Ship IMO Region Detection Based on The Wavelet Decomposition Optimized CenterNet

Current research on object detection mainly focuses on objects with specific contour features. However, there is very little research on object detection of ship IMO (International Maritime Organization) regions, which do not have apparent contour features. Object detection of ship IMO regions is an essential primary condition for unmanned aerial vehicles to patrol and monitor designated ships in seaports automatically. Meanwhile, the utilization of the existing object detection framework for ship IMO region detection results in a relatively low average precision. Therefore, we propose a ship IMO region detection method based on wavelet decomposition optimized CenterNet to address the target characteristics of ship IMO regions. The input image is decomposed using wavelet decomposition to obtain several hierarchical target feature information layers. Each layer is weighted and averaged based on its proportion in the original image signal, then combined with the feature layer of the same size in the backbone of CenterNet. The experimental results demonstrate significant performance improvements in mAP, precision, recall, and F1-Score for the wavelet decomposition optimized CenterNet compared to the original CenterNet in ship IMO region detection, with input image sizes of 512×512, 640×640, and 768×768. The code is available at https://github.com/czb7046/CenterNet-Wavelet.