=Paper=
{{Paper
|id=Vol-3925/paper09
|storemode=property
|title=Analysis of cross-sectional neural network recognition of satellite and aerial survey data
|pdfUrl=https://ceur-ws.org/Vol-3925/paper09.pdf
|volume=Vol-3925
|authors=Valerii Zivakin,Oleksandr Kozachuck,Pylyp Prystavka
|dblpUrl=https://dblp.org/rec/conf/cmigin/ZivakinKP24
}}
==Analysis of cross-sectional neural network recognition of satellite and aerial survey data==
https://ceur-ws.org/Vol-3925/paper09.pdf
Analysis of cross-sectional neural network recognition of
satellite and aerial survey data
Valerii Zivakin1,†, Oleksandr Kozachuck1,† and Pylyp Prystavka1,∗,†
1
National Aviation University, Liubomyra Huzara Ave. 1, Kyiv, 03058, Ukraine
Abstract
Modern advances in the field of neural network data recognition open new perspectives for the use of
satellite and aerial photography in aerospace research. The use of unmanned aerial vehicles (UAVs) is
becoming increasingly important in various fields, including military, environmental, and agro-industrial
applications. However, the issue of automating the processing and analysis of data obtained from satellite
and aerial surveys remains relevant in terms of the implementation of these technologies. In this study, we
focus on the analysis of the effectiveness of cross-sectional neural network recognition, which was
performed on both satellite and aerial (UAV). One of the key problems is the need to implement automated
methods of analysis and classification of images obtained from different sources. In the context of using
satellite data in aerospace applications, it is important to understand how effectively neural networks can
adapt to changes in information sources.
Keywords
neural network, data, recognition, image, dataset, satellite, aerial survey1
1. Preparation
It is important to note that both areas of data acquisition are extremely fruitful in their own right.
For example, every single set presented in this paper has already been used for research. So, with the
help of satellite data sets, various types of convolutions were studied by neural network classifiers
and their applications [1–4]. The " Aerial Survey " dataset and its separate parts have been the basis
for scientific works of various levels for several years [5–10]. A comparative element is also quite
common, for example, a large part of the work [6] contains a description of the advantages of one
type of data over another.
Recent scientific studies, such as work [7], have already emphasized the importance of cross-
training to achieve high accuracy when working with heterogeneous data. Our approach
complements these studies by focusing on the specific challenges and opportunities associated with
adapting neural networks to satellite and aerial imagery processing.
The purpose of this research is to determine the possibilities of using satellite and aerial
photography in interaction with unmanned aerial vehicles. Investigating how neural networks of
the same architecture and other hyperparameters, trained on different types of data, can adapt and
effectively recognize objects and phenomena on the ground, we set ourselves the task of improving
systems for automatic analysis of information received in real-time. For neural network recognition,
was selected a part of the " Aerial Survey " dataset [5], which was formed from data obtained from
UAVs, as well as four datasets consisting of satellite images [1–4]. The structure of each set and
conventions for convenient abbreviation are presented in Tables 1 – 5.
CH&CMiGIN’24: Third International Conference on Cyber Hygiene & Conflict Management in Global Information Networks,
January 24–27, 2024, Kyiv, Ukraine
∗
Corresponding author.
†
These authors contributed equally.
zivakin1993@gmail.com (V. Zivakin); oleksandrkozachukk@gmail.com (O. Kozachuck); chindakor37@gmail.com (P.
Prystavka)
0000-0002-0360-2459 (V. Zivakin); 0000-0002-0420-0558 (O. Kozachuck); 0000-0002-0360-2459 (P. Prystavka)
© 2025 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
It should be noted that the primary goal of the study was to check the possibility of expanding
the dataset of aerial photography. That is why the present sets were studied in this ratio (4 satellite
to 1 aerial survey).
