The Nature Conservancy Fisheries Monitoring

Group ID: 201612-13

Team Member

• Kewen Zhang (kz2246@columbia.edu)
• Pengfei Wang (pw2406@columbia.edu)
• Yueying Teng (yt2495@columbia.edu)

Introduction

This project is inspired by Kaggle.com "the-nature-conservancy-fisheries-monitoring" Link.

Environments

• Operating System: macOS Sierra
• Language: Python2.7
• Software: Pyspark, Opencv3

Outline

Train Haar-feature-based cascade classifiers

• detection_crop_train_fish.py: Crop fish in the training set
• detection_train_haar_classifier.py: Train Cascade classifier
• detection_generate_negative.py: Generate negative instances
• detection_clustering_ship.py: Cluster the fishery ships to denoise the raw images
• detection_crop_window.py: Get window of each fishery ships

BoW Model

• classification_get_sift_bow.py: Extract SIFT features and map into bag-of-words model
• classification_mllib_pyspark.py: Machine Learning with Pyspark

Convolutional Neural Network

• classification_cnn.py: Construct Convolutional Neural Network model
• classification_cnn_LAG.py: Improve result by re-weighting

Technical Presentation

https://youtu.be/dlaS8nACQ7w