Faster R-CNN with a cross-validation approach to object detection in radar images

One of the most important factors in the sustainability of the marine environment is maritime security and surveillance. On the other hand, the object detection in the ocean using radar sensors is widely used in control and monitoring activities on the sea surface. In previous work, a detection model based on the Faster R-CNN architecture achieved a good performance, in terms of accuracy and response time. However, in the present study, we propose the application of the Cross-Validation method to improve the issue of learning with a limited dataset. Through the proposed approach, the design of a more robust training model has been achieved as well as an unbiased selection of the best mo-el. In this way, it has been possible to evaluate the results of previous works by applying the statistical scope of precision with the Repeated K-Fold Cross-Validation metric.