mcgridles / Ship-Detection

Final project for EECE 5644 - Detecting ships in satellite images

Ship Detection

Final project for EECE 5644

Authors

Henry Gridley, Haley Weinstein, Jon White

Dataset

Airbus Ship Detection Challenge

Files

Matlab Functions & Scripts

• calculateError.m: Runs model on test data in test_detections.csv and calculates error percentage
• createDataMatrix.m: Loads train_detections.csv and creates a feature matrix from the training data
• drawLabeledBoxes.m: Draws labeled boxes on an image using a heatmap to determine clusters
• featureExtraction.m: Extracts color, gradient, and spatial features from a data sample
• featurePCA.m: Performs PCA on a feature matrix
• getBoundingBoxes.m: Uses the dataset and provided CSV file to create train_detections.csv and test_detections.csv
• makePredictions.m: Scans a satellite image and attempts to find ships
• ship_detection.m: Ship detection pipeline. This is the main script
• smoothImage.m: Filters an image using a median filter
• thresholdHeatmap.m: Performs thresholding on a heatmap (output of makePredictions())

Other Files

• train_detections.csv: Pre-processed data for training
• test_detections.csv: Pre-processed data for testing
• data.mat: Training data loaded from train_detections.csv
• ship_detection_model.mat: Trained SVM classifier model
• project_proposal.pdf: EECE 5644 final project proposal

Running Pipeline

1. Clone git repo to Documents/MATLAB
2. Download dataset and extract to Documents/MATLAB/Ship-Detection
3. Open ship_detection.m in Matlab and run

Results

Classification of a single satellite image