83Figure 5.17: Pencil Sketch and CNN Model predict Airport satellite image.. 86 Figure 5.18: Pencil Sketch and CNN Model predict Bridge satellite image.. 87 Figure 5.19: Pencil Sketch an
INTRODUCTION ccc ecceceeeeesesecesseeeseeececeeeeseceaeceeseeeaeeeaeseeeeaeeeneeees 10 1.1 Introduce thesis - c5 E1 111210111930 1910 9109 ngờ 10 IV.) 5 11 I9) 0n 11 Chapter 2 LITERATURE REVIEW HH HH HH HH ng Hư 13
General information about the dataset .- SH g niệ, 19 3.2 Specific information Of Objects - 5 2S 1321113111131 1 181111 1 re 20 KÝ Y0
In order to create a foundation and initial ideas to build and develop a data set for research, we decided to start with some self-proposals The first data we collected were 15,418 satellite images including 575 infrastructures (address) belonging to 11 different types (common names) and 5 different levels (classifications to cover common names).
The objects of collection are types of infrastructure in the territory of Vietnam to create a connection between the topic and the people in the country However, during the collection process, we encountered some limitations such as the location of dissolved infrastructure and the lack of exact coordinates, so we ignored it even though the list was large.
> Distribution ts i: +, Ne ằ- [International airport] +ằ [Local airport ]
Map: Created by Open Development Vietnam - Created with Datawrapper
Figure 3.2: Map of airports in Viet Nam.Š
3 Figure is taken from the website https://www.datawrapper.de/_/GfKOi/
Figure 3.4: Map of airports in Viet Nam.4
4 Figure is taken from the website https://database.earth/energy/power-plants/hydro-power/vietnam
Infrastructure type: Production and Industrial
> Distribution or Gone Wren 5a Tia
‘Thue Thanh BÍ C140 Tin Sa {420} en Ptosl (351)
Dal ae 460A) Mar Sach [tả]
- - au Thay areata “® h tan DeLuca Di Út
Sai bong B (AE) —- —— 5 : i S| eset | Cong Hoa C8i Lint 13571 |
_ | sendra: Hal Prong 1511 py Lai Gann (132) BÍ: | an Giang (09: Tân fruarng (129 S— | aes Cau thee me Met hoa - kerergrk (M6)
Dink deen ISA) Bạp Neo 5001.
‘eae EEA) lore Mau (1423 b h k Ge Dae [1EA]
Mae Chánh tiệc) i mil] He Mai L437) Tab Dae (335) = —HỊ mini
Tan fin (Lee) = Lãng Thay (4a
‘Trg Dae - Ben Lec (£34) MẸ - Leng Bình (001
_An Whut Tae (320) Nho Trace Ix)
Thạnh Duc (255) aati : ˆ Who Tewen I 648)
“lang Dien (538) i foe Tr tb Put
{pitt | Nhơn Trae’ MỊ 990) hw Wy 24) ‘One kea B2] n Tan Phk lá)
‘Chas Duc (Li501 lang San (2250) Thanh Pha {1771
Phụ My ML fo Pen toc test:
Figure 3.6: Vietnam industrial zone's map.”
> Figure is taken from the website https://kizuna.vn/en/news/difficulty-in-investing-in-vietnams-largest- industrial-zone-1020
Figure 3.8: Vietnam stadium map for V-League.
6 Figure is taken from the website https://www.google.com/maps/d/viewer?mid=1NENwlU-
AzvOSxISKXAHILOOpuwA &hl=en_US&Il.18467271686617%2C 108.42166787061942&z=6
Figure 3.10: List of water treatment plant.’
7 Figure is taken from the website https://www.unido.or.jp/files/sites/2/Water-and-Wastewater-Management- in- Vietnam.pdf
Figure 3.12: Bridges of Viet Nam.Š
8 Figure is taken from the website https://www.google.com/maps/d/viewer?mid- csFaOkCWBSRY607T0tgN1dafA &hI=en_US&II.726609164768014%2C107.68833478135812&z=6
| Maomin Rit wee \ ' 9 ¢ Hong Kong ishuangbanna b ˆ 7 i
Thailand SURIN UB0N NAKHON RATCHATHANI
KHIRI'KHAN \ LX ụ jl Et hang 8lbnfi
Map data £ 3 Google, Map Mobility imagery ©2023 TetralMe
Figure 3.14: Wind farm power plants in Viet Nam.?