Table 1
Structure of the "Aerial survey" Set
The number
Class name Description of digital Mark
images
Factory Industrial building 360 B.1.1
Forest Forests, spring-summer period (without snow) 2132 B.1.2
Vehicles Road transport 5921 B.1.3
Non vegetation field Soil fields (without vegetation) 2175 B.1.4
Not factory Not an industrial building 1979 B.1.5
Pillars Pillars (separate, roadside) 1188 B.1.6
Residential A cluster of residential buildings 386 B.1.7
Roads The roads 4236 B.1.8
Vegetation_field Fields with vegetation 1875 B.1.9
Water Reservoirs with coastline 879 B.1.10
Table 2
Structure of the "EuroSAT" Set
The number
Class name Description of digital Mark
images
Annual Crop Agricultural fields with annual crops 3000 P.1.1
Forest Forest landscape, without snow 3000 P.1.2
Herbaceous
Fields with grass vegetation 3000 P.1.3
Vegetation
Highway Roads, tracks 2500 P.1.4
Industrial Industrial building 2500 P.1.5
Pasture Pastures 2000 P.1.6
Permanent Crop Agricultural fields with perennial crops 2500 P.1.7
Residential A cluster of residential buildings 3000 P.1.8
River Rivers 2500 P.1.9
Sea Lake Reservoirs, sea canvas 3000 P.1.10
Table 3
Structure of the "AID" Set
The number
Class name Description of digital Mark
images
Areas designated for aviation services, including
Airport 360 C.2.1
runways and airport infrastructure
Areas with no significant vegetation or cover, such
BareLand 310 C.2.2
as bare ground or areas without vegetation
BaseballField Baseball pitches with a distinctive configuration 220 C.2.3
Places where the shore and the water meet, with a
Beach 400 C.2.4
sandy or rocky surface
Structures for crossing water obstacles or other
Bridge 360 C.2.5
areas, usually with a building structure
Central parts of cities or settlements with intensive
Center 260 C.2.6
construction and infrastructure
Church Religious buildings, such as churches or temples 240 C.2.7
Areas designated for commercial activities, such as
Commercial 350 C.2.8
business centers, shops and offices
Industrial Territories with industrial infrastructure 390 C.2.9
Meadow Open spaces with natural grass vegetation 280 C.2.10
Medium Areas with a moderate density of residential
290 C.2.11
Residential development and residential buildings
Mountain Mountain regions with a characteristic landscape 340 C.2.12
Park Territories intended for recreation 350 C.2.13
Parking Areas for parking vehicles 390 C.2.14
Playground Areas with children's playgrounds 370 C.2.15
Pond The reservoir is smaller in size and shallow 420 C.2.16
Port Areas of port infrastructures for sea transport 380 C.2.17
Railway Station Territories of railway hubs and infrastructure 260 C.2.18
Resort Recreation areas 290 C.2.19
River Large water streams 410 C.2.20
School Territories of educational institutions 300 C.2.21
Sparse Residential Areas with a low density of residential buildings 300 C.2.22
Square Large open areas or squares in cities 330 C.2.23
Stadium Areas with sports grounds for games and events 290 C.2.24
StorageTanks Areas with tanks for storing various substances 360 C.2.25
Viaduct Bridge-like structures that cross various obstacles 420 C.2.26
DenseResidential Areas with a high density of residential buildings 410 C.2.27
Desert Dry areas with a characteristic landscape 300 C.2.28
Farmland Territories for agricultural production with crops 370 C.2.29
Forest Areas with dense forest vegetation 250 C.2.30
Table 4
Structure of the Set "NWPU-RESISC45"
The number
Class name Description of digital Mark
images
Areas with images of aircraft in various operating
airplane 700 C.3.1
conditions
Areas of airports with runways and
airport 700 C.3.2
infrastructure for aviation services
baseball_diamond Images of baseball fields and their infrastructure 700 C.3.3
Photos of basketball courts and their
basketball_court 700 C.3.4
surroundings
Coasts and coastal areas with sandy or rocky
beach 700 C.3.5
surfaces
Structures for crossing water obstacles or other
bridge 700 C.3.6
areas, usually with a building structure
chaparral Image of areas with characteristic vegetation 700 C.3.7
church Religious buildings, such as churches or temples 700 C.3.8
circular_farmland Areas with circular or elliptical agricultural fields 700 C.3.9
cloud Areas with images of clouds or cloud formations 700 C.3.10
Areas designated for commercial activities, such
commercial_area 700 C.3.11
as business centers, shops and offices
dense_residential Areas with a high density of residential buildings 700 C.3.12
Dry, sparsely populated areas with a
desert 700 C.3.13
characteristic landscape
forest Areas with dense forest vegetation 700 C.3.14
Territories with a highway and road
freeway 700 C.3.15
infrastructure facilities
golf_course Images of golf courses and their infrastructure. 700 C.3.16
Photos of sports grounds for running and other
ground_track_field 700 C.3.17
sports
Areas with port infrastructure for receiving and
harbor 700 C.3.18
servicing ships
Territories with industrial infrastructure, factories
industrial_area 700 C.3.19
and other industrial facilities
