geospatial artificial intelligence: konsep dan aplikasinya
TRANSCRIPT
Geospatial Artificial
Intelligence: Konsep
dan Aplikasinya Pada
Daerah Tropis
Dr. Edy Irwansyah
(Bina Nusantara University)
21.07.2020
Outline
Geospatial Artificial Intelligence (GeoAI)01
History of GIS and Artificial intelligence02
Machine Learning and Deep Learning in GeoAI 03
Deep Learning in Geospasial Information
2
04
Aplikasi GeoAI pada Daerah Tropis05
Geospatial Artificial Intelligence (GeoAI)
01.1
3
Geospatial Artificial Intelligence
(Geo-AI): is an emerging scientific
discipline that combines innovations
in spatial science, artificial
intelligence methods in machine
learning (e.g., deep learning), data
mining, and high-performance
computing to extract knowledge from
spatial big data (Vopham et al, 2018)
1. Geographic Information
System: is a computer-based
system to aid in the collection,
maintenance, storage, analysis,
output, and distribution of spatial
data and information (Bolstad,
2016).
Geospatial Artificial Intelligence (GeoAI)
01.2
4
2. Artificial Intelligence Ilmu pengembangan intelligence agents
“.... any device that perceives its
environment and takes actions that
maximize its chance of successfully
achieving its goals” (Poole, Mackworth & Goebel, 1998).
“...machines (or computers) that mimic
cognitive functions that humans
associate with the human mind, such as
learning and problem solving“ (Russell
& Norvig , 2009).
Geospatial Artificial Intelligence
(Geo-AI): is an emerging scientific
discipline that combines innovations
in spatial science, artificial
intelligence methods in machine
learning (e.g., deep learning), data
mining, and high-performance
computing to extract knowledge from
spatial big data (Vopham et al, 2018)
Geospatial Artificial Intelligence (GeoAI)
01.3
5
Geo-AI: a subfield of spatial data science utilizes advancements in
techniques and data cultures to support the creation of more
intelligent geographic information as well as methods, systems, and
services for a variety of downstream tasks.
(1) Image classification, (2) Object detection, (3) Segmentation, (4)
Simulation and interpolation, (5) link prediction, (6) natural language-
based retrieval and question answering, (7) on-the-fly data integration,
(8) geo-enrichment, and many others
Janowicz, K., Gao, S., McKenzie, G., Hu, Y., & Bhaduri, B. (2020). GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge
discovery and beyond.
01.4
6
Geospatial Artificial Intelligence (GeoAI)
GeoAI: It is a spatial data processing and
analysis algorithm that integrates AI, and
is the product of AI and GIS.
AI for GIS: Using AI capabilities to
enhance the functions and user
experience of GIS software.
GIS for AI: Using visualization and
analysis technology of GIS to perform
spatial visualization and further spatial
analysis of AI output results.
https://www.supermap.com/
History of GIS and Artificial Intelligence
02.1
7
▪ ArcGIS Pro 1.0 was first
released in Jan 2015
▪GeoAI in Esri Development
Summit 2018
02.2
8
History of GIS and Artificial Intelligence
Machine Learning and Deep Learning in GeoAI
03.1
9
Source: Ragh, 2019 Source: Rose, 2019
03.2
10
Machine Learning and Deep Learning in GeoAI
Tom M. Mitchel, 1997 mengusulkan sebuah
definisi yang lebih operasional dimana
machine learning didefinisikan sebagai
algoritma yang memiliki kemampuan:
".... learn from experience E with respect to
some class of tasks T and performance measure
P, if its performance at tasks in T, as measured
by P, improves with experience E.”Gambar Perbedaan Pemrograman Tradisional dan
Machine Learning
(Sumber: Francois, 2017)
03.3
11
Machine Learning and Deep Learning in GeoAI
Gambar Metode Pembelajaran Machine Learning
03.4
12
Machine Learning and Deep Learning in GeoAIMetode Pembelajaran Machine Learning
Gambar Supervised Learning
(Sumber: Francois, 2018)
1) Supervised learning:
Gambar Unsupervised Learning
(Sumber: Francois, 2018)
2) Unsupervised learning: 3) Reinforcement learning:
Gambar Reinforcement Learning
(Sumber: Francois, 2018)
03.5
13
Machine Learning and Deep Learning in GeoAIMetode Pembelajaran Machine Learning
1) Supervised learning: 2) Unsupervised learning:
3) Reinforcement learning:
http://www.datasciencelovers.com/
03.6
14
Machine Learning and Deep Learning in GeoAI
Definisi deep learning oleh Chollet Francois (Francois, 2018) sebagai:
“a new take on learning representations from data that puts an emphasis on learning
successive layers of increasingly meningful representation.”Deep learning merupakan sebuah metode pembelajaran terhadap data yang
bertujuan untuk membuat representasi (abstraksi) data secara bertingkat
menggunakan sejumlah layer pengolahan data. Hal penting dari deep learning,
(LeCun, Bengio, & Hinton, 2015) menekankan bahwa representasi data tersebut tidak
dibuat secara eksplisit oleh manusia tetapi dihasilkan oleh sebuah algoritma
pembelajaran.
03.7
15
Machine Learning and Deep Learning in GeoAIMetode Pembelajaran Deep Learning
Ma, L., Liu, Y., Zhang, X., Ye, Y., Yin, G., & Johnson, B. A. (2019). Deep learning in remote sensing applications: A meta-
analysis and review. ISPRS journal of photogrammetry and remote sensing, 152, 166-177.
