Visual object recognition is one of the most important and tough problems in computer vision. It targets various visual objects within realistic and real-time images. In depth, deep learning has become a powerful method to extract features directly from input data, which has made great progress in identifying visual objects. Recently, machine learning methods based on deep neural networks play a pivotal role in the field of visual object recognition. In order to identify ships in digital image, the nets need to be trained with a set of labelled images. So far, great progress has been made in visual object recognition based on deep learning, but developing relevant modules is a thorny job. Therefore, in this thesis, we propose a designated methodology based on search neural structure (NAS) for the recognition of visual objects by using our own published datasets to improve the results of sailboat detection. In addition, we conducted data collection for sailboat and kayak detections so as to find the best parameters based on basic model of YOLOv5. In this thesis, we also compare the net architectures and seek the best one. We test the proposed model and compare it with others.