{
"cells": [
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"collapsed": false,
"scrolled": true
},
"outputs": [],
"source": [
"Image(url='http://www.teamnetwork.org/files/home/team-header-1000.gif')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Classifying Wildlife Camera Trap Images for Conservation International’s Tropical Ecology Assessment and Monitoring Network"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Problem Statement\n",
"\n",
"Conservation Internaitonal's TEAM Network has labeled camera trap images of wildlife species\n",
"\n",
"- Images require manual classification from experts in the field\n",
"- Highly unbalanced classes of species\n",
"- Many of the images are misclassified\n",
"- There is a class of unlabeled/unknown images\n",
"- Existence of endangered species images preclude the use of citizen scientists to help classify images"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## [Data Analysis and Exploration](#Data-Analysis-and-Exploration)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## [Classification Piepline: Image Segmentation](#Image-Segmentation)\n",
"\n",
"- Manual Cropping\n",
"- Image Segmentation with Graph Cut \n",
"http://cmp.felk.cvut.cz/~smidm/python-packages-for-graph-cuts-on-images.html (TBD)\n"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## [Classification Pipeline: Feature Extraction](#Feature-Extraction)\n",
"- Template Matching\n",
"- Histogram Comparison\n",
"- Scale Invariant Feature Transform (SIFT)\n",
"- Bag of Visual Words"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## [Classification Pipeline: Classification](#Classification)\n",
"- Linear SVM\n",
"- k-means\n",
"- Ada boost"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Set Up"
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"last updated: Sun Dec 25 2016 \n",
"\n",
"CPython 3.5.2\n",
"IPython 5.1.0\n",
"\n",
"pandas 0.19.1\n",
"numpy 1.11.2\n",
"cv2 3.1.0\n",
"sklearn 0.18.1\n",
"matplotlib 1.5.3\n",
"\n",
"compiler : GCC 4.2.1 Compatible Apple LLVM 4.2 (clang-425.0.28)\n",
"system : Darwin\n",
"release : 14.5.0\n",
"machine : x86_64\n",
"processor : i386\n",
"CPU cores : 4\n",
"interpreter: 64bit\n"
]
},
{
"name": "stderr",
"output_type": "stream",
"text": [
"/Users/Luxive/anaconda/envs/tensorflow/lib/python3.5/site-packages/sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
" \"This module will be removed in 0.20.\", DeprecationWarning)\n",
"/Users/Luxive/anaconda/envs/tensorflow/lib/python3.5/site-packages/sklearn/grid_search.py:43: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. This module will be removed in 0.20.\n",
" DeprecationWarning)\n"
]
}
],
"source": [
"# https://github.com/rasbt/watermark\n",
"%load_ext watermark\n",
"%watermark -u -n -v -p pandas,numpy,cv2,sklearn,matplotlib -m\n",
"\n",
"import pandas as pd\n",
"import numpy as np\n",
"import sys\n",
"import os\n",
"# import PIL\n",
"# import sift_pyocl # uses pyOpenCL to access GPUs and other massively parallel compute devices from Python\n",
"# import cyvlfeat as vl # a Python (cython) wrapper of the popular VLFeat library\n",
"import cv2\n",
"\n",
"import glob\n",
"import random\n",
"import warnings\n",
"\n",
"from sklearn.cluster import MiniBatchKMeans\n",
"from sklearn.svm import SVC\n",
"from sklearn.model_selection import cross_val_score\n",
"from sklearn.grid_search import GridSearchCV\n",
"from sklearn.ensemble import AdaBoostClassifier\n",
"from sklearn.externals import joblib\n",
"from sklearn import preprocessing\n",
"from sklearn.model_selection import train_test_split\n",
"from sklearn.preprocessing import LabelEncoder\n",
"from sklearn import metrics\n",
"\n",
"from matplotlib import pyplot as plt\n",
"import matplotlib.image as mpimg\n",
"%matplotlib inline\n",
"\n",
"from IPython.display import Math\n",
"\n",
"ROWS = 180 #720\n",
"COLS = 320 #1280\n",
"CHANNELS = 3\n",
"SPECIES_CLASSES = ['Loxodonta']"
]
},
{
"cell_type": "code",
"execution_count": 52,