03.8
16
Machine Learning and Deep Learning in GeoAIPerbedaan Machine Learning dan Deep Learning
Simplicity: data yang dipergunakan sebagai input proses pembelajaran tidak
membutuhkan proses rekayasa fitur sebelumnya
Scalability: proses pembelajaran model deep learning memungkinkan implementasi
secara paralel sehingga bisa memanfaatkan kapasitas Graphics Processing Units (GPU)
atau Tensor Processing Units (TPUs)
Versatility (adaptability): proses pembelajaran model
deep learning tidak selalu harus dilakukan dari awal
Reusability: sebuah model hasil yang telah di-training
menggunakan sebuah dataset berskala besar dapat
dipergunakan untuk melakukan tugas menggunakan
dataset berskala lebih kecil.Perbedaan Machine Learning (a) dan Deep Learning (b)
(Sumber: Moons, Bankman & Verhelst, 2019)
03.9
17
Machine Learning and Deep Learning in GeoAI
Model NN dan Convolutional NN (CNN) memiliki persamaan yaitu keduanya
memiliki fully connected network, yaitu struktur jaringan nodes yang saling
terkoneksi
Arsitektur CNN memiliki kemampuan
menangkap informasi kontekstual yang
terkandung didalam data, misalnya: pixel yang
saling berdekatan didalam sebuah citra atau
kata-kata yang berdekatan didalam sebuah
text
CNN memiliki kompleksitas yang lebih rendah, waktu training model yang
lebih cepat, dan membutuhkan jumlah sampel data training lebih sedikit
dari model NN.
Gambar Arsitektur Model LeNet5
(Sumber: LeCun et al., 1998)
04.1
18
Deep Learning in Geospatial Information
Use Case in Geospatial Information
Semantic Segmentation(Qin et al, 2019)
Instance Segmentation(Ji et al, 2019)
Object Detection(Prathap and Ilya, 2018)
04.2
19
Deep Learning in Geospatial Information
1. Semantic Segmentation
• Proses untuk memberikan label
semantik atau kelas objek (misalnya:
sungai, pesawat, bangunan, jalan,
pohon) terhadap setiap pixel dari
sebuah citra
• Hasil: Superpixel
04.3
20
Deep Learning in Geospatial Information
•Model U-Net
• Diusulkan oleh Ronneberger, Fischer & Brox, 2015, sebagaimodel segmentasi semantik
04.4
21
Deep Learning in Geospatial Information•Model DeepLab
• Diusulkan oleh Chen et al., 2017 dari Google sebagaimodel segmentasi semantik
(Su and Chen, 2020)
04.5
22
Deep Learning in Geospatial Information•Model Deep UNet
• Diusulkan oleh Li et al., 2018, sebagaimodel segmentasi semantikmenggunakan data citra penginderaanjauh.
04.6
23
Deep Learning in Geospatial Information• Model High Resolution Nets
• Diusulkan oleh Zhang, Lin, Ding & Bruzzone, 2020, sebagai model segmentasi semantikmenggunakan data citrapenginderaan jauh.
04.7
24
Deep Learning in Geospatial Information
2. Instance Segmentation
• Proses pemberian label,
prediksi lokasi dan
segmentation mask berbasis
pixel terhadap setiap individu
objek didalam sebuah citra.
04.8
25
Deep Learning in Geospatial Information•Model Mask R-CNN
• Diusulkan oleh He, Gkioxari, Dollár, & Girshick, 2017, sebagai model segmentasiinstance
04.9
26
Deep Learning in Geospatial Information
•Model Path Aggregation Net
• Diusulkan oleh (Liu et al., 2018) sebagai model segmentasiinstance
04.10
27
Deep Learning in Geospatial Information
3. Object Detection
• is a computer technology
related to computer vision and
image processing that deals
with detecting instances of
semantic objects of a certain
class (such as humans,
buildings, or cars) in digital
images and videos.
04.11
28
Deep Learning in Geospatial Information
•Model Spatial Pyramid
Pooling (SPP)-Net
• Diusulkan oleh (He et al., 2015) sebagai model Object detection
04.12
29
Deep Learning in Geospatial Information
Dua tipe frameworks di Object Detection: region proposal-based and regression/classification based
(Zhao and Zheng, 2019)
05.1
30
Aplikasi GeoAI Pada Daerah Tropis1. Semantic Segmentation
Semantic Segmentation
using U-Net and
U-Net+ResNet-18
05.2
31
Aplikasi GeoAI Pada Daerah Tropis
Semantic Segmentation
using ResNet and SatNet
1. Semantic Segmentation
SatNet Architecture
(Gunawan et al, 2019)
05.3
32
Aplikasi GeoAI Pada Daerah Tropis
Instance Segmentation
using PSPNet, Unet and
Classification Using
ResNet50
2. Instance Segmentation
Hasil Awal PSPNet
Hasil Awal UNet
Hasil Klasifikasi (1 Klas)
Hasil Klasifikasi (2 Klas)
05.4
33
Aplikasi GeoAI Pada Daerah Tropis3. Object Detection
Object Detection using
AlexNet dan ResNet18
Link video
https://youtu.be/suL1YWbchOA
AlexNet Architecture
(Krizhevsky et al, 2012)
05.5
34
Aplikasi GeoAI Pada Daerah Tropis3. Object Detection
Object Detection using
ResNet with SSD
Architecture
https://learn.arcgis.com/
00.0
35
Terima kasih
36