intersection Road intersection areas with road infrastructure 700 C.3.20
Areas with the image of islands surrounded by
island 700 C.3.21
water bodies
lake Areas with large bodies of water, such as lakes 700 C.3.22
meadow Open spaces with natural grass vegetation 700 C.3.23
Areas with a moderate density of residential
medium_residential 700 C.3.24
development and residential buildings
mobile_home_park Territories with stationary or mobile homes 700 C.3.25
High mountain regions with a characteristic
mountain 700 C.3.26
landscape
Structures that cross and pass over other roads or
overpass 700 C.3.27
objects
Areas with images of palaces or historical
palace 700 C.3.28
buildings
parking_lot Photos of areas for parking vehicles 700 C.3.29
railway Territories with railway infrastructure and roads 700 C.3.30
railway_station Areas of railway hubs and infrastructure 700 C.3.31
Areas with rectangular or square agricultural
rectangular_farmland 700 C.3.32
fields
river Large water streams 700 C.3.33
roundabout Areas with roundabouts 700 C.3.34
runway Photos of airport runways 700 C.3.35
sea_ice Areas with ice cover on seas and oceans 700 C.3.36
ship Areas with the image of large water vessels 700 C.3.37
Territories with the image of large mountain
snowberg 700 C.3.38
snow landscapes
sparse_residential Areas with a low density of residential buildings 700 C.3.39
Territories with large sports grounds for holding
stadium 700 C.3.40
games and events
tennis_court Photos of tennis courts and their surroundings 700 C.3.41
terrace Territories with a terraced relief structure 700 C.3.42
Territories with thermal power plants and energy
thermal_power_station 700 C.3.43
infrastructure
wetland Areas depicting wetlands and swamps 700 C.3.44
storage_tank Areas with tanks for storing various substances 700 C.3.45
A neural network was used to conduct the research coagulation classifier (a classifier based on a
convolutional artificial neural network, created with [11]), the structure of which is illustrated in
Figure 1. We also note that all images undergo reprocessing, during which their size is also unified,
which at the beginning of processing is 64 by 64 pixels.
During the experiment, each of the satellite datasets was pairwise compared with the UAV set.
The essence of the comparison was that in each pair of singulars, the data set first served as a training
one, and the second as a test one. Thus, we were able to observe how the data from the test set will
be reflected on the data from the set for training the network (training).
After receiving the result, the datasets in pairs changed roles and the learning and testing process
began anew. Thus, we were able to evaluate which classes from the satellite sets are better displayed
(recognized) in the classes from the UAV set and vice versa.
For convenience, these data are presented in the form of matrices with a color scale (Table 6), in
which you can see which frequency of a particular class was reflected on the classes from the training
set.
Table 5
Structure of the Set "UCMerced _ LandUse"
The number
Class name Description of digital Mark
images
Areas used for agricultural activities, such as
agricultural 100 C.4.1
fields where different crops are grown
airplane Images of aerial vehicles 100 C.4.2
Image of baseball fields with a characteristic
baseballdiamond 100 C.4.3
configuration
Places where the shore meets the water, with a
beach 100 C.4.4
sandy or rocky surface
Areas with a concentration of buildings,
buildings including residential, commercial and other 100 C.4.5
structures
chaparral Regions with dense shrub vegetation 100 C.4.6
Areas with a high density of residential
dense residential 100 C.4.7
buildings
forest Territories with dense forest vegetation 100 C.4.8
Motorways and expressways for automobile
freeway 100 C.4.9
traffic
golfcourse Golf courses 100 C.4.10
Areas of port infrastructure for parking and
harbor 100 C.4.11
maintenance of ships
intersection Images of road intersections and crossed roads 100 C.4.12
Areas with a moderate density of residential
mediumresidential 100 C.4.13
buildings
Areas with parks for mobile homes and
mobilehomepark 100 C.4.14
residential facilities
Above-ground structures for crossing other
overpass 100 C.4.15
roads or obstacles
parkinglot Territories for parking vehicles 100 C.4.16
river Large water streams 100 C.4.17
runway Runways for aircraft 100 C.4.18
sparseresidential Areas with a low density of residential buildings 100 C.4.19
storagetanks Areas with tanks for storing various substances 100 C.4.20
tenniscourt Tennis courts 100 C.4.21
Figure 1: Architecture of the used classifier.
Table 6
Display Level Scale
Scale
1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
2. Experiment results
Tables 7–14 present the results of a pairwise mapping study. Since some satellite datasets contain a
significant number of classes, all tables are reduced to a view where satellite classes are arranged
horizontally (rows) and UAV classes vertically (columns).
