.rst .pdf

Kano & Yoshii river: Land cover classification using machine learning

---

Overview

In this project, I try to develop supervised machine learning model in Python to classify the land cover type within the river channel. The model is developed by using the training data from 4 seasons of PLANET satellite images with 3-m spatial resolution at Kano and Yoshii river in Japan.

The land cover is classified into 7 clases:

1. Herbecous

2. Shrub

3. High tree

4. Bamboo

5. Other vegetation

6. Built-up

7. Water surface

The input data for model development is the combination of Kano river and Yoshii river, and totally consists of 1,845,237 data or pixels. Then this data is split into 75% for training and 25% for testing as follows:

• Training data: 1,383,093 pixels

• Testing data: 462,144 pixels

Figure 1: Training data of Kano river.

Figure 2: Training data of Yoshii river.

Codes for model development:

• 1. Prepare input data of Kano River
• 2. Prepare input data of Yoshii River
• 3. Supervised learning model development
• 4. Predicting the whole Kano and Yoshii River

Visualize result

Figure 3: a. Predicted land cover of Kano river b. Predicted land cover of Yoshii river.

Figure 4: a. Training land cover of Kano river b. Training land cover of of Yoshii river.

Figure 5: a. Predicted land cover of Kano river b. Predicted land cover of of Yoshii river.

previous

2022

next

1. Prepare input data of Kano River