It is important to find the object as soon as possible for search and rescue operations in maritime accidents. However, there are two problems with search and rescue operations for maritime accidents: (1) data shortage problem and (2) accuracy and recognition speed problem due to wide search range. This paper proposes a methodology to use virtual image data acquired from the commercial game as training data of deep learning algorithms. The proposed methodology shortens the detection time to detect small object by combining image segmentation, enhancement and convolutional neural networks based on game dataset. In order to verify the performance of the proposed methodology, a case study was conducted using real image data. A case study compares the performance of the proposed methodology with the deep learning such as original deep convolutional neural networks. This paper anticipate that the proposed methodology can improve the accuracy and detecting speed by using game dataset and distributed deep learning. It can be applied to high altitude reconnaissance aircraft or offshore structure monitoring where object detection can be applied.