Ship image classification in ocean background is of great significance for military and civilian domains which will further improve the automation in ship identification and naval domain perception. Deep stacked layers of neurons are extensively employed in recent years due to their ability to recognize the high-level features from an image in a hierarchal way. However, CNN lacks the capability of dealing with global rotation in an image of large size. This limits the accuracy of the CNN algorithm. Therefore, we have proposed a deep learning framework which uses a few layers of AlexNet for initial feature extraction and subsequently, KNN classifiers have been utilized for measuring the accuracy in classifying ships according to various categories. © Springer Nature Singapore Pte Ltd. 2020.