You are on page 1of 31

“VIII Conferencia por el día Mundial del Medio Ambiente”

DETECCIÓN DE LA TALA CON


DEEP LEARNING

Mg. Ing. Lucio Villa


GIS & Remote Sensing Specialist
E-mail: luciovilla@lamolina.edu.pe 1
Estructura de la Presentación

1. Introducción
2. Teledetección e Inteligencia Artificial
3. Deep Learning para la Detección de Tala

2
1. INTRODUCCIÓN
1. Introducción

• Tala: actividad selectiva.


• Zonas de difícil acceso.
• Imperceptibles para muchas
imágenes satelitales.
• Más acceso a mejores imágenes
aerosatelitales (Big Earth Data) y
recursos informáticos (Cloud
Computing).
• Evolución de la Inteligencia Artificial.
4
https://maaproject.org/2019/detect_illegal_logging/

5
2. Teledetección e Inteligencia Artificial
Teledetección e Inteligencia Artificial

7
8
Principales sensores remotos satelitales

9
Identificación de Patrones espaciales

10
From Artificial Intelligence to Deep Learning

What is Intelligence?

“... when a washing


machine...can decide on the
right amount of water to use
and on the required time for
soaking, washing, and
spinning”

Reference: Moolayil J.(2019). Learn Keras for Deep Reference: Babu, A.D. (Nov 5, 2019). Artificial Intelligence vs Machine Learning vs Deep Learning (AI vs ML vs DL)
Neural Networks https://medium.com/@alanb_73111/artificial-intelligence-vs-machine-learning-vs-deep-learning-ai-vs-ml-vs-dl-e6afb7177436

11
From Artificial Intelligence to Deep Learning
- Artificial intelligence (AI): Techniques which allows
to induce intelligence .
… with or without explicit programming (e.g. If-Else)

Password == OK Balance > $ 20

12
From Artificial Intelligence to Deep Learning
- Artificial intelligence (AI): Techniques which
allows to induce intelligence .
… with or without explicit programming (e.g. If-Else)
- Machine Learning (ML): Algorithms which enable
computer systems to infer patterns (to learn) from
data (without explicit programming)
Supervised / Unsupervised/ Reinforcement
learning (e.g. RF, NN, SVM)

Techniques applied in Remote Sensing Analysis


(1990 - 2021)

13
From Artificial Intelligence to Deep Learning
- Artificial intelligence (AI): Techniques which allows
to induce intelligence .
… with or without explicit programming (e.g. If-Else)
- Machine Learning (ML): Algorithms which enable
computer systems to infer patterns (to learn) from
data (without explicit programming)
Supervised / Unsupervised/ Reinforcement learning (e.g.
RF, NN, SVM)
- Deep Learning (DL): Algorithms inspired by - our understanding of-
the biological functioning of the human brain to obtain knowledge.

14
From Artificial Intelligence to Deep Learning
- Artificial intelligence (AI): Techniques which allows
to induce intelligence .
… with or without explicit programming (e.g. If-Else)
- Machine Learning (ML): Algorithms which enable
computer systems to infer patterns (to learn) from
data (without explicit programming)
Supervised / Unsupervised/ Reinforcement learning (e.g. RF, NN, SVM)
- Deep Learning (DL): Algorithms inspired by - our understanding of-
the biological functioning of the human brain to obtain knowledge.
Neural Network Architecture (e.g. Classifying an image as a cat or dog or
Building a “face of a person who doesn’t exit”)

15
https://thispersondoesnotexist.com/
16
Deep Learning y Computer Vision

Object Detection Semantic Segmentation


https://towardsdatascience.com/semantic-segmentation-with-deep-learning-a-guide-and-code-e52fc8958823
17
Input Data for training a Model
- Benchmark dataset: datasets that
represent real-world data science problems
and it’s used to training a ML model.
… of all sizes with various levels of difficulty.

- Satellite image: images of Earth


collected by imaging satellites. It’s
minimum unit it’s called pixel.
Reference: Mikaela Weisse (Mar 2, 2021). New Radar Alerts Monitor Forests
Through the Clouds: https://blog.globalforestwatch.org/data-and-
research/radd-radar-alerts/
… some examples include Sentinel-2, Landsat-8,
Planet, SkySat.

18
Deep Learning Models
- Neural network: computational model
inspired by the structure of the human
brain.
… used for predictive modeling and applications
where they can be trained via a dataset

- Convolutional neural network: type of


neural network generally used in images,
able to extract the features of them.
… like borders, line colors, etc.
Reference: Madhivarman (May 30, 2018). How to calculate the number of parameters in
the CNN?:
https://medium.com/@iamvarman/how-to-calculate-the-number-of-parameters-in-the-
cnn-5bd55364d7ca

19
3. Deep Learning para la Detección de Tala
ACCA Empoderamos a la
• Organización peruana sin fines de lucro que, gente
desde 1999, trabaja en la conservación y
puesta en valor de los bosques amazónicos. Protegemos al medio
• Genera alternativas sostenibles para los ambiente
pobladores y apoya la educación de niños, jóvenes
y adultos. Ponemos la ciencia a
• Pioneros en el trabajo para la protección de la trabajar
diversidad biológica de uno de los espacios más
ricos del planeta: la región Andino – Amazónica sur
peruana.
• Emplean la ciencia para guiar el enfoque
multifacético hacia la conservación ambiental y la
restauración de los bosques del Perú.

21
Methodology

22
Methodology
Google Colab

- For the code generation (Tensorflow)

Google Cloud Platform

- For storing and deploying the model

Google Earth Engine

- For the visualization

23
CASE 1: Logging and Roads

24
CASE 1: CNN applied to Sentinel-2 Imagery (10 m)

Logging
Roads

25
CASE 2

PlanetScope PlanetScope
8-Jun-2020 17-Aug-2020

26
CASE 2

PlanetScope SkySat
09-Oct-2020 17-Oct-2020

27
28
CNN applied to Planet Imagery (NICFI - 4.7 m)

29
GRACIAS, PREGUNTAS?
“VIII Conferencia por el día Mundial del Medio Ambiente”

DETECCIÓN DE LA TALA CON DEEP


LEARNING

Mg. Ing. Lucio Villa


GIS & Remote Sensing Specialist
E-mail: luciovilla@lamolina.edu.pe
31

You might also like