Multi-Site and Multi-Scale Unbalanced Ship Detection Based on CenterNet

Object detection plays an essential role in the computer vision domain, especially machine learning-based approach, which has developed rapidly past decades. However, development of convolutional neural networks marine field is relatively slow, such as ship classification and tracking. In this paper, considered a central point regression task but discards non-maximum suppression operation. We first improved deep layer aggregation network to enhance feature extraction capability tiny targets, then reduced number parameters through lightweight convolution module, finally employed unique activation function nonlinearity model. By doing this, not only suits unbalanced sample ratios classifying, more robust scenarios where both resolution samples are unstable. Experimental results demonstrate that proposed approach obtains outstanding performance object compared with current advanced methods. Furthermore, contrast original CenterNet framework, mAP increased by 5.6%.