In this case, you can understand which classes were included in the educational sample (the one
in which it is displayed) from the name of the table. For example, in Table 8 it is B.1, vertical, and in
Table 7 it is C.1, horizontal.
Table 7
The Result of Mapping Aerial Photographs from (B.1) into Classes from (C.1)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
P.1.1 0 0.33 0.07 0.19 0 0.10 0 0.08 0.05 0.02
P.1.2 0 0.01 0 0.03 0 0 0 0 0.30 0.20
P.1.3 0.26 0.55 0.40 0.42 0.17 0.38 0 0.38 0.13 0.23
P.1.4 0 0 0.02 0 0.03 0.03 0 0.20 0 0.01
P.1.5 0.57 0 0.13 0 0.71 0.01 0.01 0.07 0.05 0.03
P.1.6 0 0 0 0.02 0 0.02 0 0 0 0
P.1.7 0 0.01 0 0.24 0.03 0 0 0.02 0 0.02
P.1.8 0.16 0.08 0 0.01 0.05 0.02 0.97 0 0 0.04
P.1.9 0 0 0.14 0 0.01 0.05 0.01 0.14 0 0.19
P.1.10 0.01 0 0.23 0.09 0 0.39 0.01 0.11 0.47 0.43
Table 8
The Result of Mapping Satellite Images from (C.1) into Classes from (B.1)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
P.1.1 0 0.01 0.02 0.05 0.17 0.04 0 0.13 0.01 0.57
P.1 .2 0 0.06 0 0.01 0 0.11 0 0 0.03 0.79
P.1.3 0.09 0.07 0.01 0.45 0.05 0.17 0.01 0.04 0.01 0.11
P.1.4 0.02 0.01 0.04 0.02 0.09 0.38 0.03 0.27 0 0.14
P.1.5 0.05 0 0.43 0 0.39 0.01 0.10 0.02 0 0
P.1.6 0 0.06 0.01 0.02 0 0.35 0 0 0.02 0.53
P.1.7 0.05 0.07 0.02 0.25 0.15 0.16 0.02 0.07 0.01 0.20
P.1.8 0.09 0.05 0.02 0.11 0.08 0.01 0.63 0 0 0.01
P.1.9 0 0.01 0.03 0 0.01 0.16 0.01 0.03 0.01 0.73
P.1.10 0 0 0 0 0 0.08 0 0 0.02 0.90
Table 9
The Result of Mapping Aerial Photographs from (B.1) into Classes from (S.2)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.2.1 0.24 0 0.05 0 0.19 0.02 0 0.14 0 0.02
C.2.2 0.03 0.01 0.05 0.35 0 0.12 0 0.11 0.01 0.21
C.2.3 0 0 0.01 0 0 0 0 0 0 0
C.2.4 0.04 0 0.07 0.01 0.02 0.08 0.52 0.15 0.11 0.16
C.2.5 0.10 0 0.02 0.01 0.02 0.11 0 0.19 0 0.21
C.2.6 0.20 0 0.07 0 0.13 0.01 0 0 0 0.01
C.2.7 0.03 0 0 0 0.06 0 0 0 0 0
C.2.8 0 0 0 0 0.01 0 0 0 0 0
C.2.9 0.03 0 0 0 0.07 0 0 0 0 0.07
C.2.10 0 0 0 0.03 0 0.04 0 0.01 0.77 0
C.2.11 0 0 0.02 0 0.01 0.02 0.01 0 0 0.01
C.2.12 0 0.17 0 0 0.03 0 0 0.02 0 0.01
C.2.13 0 0.02 0 0 0 0 0 0 0 0
C.2.14 0.04 0 0 0 0.04 0 0 0 0 0
C.2.15 0 0 0.02 0 0.04 0.01 0 0.02 0.01 0
C.2.16 0 0 0.02 0 0.01 0 0 0 0 0.06
C.2.17 0.02 0 0 0 0.01 0 0.01 0 0 0.04
C.2.18 0.01 0 0 0.01 0.01 0 0 0.03 0 0.01
C.2.19 0 0.01 0.01 0 0.02 0 0.01 0.01 0 0.03
