AI & ROBOTICS HACKATHON 2022 - 18-19/6/2022**

GROUP: KLPM-Untouchables

Members:

• Mamadou Keita
• Ningyang Li
• Pham Duc Thinh
• Mohib Ullah

Dataset

The dataset from VARUNA company, image of Satelite.

Purpose:

Crop Classification problem. We have 4 classes for classification:

Preprocessing-data

Step 1: Statistical the data => Imbalance data

Step 2: Balance data

Model

Environment: on Google Colab environment

Library:

geopandas, rasterio, libpysal, contextily, mapclassify,etc

  https://pypi.org/project/rasterio/

Training + validation:

Step 1: Split into 20% for validation and 80% for training

Step 2: Select model: HyperResNet34 with CBAM block => get train accuracy and validation accuracy: ~95%

Accuracy: 97.583 | Jaccards: 0.954 | F1 Score: 0.976

Testing:

Send the result to VARUNA team.

Result for the folder 20211028 can be find in KLPM_Scores.csv