-
Notifications
You must be signed in to change notification settings - Fork 2
/
instructor_notes_raster_classification.Rmd
74 lines (47 loc) · 2.46 KB
/
instructor_notes_raster_classification.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
---
title: 'Instructor Notes: Raster Classification with Hurricane Rita'
output:
pdf_document: default
html_document: default
---
## Where?
Rita passed over the Gulf of Mexico in 2005 and headed north, impacting Texas and Louisiana most directly.
We will be looking at a small portion of land on the border between Louisiana and Texas for today's lesson.
## Brick
A raster brick is a multi-layer raster object. It's used because it decreases processing time.
There are 7 bands in this raster here.
## MODIS bands
Lines 48 - 60:
NIR = near infrared
SWIR = shortwave infrared - useful for looking at water-logged ground.
MNDWI = modified normalized difference water index (one of many water indicies)
NDVI = normalized difference vegetation index
Spectral bands from MODIS are sensed (raw) at 500m, but for this class they have bene processed to 1 km to agree with scale of other products.
Using bands allows you to create your own indicies, but comes with challenges.
## Writing out rasters
Lines 78 - 86:
Need to know about datatype when writing out a raster.
## sf = simple features
Lines 94 - 96:
We are using the st_read() function, which is from the sf package. It reads in the files as dataframes with a
geometry column. Therefore, we can use simple functions like rbind() to manipulate files.
-------------
## CART:
Classification and Regression Trees
Split the data by minimizing the variance in the branches to build the tree.
Unsupervised I think - attempts to find natural clustering as stated above.
\pagebreak
## SVM:
Supervised Learning Models
Analyze data used for classification and regression. Performs probabilistic binary linear classification.
Tries to split the data with widest gap possible, i.e the maximum distance between data points of both classes.
Finds an optimal solution.
Hyperplanes are decision boundaries that help classify the data points. These are lines in 2D, and planes in 3D.
Good visuals here:
https://towardsdatascience.com/support-vector-machine-introduction-to-machine-learning-algorithms-934a444fca47
Support vectors are the data points that lie closest to the decision surface (or hyperplane).
-------------
## Confusion Matrix:
Counts up the number of pixels in each class. What's correctly classified is on the diagonal - what's off diagonal is incorrectly classified.
## Accuracy metrics
Overall accuracy can be mis-leading. Need to look at the map, or other metrics of accuracy such as producer's accuracy.