? Figure is taken from the website https://www.google.com/maps/d/viewer?mid=1 NHY5MwPrzOgHvKXpWI13D9aDulas&hl=en_US&ll. 676849959756382%2C 105.266940592727 15&z=6
Chuỗi Công viên phan mềm Quang Trung _—.
Trung tâm Công nghệ thong tin
HUE All tinh Thtfa Thiên Huế
Trung tam Công nghệ thông tin - Truyễn thông tinh Binh Binh
“Quilty eehseuionconpies Thua Thien Hue Center of Information
Technology - Hue CIT Information and Communication
- Technology Center of Binh Dinh Province - ICTBD
Khu Công nghệ phan mễm Đọi học Quốc Gio TP.HCM
Công viên phdn mém Quan dội tinh Khanh Hòa
Công viên phồn mềm Mekong tinh Tiền Giang
Mekong Innovation Technology Park - Mekong ITP ° s_a Thanh viên chính thức
Decal Member Khu Công nghệ số tinh Hau Giang xã Thanh viên dự liển Hau Giang Digital Technology Park Prospective Member Ea ee Thanh viên tiém năng
Khu Công viên phdn mềm Quang Trung ` = - mm ^
Quality Tech Solution Complex - QTSC ie Trung tam Công nghệ phan mém Pai học Can Thơ
@ Q1SC al "7 ae ` Cantho Universitu Software Center - CUSC ` — E °
Figure 3.16: Map of technology parks in Viet Nam.!9
!9 Figure is taken from the website https://www.google.com/maps/d/viewer?mid=1 NHY5MwPrzOgHvKXpWI13D9aDulas&hl=en_US&ll.
Figure 3.18: Students gather in shape of Vietnam map.
!! Figure is taken from the website https://news.baokhanhhoa.vn/socio_politic/202203/1000-students- gather-in-shape-of-vietnam-map-8246482/
Biển Bong quan đảo Trưởng Sa
Figure 3.20: List of water reservoir in Vietnam.!2
Figure is taken from the website https://www.touristlink.com/vietnam/cat/lakes/map.html
TUYẾN MỖI CHỢ QUANG THÁI
UU TIEN PHAT TRIEN ằ ———— QUANG BINH
QUANG TR! À HUE HOANG SA
Hướng tuyến dự kiến theo quy hoạch: mee DL[ÙII da Khai thác BINH ĐỊNH
Dự kiến hoàn thành ses BUÔN MA —-e Sau năm 2023 THUỘT sấi 4° NHA TRANG
VÀNH DAIS oem PHAN THIẾT
CHAU ĐỐC —a h PX VŨNG TAU one eth —— BEN LUC - LONG THÀNH ví 6 STRUNG LƯƠNG - MỸ THUẬN
LỘ TẺ - RẠCH SỎI “MY THUẬN - CAN THƠ
`SÓC TRANG oa TRƯỜNG SA ° CAN THƠ - CÀ MAU
Figure 3.22: List of water reservoir in Vietnam.
'3 Figure is taken from the website https://www.dothanhauto.com/he-thong-duong-cao-toc-viet-nam-den- nam-2025
3.3 Evaluation of collected data sets
3.3.1 Evaluation of data sets collected from sources
In the course of this research, we utilized satellite imagery data sources and fundamental classification challenges available on the Kaggle platform [8] This provided a valuable opportunity to delve into deep learning methodologies and rudimentary image processing techniques Inspired by the nation of our origin, Vietnam, we endeavored to locate relevant satellite imagery datasets Our efforts led us to GeoViet Consulting [9], offering a variety of satellite data including MODIS, LandSat, SPOT, ASTER, and THEOS images Upon thorough evaluation, we selected the MODIS and ASTER datasets pertaining to Vietnam Regrettably, access to these datasets was hindered due to non-functional links This obstacle prompted the pursuit of alternative sources, which ultimately did not yield results. Consequently, this research pivoted to developing a novel dataset, focused on Vietnam's infrastructure, thereby contributing to the broader field of satellite imagery data.
