Inshore and Offshore Ship Detection and Categorization System using Mask R-CNN

Abstract

A regional convolutional neural network (RCNN) can detect ships using high-resolution satellite images. False alarms created by onshore ship-like phenomena, on the other hand, can impair the detection framework's accuracy and feasibility. In this study, we propose Mask R-CNN is a comprehensive technique for minimizing onshore false alarms while also enhancing accuracy and practicality. The mask extraction network is a new addition to the detection system for scene segmentation. An inference mask is used to point out the non target position, which aids in ship detection. By merging the feature map created by the feature pyramid network (FPN) with the inferred mask, the non-target area, spurious candidate targets are eliminated. Mask RCNN-based ship detection uses a proprietary approach to limit false alarms in non-target areas while preserving detection accuracy. Finally, a ship dataset built from high- resolution optical remote sensing photos was used to validate the method's validity and accuracy. The categorizing of the ships is likewise accomplished by web scraping. The features of the input image are extracted using a convolution neural network. This will categorize the ship into one of the following categories of ships namely: accommodation ship, container ship, war ship and cruise ship.