"metadata": {
"collapsed": false
},
"outputs": [],
"source": [
"def get_files(directory, species, filetype, random=0):\n",
" \"\"\"Load files from directory folder\"\"\"\n",
" \"\"\"Use os.walk() traversal method\"\"\"\n",
" images = []\n",
" for root, dirs, files in os.walk(directory+species):\n",
" for file in files:\n",
" if file.endswith(filetype):\n",
" images.append(os.path.join(root,file))\n",
" return images\n",
"\n",
"def get_metadata(directory, species):\n",
" \"\"\"Load metadata file from directory folder\"\"\"\n",
" \"\"\"Use os.walk() traversal method\"\"\"\n",
" for root, dirs, files in os.walk(directory+species):\n",
" for file in files:\n",
" if file.endswith(\".csv\"):\n",
" meta = pd.read_csv(os.path.join(root,file), header=46,low_memory=False)\n",
" return meta\n",
"\n",
"def read_image(src,resize=False):\n",
" \"\"\"Read and resize individual images\"\"\"\n",
" im = cv2.imread(src, cv2.IMREAD_COLOR)\n",
" if resize:\n",
" im = cv2.resize(im, (COLS, ROWS), interpolation=cv2.INTER_CUBIC)\n",
" return im\n",
"\n",
"def show_rgb_img(img):\n",
" \"\"\"Display a typical color image without resizing\"\"\"\n",
" return plt.imshow(cv2.cvtColor(img, cv2.CV_32S))\n",
"\n",
"def to_gray(color_img):\n",
" \"\"\"Convert a 3-channel RGB color image to grayscale image\"\"\"\n",
" gray = cv2.cvtColor(color_img, cv2.COLOR_BGR2GRAY)\n",
" return gray\n",
"\n",
"def fd(algorithm):\n",
" if algorithm == \"SIFT\":\n",
" return cv2.xfeatures2d.SIFT_create()\n",
" if algorithm == \"SURF\":\n",
" return cv2.xfeatures2d.SURF_create(float(sys.argv[3]) if len(sys.argv) == 4 else 4000)\n",
"\n",
"def gen_fd_features(img, # input image\n",
" compute=False, # toggle compute on or off\n",
" nfeatures=0, # number of best features to retain, ranked by local contrast score\n",
" nOctaveLayers=3, # number of layers in each octave, computed automatically from the image resolution\n",
" contrastThreshold=0.04, # contrast threshold used to filter out weak features in semi-uniform regions. \n",
" # The larger the threshold, the less features are produced by the detector.\n",
" edgeThreshold=10, # threshold used to filter out edge-like features\n",
" # the larger the edgeThreshold, the less features are filtered out (more features are retained).\n",
" sigma=1.6, # sigma of the Gaussian applied to the input image at the octave #0\n",
" # images captured with a weak camera and soft lenses require smaller numbers\n",
" mask=None):\n",
" descriptors = None\n",
"\n",
" # Create a SIFT Object\n",
" sift = cv2.xfeatures2d.SIFT_create(nfeatures=nfeatures,\n",
" nOctaveLayers=nOctaveLayers,\n",
" contrastThreshold=contrastThreshold,\n",
" edgeThreshold=edgeThreshold,\n",
" sigma=sigma)\n",
" # descriptors are the 128-dimensional vectors of SIFT dewscriptors\n",
" if compute:\n",
" keypoints, descriptors = sift.detectAndCompute(image=img,mask=mask) # 1. grayscale image, 2. optional input mask\n",
" else:\n",
" keypoints = sift.detect(image=img,mask=mask) # 1. grayscale image, 2. optional input mask\n",
" return keypoints, descriptors\n",
"\n",
"def show_sift_features(img, kp): # 1. grayscale image, 2. keypoints\n",
" \"\"\"Overlay SIFT keypoints onto grayscale image\"\"\"\n",
" plt.imshow(cv2.drawKeypoints(img, kp, None, flags=cv2.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS))\n",
" return plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Data Analysis and Exploration\n",
"### WARNING: DO NOT RUN. THESE FILES ARE ON A LOCAL DROPBOX FOLDER AND TOTAL 7.3GB"
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"31485 expected photos of Loxodonta\n"
]
}
],
"source": [
"# define images directory\n",
"DROPBOX = \"/Users/Luxive/Dropbox/TEAM_Machine Learning/\"\n",
"\n",
"nnn_metadata = get_metadata(DROPBOX, SPECIES_CLASSES[0])\n",
"print(\"{0} expected photos of {1}\".format(nnn_metadata.Species.count(),SPECIES_CLASSES[0]))"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {
"collapsed": false
