Landslide4Sense( Landslide Detection using Deep Learning )

Landslide4Sense( Landslide Detection using Deep Learning )

LandSlide4Sense

"Landslides are a natural phenomenon with devastating consequences, frequent in many parts of the world. Thousands of small and medium-sized ground movements follow earthquakes or heavy rainfalls. Landslides have become more damaging in recent years due to climate change, population growth, and unplanned urbanization in unstable mountain areas. Early landslide detection is critical for quick response and management of the consequences. Accurate detection provides information on the landslide's exact location and extent, which is necessary for landslide susceptibility modeling and risk assessment.

Recent advances in machine learning and computer vision combined with the growing availability of satellite imagery and computational resources have facilitated rapid progress in landslide detection. Landslide4Sense aims to promote research in this direction and challenges participants to detect landslides around the globe using multi-sensor satellite images. The images are collected from diverse geographical regions offering an important resource for remote sensing, computer vision, and machine learning communities."


Devoted myself into AI field in Geodata, I expand my knowledge and experience into disaster field of study in this case landslide. Despite ranking not too good, I am really thrilled as this is my first solo project in international challenge of GeoAI field.

The below are the approach that I have used for this particular challenge.

No alt text provided for this image


The Data

The Landslide4Sense data consists of the training, validation, and test sets containing 3799, 245, and 800 image patches, respectively. Each image patch is a composite of 14 bands that include:

  • Multispectral data from?Sentinel-2: B1, B2, B3, B4, B5, B6, B7, B8, B9, B10, B11, B12.
  • Slope data from?ALOS PALSAR: B13.
  • Digital elevation model (DEM) from ALOS PALSAR: B14.


All bands in the competition dataset are resized to the resolution of ~10m per pixel. The image patches have the size of 128 x 128 pixels and are labeled pixel-wise.

Datasets come in h5 format for both training data and label data. These are being read with h5py python modules and were converted into numpy array for better manipulation.

Out of 14 bands, the below are the first 3 bands of training data tagging together along with label data.

No alt text provided for this image

Bands 2 and 4 of sentinel 2 were chosen with slope data for features. Features were normalized and nulls were removed. Better feature engineering could be done here for better model performance such as indices calculations (NDVI,NDWI,EVI,etc).


Binary cross entropy loss is used for Unet model in tensorflow framework with metrics such as F1, precision and recall are used . Multiple tensorflow frameworks were tested such as CNN Unet, Res34Unet , InceptionV3Unet. Amongst them Res34Unet show the best results. the below are the predictions and classification reports.

No alt text provided for this image


Predictions are as follow:

No alt text provided for this image
No alt text provided for this image
No alt text provided for this image
No alt text provided for this image


Classification report is as follow :

precision :72.67778515815735

f1 score :52.74485945701599

recall :57.87952542304993

Loss :0.24824434518814087

accuracy : 98.6082136631012

And again thanks for your time to read my articles. I hope the slight work I have done would help you out in your research, papers or work. Please feel free to connect me or discuss with me at [email protected].

要查看或添加评论,请登录

Zaw Thu Htet ( Toby )的更多文章

社区洞察

其他会员也浏览了