal solitary waves amplitude in both deep and shallow areas of Northern South China Sea.