3.3.2 Essential in building new datasets
Identify more diverse and detailed infrastructure by combining level division and storing images of object directions, this helps us identify insects of diverse types such as: Airports, Industrial parks, Schools,
Increased accuracy for infrastructure recognition in new datasets: The model can specifically recognize certain types of images Although some datasets have classification levels only at infrastructure types without lower infrastructure classification levels, they contain satellite images with good identification capabilities that other datasets do not have.
The dataset will be able to evolve to a higher level of infrastructure satellite image classification for improved infrastructure identification and classification. The data set clearly classifies infrastructure at levels such as: Transportation,
Energy, Water, Production and Industrial This can help the dataset assign more infrastructure classification labels to higher-level subclasses, not just limited to the distinction level.
We initially collected and searched 11 types of infrastructure with different classification levels The data is fragmentary and has not been reorganized and has no connection.
The objects of collection are types of infrastructure in the territory of Vietnam to create a connection between the topic and the people in the country However, during the collection process, we encountered some limitations such as the location of dissolved infrastructure and the lack of exact coordinates, so we ignored it even though the list was large Information of the above 11 categories was collected from many different sources along with multiple comparisons between images and found information.
The basis for implementing the combination is: the classification information of the infrastructure After collecting information on 11 types of infrastructure, we reviewed each classification level and then grouped which insects belong to the same group and the same infrastructure classification level according to the Report. World Economic Forum Global Competitiveness Report 2019 [10] It is worth noting that careful comparison must be made to avoid placing the type of infrastructure at the wrong level Because among the 11 types, some types of data belonging to the infrastructure hierarchy in Vietnam are higher than other types but do not belong to the same subordinate group.
3.4.1.1 Process for building new datasets
They synthesized the collected data based on information regarding the classification of Vietnam's infrastructure, following these steps: e Step 1: Using information on the classification of 11 types of infrastructure, however, we take data from 3 selected types to illustrate cross-referencing, including Bridge, Highway and Hospital.
Table 3.1: Data from 3 selected types for cross-reference. e Step 2: Comparing the classifications of infrastructure types to identify the commonality among them.
= Considering two types of infrastructure, Highway and Hospital, we observe that their common feature is the shared location: Vietnam However, due to their distinct infrastructure types, they are not grouped together.
Grouping is not conducted as these do not belong to the same category of infrastructure
Figure 3.23: Considering two types of infrastructure Highway and Hospital.
= Considering two types of infrastructure, Bridge and Highway, we find that they share the common feature of Location: Vietnam Due to the nature of infrastructure classification, both types fall under the infrastructure type: Transportation; hence, they are grouped together.
Conduct grouping based on similar classification criteria
Figure 3.24: Considering two types of infrastructure Bridge and Highway.
After the data has been organized into each infrastructure group, we perform manual data labeling using the names of the infrastructure groups to create a file and store the corresponding data in those groups.
The primary label is the name of the infrastructure For example:
Table 3.2: Example to label the data.
3.5 Statistics for groups of data
The dataset was individually created by the team members, who conducted their own research, collected, and organized the data; thus, it has not been validated The dataset quality exhibits a noticeable imbalance among labels The dataset encompasses the classification of various types of Vietnam's infrastructure, including transportation, energy, water, public, production and industry The total number of image data used is 15,418.
The distribution of data in each hierarchy is as follows:
Infrastructure type | The number of labels within) Total amount of data an Infrastructure type (equivalent to the number of infrastructure entities) Transportation 3 5825
Data on labels in the data set of 11 infrastructure types:
Numerical order | Infrastructure name Amount of data
Table 3.4: Data for each infrastructure types. o Transportation:
Bridge 1485 Expressway 2063 Table 3.5: Data for each transportation types. o Energy:
Hydroelectric power plant | 809 Wind power plant 754
Table 3.6: Data for each energy types. o Water:
Water treatment plant | 1000 Water reservoir 1679
Table 3.7: Data for each water types.