C.2.20 0.01 0.01 0.05 0 0.01 0.01 0 0.07 0 0.04
C.2.21 0.01 0.04 0 0 0.11 0 0.27 0 0 0
C.2.22 0 0 0.07 0 0 0 0 0 0 0
C.2.23 0.08 0.17 0.14 0 0.14 0.07 0 0.07 0 0.02
C.2.24 0.01 0 0.01 0 0.02 0 0 0 0 0
C.2.25 0.01 0 0.20 0 0.01 0 0 0 0 0
C.2.26 0.03 0 0 0 0.02 0 0 0.02 0 0
C.2.27 0 0.20 0 0.13 0.01 0 0.02 0 0 0
C.2.28 0 0 0.15 0.17 0 0.37 0 0.04 0.01 0
C.2.29 0.10 0.38 0.02 0.28 0.01 0.12 0 0.09 0.07 0.05
C.2.30 0 0.15 0 0.01 0 0 0.15 0 0.01 0.04
Table 10
The Result of Mapping Satellite Images from (S.2) into Classes from (B.1)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.2.1 0.16 0.01 0.04 0 0.59 0.01 0.01 0.12 0 0.06
C.2.2 0.02 0.14 0.09 0.32 0.10 0.08 0 0.14 0.04 0.07
C.2.3 0.09 0.02 0.05 0 0.35 0.03 0 0.22 0 0.25
C.2.4 0.10 0.01 0.12 0 0.16 0.02 0.03 0.14 0.09 0.33
C.2.5 0.25 0.01 0.02 0 0.12 0.15 0.01 0.16 0 0.28
C.2.6 0.19 0 0.03 0 0.78 0 0 0 0 0
C.2.7 0.30 0.04 0.01 0 0.63 0 0 0 0 0.02
C.2.8 0.07 0.08 0 0 0.73 0 0.11 0 0 0.01
C.2.9 0.05 0.01 0.05 0 0.83 0 0.03 0.01 0.01 0.01
C.2.10 0 0.06 0.26 0.28 0 0.01 0 0.04 0.34 0
C.2.11 0.18 0.27 0.02 0.02 0.30 0 0.18 0.02 0 0.02
C.2.12 0.04 0.27 0.25 0.03 0.17 0.02 0.01 0.08 0.01 0.12
C.2.13 0.18 0.16 0.07 0.01 0.20 0.03 0.22 0.03 0.01 0.10
C.2.14 0.06 0.02 0.02 0 0.85 0.01 0.03 0 0.01 0.01
C.2.15 0.15 0.02 0.02 0 0.36 0.01 0.01 0.18 0.01 0.24
C.2.16 0.08 0.01 0 0 0.08 0.02 0 0.02 0 0.79
C.2.17 0.26 0 0.03 0 0.33 0.05 0.14 0 0 0.19
C.2.18 0.10 0.05 0.07 0.01 0.37 0.03 0.22 0.11 0 0.03
C.2.19 0.07 0.04 0.05 0.01 0.70 0.01 0.06 0.03 0 0.03
C.2.20 0.03 0.18 0.09 0.01 0.06 0.11 0.02 0.18 0 0.30
C.2.21 0.22 0.05 0.04 0 0.59 0.01 0.05 0.02 0 0.01
C.2.22 0.09 0.35 0.31 0 0.03 0.09 0.01 0 0 0.10
C.2.23 0.12 0.07 0.03 0 0.70 0.01 0.01 0.03 0 0.04
C.2.24 0.24 0 0.01 0 0.72 0 0 0.01 0 0.01
C.2.25 0.03 0.01 0.25 0.01 0.69 0 0 0.01 0 0.01
C.2.26 0.23 0.08 0.03 0 0.31 0.03 0.03 0.22 0 0.07
C.2.27 0.03 0.22 0.10 0.06 0.44 0 0.10 0 0.05 0
C.2.28 0 0.03 0.38 0.47 0.02 0.01 0 0.06 0.02 0.01
C.2.29 0.02 0.26 0.05 0.08 0.07 0.20 0.04 0.10 0.03 0.16
C.2.30 0 0.68 0.21 0 0 0 0.04 0.01 0.02 0.03
Table 11
The result of Mapping Aerial Photographs from (B.1) into Classes from (S.3)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.3.1 0.19 0 0.05 0.08 0.05 0.15 0 0.07 0 0.02
C.3.2 0.03 0 0 0 0.02 0.01 0 0.01 0 0
