A system for automatic monitoring of wave conditions in littoral waters was constructed. Based on previously reported findings using daytime ocean surface images, the ability of the system to use nighttime images was investigated. Texture analysis of ocean surface images was performed on infrared images, and the characteristic quantities of these results were compared with wind force. Two analytical methods were used to extract the image characteristics of the quantitative result: the spatial gray-level dependence method and the gray-level difference method. A multilayer neural network was trained to predict ocean wind velocity by error back-propagation. The specified conditions and optimal ocean surface images were then compared to the network structure and learning conditions to evaluate the effectiveness of this system and to assess its limitations.