Transportation | Amount of data Stadium 989
Table 3.8: Data for each public types.
> Technology park o Production and Industrial:
Table 3.9: Data for each production and industrial types.
Transportation | Amount of data Industrial zone 3284
4.1.1 Classification levels in Vietnam's infrastructure
Investment in public infrastructure has been one of the key driving forces for Vietnam’s economic development over recent decades Infrastructure accounted for 53% of total Official Development Assistance (ODA) received between 2010-2017. Vietnam has heavily invested in transportation, particularly roads, airports, and seaports Vietnam’s public and private investment in infrastructure reached 5.7% of GDP in recent years, the highest in Southeast Asia and second highest in Asia after only China (6.8% of GDP) On the one hand, boosting infrastructure development satisfies requirements for investment projects; on the other hand, it will help create higher economic growth and more employment In addition, rapid urbanization in Vietnam is a strong driver for developing the transport and utilities sector [10]
According to the World Economic Forum Global Competitiveness Report 2019 [11], Vietnam ranks 77th out of 141 in terms of overall quality of infrastructure, 66th in transport infrastructure, and 87th in utility infrastructure.
Source: World Economic Forum Global Competitiveness Report 2019
Figure 4.1: Viet Nam ranking about infrastructure in the world.!4
EXPERIMENT AND EVALUATION -ccccsecseeeirserske 58 5.1 Comparison each data and Performance of modeÌs ôô ô+ 58
Metrics of each Class ceceecceeceseeeseeneceeceseceeeeaecececeeeeceseeeaeeeeeeeeeaeeees 58 5.1.2 Evaluate of each imO(el - ‹- c s11 +3 191119119 1191 vn HH it 60 5.1.2.1 COIN rrugmmms eee ee gggmmmmes
SVM | CNN with with Pencil | Pencil CNN ResNet50 | VGG16
Table 5.1: Table of Measurement results of the models for Transportation.
SVM | CNN with with Pencil | Pencil CNN ResNet50 | VGG16
Table 5.2: Table of Measurement results of the models for Energy.
Pencil | Pencil CNN ResNet50 | VGG16
Table 5.3: Table of Measurement results of the models for Water.
SVM | CNN with with a /s4 CNN ResNet50 | VGG16
Table 5.4: Table of Measurement results of the models for Public.
SVM | CNN with with pencil | Pencil CNN ResNet50 | VGG16
Table 5.5: Table of Measurement results of the models for Production and
Evaluation Metrics by Infrastructure Type Using CNN Model
0.8 mmm Accuracy 0.8 4 mm Precision mm Recall mmm Fl-score m8 ROC AUC
Figure 5.1: Performance Comparison of a CNN Model Across Different Infrastructure
The chart is a bar graph titled "Evaluation Metrics by Infrastructure Type Using CNN Model" It compares five different metrics across five types of infrastructure: Water, Public, Transportation, Production and Industrial, and Energy The metrics compared are Accuracy, Precision, Recall, Fl-score, and ROC AUC Here's a detailed analysis:
- Water Infrastructure: The scores for all metrics are consistent at 0.667, indicating that the CNN model performed equally across all evaluation metrics for water infrastructure.
- Public Infrastructure: There is a variation in the scores with the highest being Precision (0.463) and the lowest being Fl-score (0.312) This suggests that while the model is good at identifying true positives, it is less effective at balancing precision and recall for public infrastructure.
- Transportation Infrastructure: The scores are relatively low compared to other types, with the highest metric being Precision (0.405) and the lowest being Fl-score (0.317) This implies that the model's performance on transportation infrastructure is not as reliable as other types.
- Production and Industrial Infrastructure: This category has the highest Accuracy score at 0.8, but the other metrics are significantly lower, with the ROC AUC being the lowest at 0.315 This could indicate that while the model correctly identifies a high percentage of results, it struggles with the balance between true positive and false positive rates, as well as the balance between precision and recall.
