Kaggle: Whale Challenge 張智星 jang@cs.nthu.edu.tw http://www.cs.nthu.edu.tw/~jang 多媒體資訊檢索實驗室 台灣大學 資訊工程系
Whale Challenge Problem definition Characteristics: Imbalance data Identify the existence of whales from sensor recordings Characteristics: Imbalance data Some recordings are hardly recognizable by non-experts
Dataset Training set Test set Recording format 47,844 recordings of 2 seconds 88.97% (42,565 recordings): w/o whales 11.03% (5,276 recordings): with whales Test set 25,468 recordings of 2 seconds Recording format 2000-Hz sample rate, 16-bit resolution
Preprocessing Potential preprocessing Trend removal Noise removal Trend estimation via polynomial fitting Noise removal Band-pass filter Removal of “non-whale” part Linear prediction?
Spectrogram kwcPreprocess.m W/o band-pass filter W/ band-pass filter
Potential Features Acoustic features Volume Pitch Spectrum MFCC … Visual features (obtained from spectrogram) Radon transform Hough transform Gabor filters …
Pitch Tracking kwcPitchTracking.m
Volume kwcVolume.m
Spectrogram kwcSpectrogram.m
Visual Features via Radon Transform Projection onto lines at various angles For grayscale images only Detection objects at a specific angle
Example of Radon Transform Source http://www.mathworks.com/help/images/ref/radon.html Output Code: goRadon.m
Example of Radon Transform (2) Source image Output Code: goRadon2.m
Visual Features via Hough Transform Commonly used for detection lines and circles For BW images only (after edge detection)
Visual Features via Hough Transform (2) Point to curve mapping Two points Two sine curves The intersection is the right θ and ρ for the line connecting these two points
Example of Hough Transform Source http://www.ebsd-image.org/documentation/reference/ops/hough/op/houghtransform.html Image Hough space and its maxima Detected lines
Example of Hough Transform (2) Source http://www.mathworks.com/help/images/analyzing-images.html (MATLAB code available) Image Edge image Hough space and its maxima Detected lines
Methods Thresholding Static classifiers Sequence classifiers Volume variance Pitch variance Static classifiers Naïve Bayes classifiers GMM SVM … Sequence classifiers HMM CRF …
HMM Training kwcHmmTrain.m
HMM Evaluation kwcHmmEval.m
HMM Basic models Advanced models Class 1: sil Class 2: sil-whale-sil sil-whale-sil-whale-sil … 1.0 sil 0.9 0.4 1.0 sil w sil 0.1 0.6
HMM (2) Other approach Train HMM models Align each recording with the HMM Extract features from the whale part for other static classifiers Duration (no. of frames) Average log likelihood per frame 0.9 0.4 1.0 sil w sil 0.1 0.6
Performance Evaluation Performance evaluation of methods based on thresholding (http://en.wikipedia.org/wiki/Receiver_operating_characteristic): ROC, DET AUC