},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"31415 actual photos of Loxodonta\n"
]
}
],
"source": [
"# get all image filenames for each species\n",
"files = []\n",
"y_all = []\n",
"\n",
"for species in SPECIES_CLASSES:\n",
" species_files = get_files(DROPBOX,species,\".JPG\")\n",
" files.extend(species_files)\n",
"\n",
" y_species = np.tile(species, len(species_files))\n",
" y_all.extend(y_species)\n",
" print(\"{0} actual photos of {1}\".format(len(species_files), species))\n",
" \n",
"y_all = np.array(y_all)"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {
"collapsed": false
},
"outputs": [
{
"data": {
"text/html": [
"\n",
"
\n",
" \n",
" \n",
" | \n",
" actual images | \n",
" expected images | \n",
" difference | \n",
"
\n",
" \n",
" \n",
" \n",
" | 0 | \n",
" 31415 | \n",
" 31485 | \n",
" 70 | \n",
"
\n",
" \n",
"
\n",
"
\n",
"
\n",
" \n",
" \n",
" | \n",
" count | \n",
" data type | \n",
" number of null values | \n",
"
\n",
" \n",
" \n",
" \n",
" | ID | \n",
" 31485 | \n",
" int64 | \n",
" 0 | \n",
"
\n",
" \n",
" | Site Name | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Sampling Unit Name | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Latitude | \n",
" 31485 | \n",
" float64 | \n",
" 0 | \n",
"
\n",
" \n",
" | Longitude | \n",
" 31485 | \n",
" float64 | \n",
" 0 | \n",
"
\n",
" \n",
" | Sampling Period | \n",
" 31485 | \n",
" float64 | \n",
" 0 | \n",
"
\n",
" \n",
" | Photo Type | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Photo Date | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Photo Time | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Raw Name | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Photo Notes | \n",
" 0 | \n",
" float64 | \n",
" 31485 | \n",
"
\n",
" \n",
" | Photo ID URL | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Class | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Order | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Family | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Genus | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Species | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Number of Animals | \n",
" 31485 | \n",
" int64 | \n",
" 0 | \n",
"
\n",
" \n",
" | Person Identifying the Photo | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Camera Serial Number | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Memory Card Serial Number | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Camera Notes | \n",
" 4503 | \n",
" object | \n",
" 26982 | \n",
"
\n",
" \n",
" | Camera Start Date and Time | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Camera End Date and Time | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Person setting up the Camera | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Person picking up the Camera | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Camera Make | \n",
" 25824 | \n",
" object | \n",
" 5661 | \n",
"
\n",
" \n",
" | Camera Model | \n",
" 25824 | \n",
" object | \n",
" 5661 | \n",
"
\n",
" \n",
" | Flash | \n",
" 25824 | \n",
" float64 | \n",
" 5661 | \n",
"
\n",
" \n",
" | Exposure Time | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Sequence Info | \n",
" 27212 | \n",
" object | \n",
" 4273 | \n",
"
\n",
" \n",
" | Triplet Number | \n",
" 27212 | \n",
" object | \n",
" 4273 | \n",
"
\n",
" \n",
" | Moon Phase | \n",
" 27212 | \n",
" object | \n",
" 4273 | \n",
"
\n",
" \n",
" | Temperature | \n",
" 27212 | \n",
" float64 | \n",
" 4273 | \n",
"
\n",
" \n",
" | Camera Firmware Version | \n",
" 21551 | \n",
" object | \n",
" 9934 | \n",
"
\n",
" \n",
" | Protocol Version | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Data Set Creator Institutions | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Data Set Creator Scientist | \n",
" 29037 | \n",
" object | \n",
" 2448 | \n",
"
\n",
" \n",
" | Data Set Contact | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
" | Data Level | \n",
" 31485 | \n",
" object | \n",
" 0 | \n",
"
\n",
" \n",
"
\n",
"