- Energy Infrastructure: The scores are more balanced but generally moderate, with the highest being Accuracy (0.583) and the lowest being ROC AUC (0.556) The model seems to perform fairly consistently across metrics for energy infrastructure.
The model appears to be most effective in "Production and Industrial" infrastructure in terms of accuracy but shows some weaknesses in other metrics For
"Public" and "Transportation" infrastructure, the model's precision is comparatively better than its ability to balance precision and recall (F1-score).
5.1.2.2 CNN combine with Pencil Sketch
Evaluation Metrics by Infrastructure Type Using CNN Model and Pencil Sketch Method
BH Fl-score mm ROC AUC
Figure 5.2: Performance of CNN Model Using Pencil Sketch for Feature Extraction.
This bar graph, titled "Evaluation Metrics by Infrastructure Type Using CNN Model and Pencil Sketch Method", displays the performance of a convolutional neural network (CNN) model across various infrastructure types, with the addition of a method referred to as "Pencil Sketch" Here's a detailed analysis of the graph:
- Water Infrastructure: Shows relatively low performance across all metrics with the highest being Accuracy at 0.308, suggesting the model is not very accurate in this domain Precision, Recall, Fl-score, and ROC AUC are all below 0.3, indicating the model has significant room for improvement in all areas for water infrastructure.
- Public Infrastructure: This sector shows a balanced but moderate performance with Accuracy and F1-score both at 0.5 The ROC
AUC is not specified but would likely be around the same range, suggesting a consistent but not high performance.
- Transportation Infrastructure: Here, the highest metric is Fl-score at 0.418, with other metrics like Precision and Recall being lower, at 0.309 and 0.333 respectively This suggests the model has a relatively better balance of precision and recall in this domain compared to Water or Public infrastructure.
- Production and Industrial Infrastructure: This sector shows varied performance, with Accuracy at 0.8 being the highest across all infrastructure types However, other metrics are considerably lower, with Precision at 0.283 and ROC AUC at 0.295, indicating a discrepancy between the number of correct predictions and the model's confidence in those predictions.
- Energy Infrastructure: The performance is relatively high and balanced across all metrics, with Accuracy leading at 0.583 The other metrics are closely grouped around the mid-0.55 range, suggesting a more reliable performance by the model in the energy sector.
Overall, the model performs best in the Production and Industrial sector in terms of Accuracy but is less consistent across the other metrics The model's performance in the Water sector is the lowest The use of the Pencil Sketch method is not directly commented on in terms of its effect on the results, but it may be a preprocessing or feature extraction technique used before applying the CNN.
Evaluation Metrics by Infrastructure Type Using RESNET-50 Model
Figure 5.3: RESNET-50 Model Performance Across Infrastructure Sectors.
The bar graph is titled "Evaluation Metrics by Infrastructure Type Using RESNET-50 Model" It presents a comparison of five evaluation metrics — Accuracy, Precision, Recall, Fl-score, and ROC AUC — across five different types of infrastructure: Water, Public, Transportation, Production and Industrial, and Energy Here are some observations from the chart:
- Water Infrastructure: Scores perfect 1.0s across all metrics, indicating that the RESNET-50 model performs exceptionally well for this category, with ideal accuracy, precision, recall, and an excellent balance between precision and recall (Fl-score), alongside a perfect ROC AUC score.
- Public Infrastructure: Scores are high but slightly lower than Water, with the highest metric being Precision at 0.875 and the lowest being Fl-score at 0.829 This suggests that while the model is quite precise, there's a slight drop in the balance between precision and recall.
- Transportation Infrastructure: Shows lower performance compared to Water and Public, with the highest metric being F1-score at 0.656 and the lowest being Recall at 0.648 The model is less effective in this sector but still provides reasonably balanced results.
