The main objective is to detect the ships in satellite images. Dataset is made up of 4000 80x80 RGB photos labeled with "ship" or "no-ship" classifications and saved as a 19200-item list. These images were extracted from ‘Planet satellite imagery collected over the San Francisco Bay and San Pedro Bay areas of California’.
Dataset : Ships in Satellite Imagery
Multi-step approach in R-CNN:
- Generate region proposals: selective search
- classify the content of the region proposals within a refined bounding box.
Main steps process to implementing an R-CNN object detector:
- Build an object detection dataset using Selective Search : data_preprocessing
- Apply rcnn algorithm :
- Basic RCNN based on region proposal
- RCNN using bounding box
Basic RCNN implementation in detail:
