# Identifying Land Patterns from Satellite Imagery in Amazon Rainforest using Deep Learning

**Authors:** Somnath Rakshit, Soumyadeep Debnath, Dhiman Mondal  
**arXiv:** [1809.00340](https://arxiv.org/abs/1809.00340)  
**Date:** 2018-09-02  
**Summary page:** https://www.summarizepaper.com/en/paper/1809.00340v1/

## Key points

- The Amazon rainforests are experiencing damage due to natural and artificial means
- Forest cover equivalent to 48 football fields is lost every minute globally
- Deforestation in the Amazon has led to a reduction in biodiversity, loss of habitat, climate change, and other biological losses
- Image classification using deep learning techniques can track changes in the nature of these forests over time
- Convolutional neural networks (CNNs) can automate the task of classifying each image and speed up this process
- Researchers have demonstrated how CNNs can be used to track changes in land patterns in the Amazon rainforests with an impressive testing accuracy of 96.71%
- Deep convolutional networks gained popularity after Krizhevsky et al.'s AlexNet won the ImageNet challenge by a significant margin in 2012
- Transfer learning is a useful technique for speeding up development using neural networks by reusing weights from models trained on specific datasets and retraining only the last layer before passing its output through a classifier
- Researchers have used various data augmentation and ensemble techniques with different models such as ResNet50 and deep residual architecture to classify land patterns in the Amazon rainforests using CNNs
- The dataset for this work was derived from Planet's full-frame analytic scene products using its four-band satellites in sun-synchronous orbit (SSO) and International Space Station (ISS) orbit.
- Tracking changes in land patterns in the Amazon rainforests is crucial for preserving biodiversity and mitigating climate change effects effectively.
- The use of CNNs and other deep learning techniques can help automate this task, making it faster and more accurate while providing more reliable results than manual methods or traditional machine learning algorithms alone could provide.

## Summary

The Amazon rainforests have been experiencing widespread damage due to both natural and artificial means. According to estimates, the world loses forest cover equivalent to 48 football fields every minute. Deforestation in the Amazon has led to a significant reduction in biodiversity, loss of habitat, climate change, and other biological losses. Therefore, it is essential to track changes in the nature of these forests over time. One way to do this is through image classification using deep learning techniques. Convolutional neural networks (CNNs) can be used to speed up this process by automating the task of classifying each image. In recent work, researchers have demonstrated how CNNs can be used to track changes in land patterns in the Amazon rainforests with an impressive testing accuracy of 96.71%. Neural networks have been successfully applied to a wide range of problems such as diabetes prediction, face detection, object localization, and visual image recognition. Deep convolutional networks gained popularity after Krizhevsky et al. 's AlexNet won the ImageNet challenge by a significant margin in 2012. Since then, all winners of ImageNet challenges have used deep convolutional neural network architecture. Transfer learning has also emerged as a useful technique for speeding up development using neural networks. It involves reusing weights from models trained on specific datasets and retraining only the last layer before passing its output through a classifier. To classify land patterns in the Amazon rainforests using CNNs, researchers have used various data augmentation and ensemble techniques with different models such as ResNet50 and deep residual architecture. The dataset for this work was derived from Planet's full-frame analytic scene products using its four-band satellites in sun-synchronous orbit (SSO) and International Space Station (ISS) orbit. Tracking changes in land patterns in the Amazon rainforests is crucial for governments and other agencies concerned about preserving biodiversity and mitigating climate change effects effectively. The use of CNNs and other deep learning techniques can help automate this task, making it faster and more accurate while providing more reliable results than manual methods or traditional machine learning algorithms alone could provide.