CNN and Pencil SkefCH - -.- - s1 191 1 vn ng nến 84
The CNN model and Pencil sketch method have the following classification results:
Input Images out of dataset True Labels | True Classes | Predict results
Screenshot Camranh.png Airport Transportation | True
Screenshot Andong.png Bridge Transportation | True
Screenshot HCM.png Expressway | Transportation | False
Screenshot Huongdien.png Hydroelectric | Energy False power plant
Screenshot EANA.png Wind power | Energy False plant
Screenshot HaKhanhWT.png Water Water True treatment plant
Screenshot CamuS.png Stadium Public False
Screenshot UEL.png School Public True
20200501_CVTQG_NB04_ Industrial Production True
Screenshot ICTH.png Technology | Production False park and Industrial
Table 5.7: CNN Model and Pencil Sketch Method Classification Results for
The table appears to be a summary of the classification results from a convolutional neural network (CNN) model paired with a Pencil Sketch method for feature extraction It provides information on how well the CNN model performs in classifying images from a dataset into predefined infrastructure categories.
Here's a detailed interpretation of the table:
- Input Images out of dataset: Lists the filenames of the images that were input into the CNN model for classification.
- True Labels: Indicates the actual, real-world category of the infrastructure depicted in each image.
- True Classes: Specifies the category that the CNN model is expected to classify the image into.
- Predict results: Shows whether the CNN model's prediction was correct
(True) or incorrect (False) compared to the true labels.
- The model correctly classified images such as "Screenshot Camranh.png" as an Airport and "Screenshot Andong.png” as a Bridge.
- The model made incorrect predictions for "Screenshot HCM.png"
(classified as an Expressway) and "Screenshot Huongdien.png" (classified as a Hydroelectric power plant), as well as several others.
- Notably, the model seems to have issues with certain categories, misclassifying both energy-related images ("Screenshot EANA.png" and
"Screenshot Huongdien.png") and one water-related image ("20150325_CucVTQG20150220_17_NamBo_001_58878_1A_MS.jpg ).
- The model's performance is mixed across different categories, suggesting that while it can classify some infrastructure types with high accuracy, there may be challenges with others. drive/MyDrive/ThesisProject/TestData_Option2?/Demo/Screenshot Camranh.png 1/1 [mmmmmmmmmnnnmmunmnnnnmnnmnmmmmmj = 15 590m5/step
True label: Industrial zone (Production and Industrial}
Predicted label: Industrial zone {Production and Industrial}
Figure 5.17: Pencil Sketch and CNN Model predict Airport satellite image.
86 drive/MyDrive/ThesisProject/TestOata_Option2?/Demo/Screenshot Andong'.png
- @s 1ÊBms/step True label: Hydroelectric power plant (Energy)
Predicted label: Wind power plant {Energy}
Predict result: False 1/1 [ue | - @s 155ms/step
True label: Bridge (Transportation) Predicted label: Bridge (Transportation)
Figure 5.18: Pencil Sketch and CNN Model predict Bridge satellite image.
87 drive/MyDrive/ThesisPrnject/TestData_0ption2/Demn/5creenshot HCM png
Predicted label: Stadium (Public) Predict result: False
1/1 [ ] = Os 79ms/step True label: Hydroeclectric power plant (Energy)
Predicted label: Wind power plant (Energy) Predict result: False
Figure 5.19: Pencil Sketch and CNN Model predict Expressway satellite image.
88 dr1we/Myriwe/ThesisFrn†ect/Test0ata_ ủpt inn2/Ere/5rreenshat Hưang3en an
True Labelt Mater reservoir (Water) Predicted el: Water treatment plant (Water) Predict r False
True labelr Stadiua (Pubtirl Predicted tael: School (Public)
True label: Technology park [Production and Industriat) Predicted lapel: Industrial zone (Production and Industrial)
True lade Hydroelectric power pl (Energy!
Predicted label: Wind power plant (Energy!
True lapel: Bridge (Transportation) Predicted tamel: Expressway [Transportation}
Figure 5.20: Pencil Sketch and CNN Model predict Hydroelectric power plant satellite image.
