Marine ship recognition is a challenging Fine-Grained Visual Categorization (FGVC) problem due to the large visual variations caused by motion blur, occlusion, lighting changes, and etc. The visual distinction between similar categories is usually very small, so it is difficult to solve it with general recognition algorithm. It demands an advanced discriminative model to accurately segment marine ships from the backgrounds and classify the type of the ship. However, effective models for the problem tend to be computationally prohibitive. To address these two conflicting challenges, we propose to recognize marine ship based on two cascade CNNs (convolutional neural networks), a shallow CNN and a deep CNN. The shallow CNN is used to quickly remove most of the background regions to reduce the computation cost, and the deep CNN is used to classify the type of ship in the remaining regions. The two CNNs are trained end-to-end, and they are complementary to each other to guarantee the recognition precision with low computation cost. Experimental results show that the proposed method is promising for marine ship recognition.
|