89 drive/MyDrive/ThesisProject/TestData_Option2/Demo/Screenshot EANA.png 1/1 [mmmxmmmmmmnnmmmunnnrnn=nnnnmrmmzj - 5 203n5/step
True Label: School (Public) Predicted label: School (Public)
Predict result: True 17/1 (eee ===z| = @5 119ms/step
True label: Wind power plant (Energy) Predicted label: Hydroelectric power plant (Energy)
Predict result: False 1/1 [mmnuummummunnnnmmmmnnnnmnnmnmmnmmj - Os 14Bms/step
True label: Bridge (Transportation) Predicted label: Airport {Transportation}
Figure 5.21: Pencil Sketch and CNN Model predict Wind power plant satellite image.
90 drive/MyDraw ‘ThesisProject/Testiata_Option2/Deno/Screenshot HAKhanhMT pog
True labels Water treatment plant (Water) Predicted tabel: Water treatment plant (Water!
True Label: Mind power plant (Energy!
Predicted label: Wind power plant (Energy!
True labe Technology park [Production and Industriat) Predicted label: Industrial zone (Production and Industrial)
True label: Bridge [Transportation Predicted tabel: Airport (Tramsportation) Predict resu False
Figure 5.22: Pencil Sketch and CNN Model predict Water treatment plant satellite image.
91 driwe/Myriwe/ThesisFrn†ert/Tesxtata_Iptinn2/0Ero/zB150325_EucWT62815822B_17_Nara B@1_5BBS7TBH_1A_H5 jpg
True label: Mater reservoir |lwater}
Predicted labels Water treatment plant (Water!
171 [== =] - @s 5ữns/stepn True labels Hydroelectric power plant (Energy!
Predicted label: Hydroelectric power plant (Energy!
True lapelr Industrial zone (Productiom and Industrial) Predicted tabel: Industrial zone (Production and Industrial!
Predicted label: Airport (Transportation) Predict result: False
Figure 5.23: Pencil Sketch and CNN Model predict Water reservoir satellite image.
92 drive/MyDrive/ThesisProject/TestData_Option2/Demo/Screenshot CamuS.png 1/1 [=mmummmmunmummummnnnnmnnnnnmammj = 85 471ms/step
Predict result: False 1/1 [mmmnmmnmmumrmmunmnnnnnunxmmnmmmmmj - 65 307ms/step
True label: Hydroelectric power plant (Energy) Predicted label: Wind power plant (Energy}
Predict result: False 1/1 [mmmmmmnmmmnnmmmunnnnnmunnnnrnnmmj - 65 359n5/step True label: Airport (Transportation)
Predicted label: Airport (Transportation) Predict result: True
ee ee eeee rT eT
‘pai al „ 3 : tả By aX : veg sở
Figure 5.24: Pencil Sketch and CNN Model predict Stadium satellite image.
93 dr1uE/MWyriwe/ThesisPrn†ect/Test0ata_ IEtinn2/0Ere/5creenshat UEL png
True Label: Water treatment plant (Water) Predicted label: School (Water)
True Label: School (Public) Predicted tai School (Public) Predict resu True
True label: Wind power plant (Energy!
Predicted label: Wind power plant (Energy)
1/1 [= ELSEHDETHEEE-EEEE =) - Os I1Eỉ/stEn
True lade Expressway | Transpo tini Predicted tabel: Airport (Transportation) Predict result: False fi ; “its
Figure 5.25: Pencil Sketch and CNN Model predict School satellite image.
94 driwe/Mynriwe/ThesisFrn†EcE/Test0ata_ ủEt inn2/VMEDSAT1/ZB2005B1_CWTG NH4 _22584B_1Á_MS jpg 1/1 [=e=erresurraunzstmnnnrenaum=mrrmj — Os 343a5/step
True Labelt Mater treatment plant (Mater) Predicted label: Water treatment plant (Water) Predict result: True
True label: Stadius (Public) Predicted label: School (Public) Predict result: False
1/1 [====z====zzx=zz=z=zz===========] - Os 1355/SLPP
True label: Industrial zứme (Production and Industriat) Predicted tabel: Industrial zone (Production and 1ndưstrial]
1/1 [===sr===rn=nrsrnnmnnnr=rnmrrmmmrl — lềS 130ns/step
True labELt Hydroelectric power plant (Energy) Predicted label: Wind power plant (Energy!
Predict result: False WARNING: tersorflow:S out of the last 5 calls to