C.3.3 0 0 0 0 0 0 0 0 0 0
C.3.4 0.07 0 0 0.04 0.09 0 0 0.01 0 0
C.3.5 0 0 0.01 0 0.01 0.04 0 0.06 0.02 0.16
C.3.6 0.03 0 0 0.01 0 0.09 0 0.04 0.08 0.05
C.3.7 0 0 0 0 0 0 0 0 0 0
C.3.8 0.02 0 0.03 0 0.10 0 0 0 0 0
C.3.9 0 0.01 0 0 0 0 0 0 0 0
C.3.10 0.03 0.22 0.49 0 0.11 0.14 0 0.08 0 0.02
C.3.11 0.01 0.04 0 0 0.02 0 0 0 0 0
C.3.12 0 0.01 0 0 0.03 0 0.01 0 0 0
C.3.13 0 0.01 0.02 0.20 0 0.03 0 0.03 0.03 0.01
C.3.14 0 0.09 0 0 0 0 0 0 0 0
C.3.15 0.01 0 0 0.01 0.01 0 0.01 0.06 0 0.01
C.3.16 0 0 0 0 0 0 0 0 0 0
C.3.17 0 0 0 0 0 0 0 0 0 0
C.3.18 0 0 0 0 0 0 0.13 0 0 0
C.3.19 0.02 0.01 0 0 0.08 0 0.01 0 0 0
C.3.20 0 0 0 0 0.01 0 0 0 0 0
C.3.21 0 0 0.09 0.02 0 0.15 0 0.03 0.20 0.09
C.3.22 0 0 0 0 0.01 0 0 0 0 0.04
C.3.23 0 0 0 0 0 0 0 0 0.45 0
C.3.24 0 0.02 0 0 0.01 0 0 0 0 0
C.3.25 0 0 0 0 0 0 0.03 0 0 0
C.3.26 0 0.13 0 0.01 0 0.01 0 0.01 0 0
C.3.27 0.01 0 0 0 0.02 0 0 0.01 0 0
C.3.28 0.15 0 0.01 0 0.19 0 0 0.01 0 0.01
C.3.29 0.12 0 0 0 0.06 0 0.19 0.01 0.09 0.01
C.3.30 0.01 0 0 0.01 0 0 0.01 0.01 0 0.02
C.3.31 0.03 0 0 0 0.03 0 0.09 0.02 0 0.01
C.3.32 0 0 0 0.01 0 0 0 0 0 0
C.3.33 0 0 0 0 0 0 0 0 0 0.03
C.3.34 0 0 0 0 0 0 0 0 0 0
C.3.35 0.03 0.05 0.08 0 0.04 0.22 0.01 0.40 0.01 0.07
C.3.36 0.02 0.07 0.02 0.10 0 0.02 0.01 0 0.07 0.09
C.3.37 0.19 0 0.04 0.49 0.01 0.08 0.04 0.06 0.01 0.32
C.3.38 0.03 0.12 0.01 0.01 0.05 0.03 0 0.01 0 0
C.3.39 0 0 0 0 0.01 0 0 0.01 0 0
C.3.40 0 0.07 0.01 0 0.01 0 0.29 0 0 0
C.3.41 0.01 0 0 0 0 0 0 0 0 0
C.3.42 0 0.01 0 0 0 0.01 0 0 0 0.02
C.3.43 0.02 0 0.13 0 0.03 0.01 0.14 0.02 0 0
C.3.44 0 0.05 0 0 0 0 0 0 0 0.01
C.3.45 0 0.06 0.01 0 0.01 0 0.01 0 0 0
Table 12
Display Result Satellite Pictures from (C.3) into Classes from (B.1)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.3.1 0.16 0 0.28 0 0.38 0.04 0 0.09 0.01 0.04
C.3.2 0.06 0.02 0.04 0.01 0.31 0.09 0.01 0.29 0 0.15
C.3.3 0.06 0.13 0.03 0 0.16 0.06 0 0.12 0.01 0.44
C.3.4 0.16 0.08 0.05 0 0.30 0.04 0 0.08 0 0.29
C.3.5 0.10 0 0.09 0 0.20 0.04 0.02 0.28 0.02 0.25
C.3.6 0.14 0 0.01 0 0.05 0.22 0 0.09 0 0.49
C.3.7 0 0.25 0.06 0.20 0.41 0.04 0 0.02 0.01 0.01
C.3.8 0.26 0.02 0.04 0 0.64 0.01 0 0 0 0.04
C.3.9 0.03 0.19 0.02 0 0.16 0.12 0 0.05 0.01 0.42
C.3.10 0.04 0.01 0.37 0 0.44 0.04 0 0.06 0.01 0.04
C.3.11 0.23 0.06 0.02 0 0.60 0.04 0.01 0 0.01 0.02
C.3.12 0.13 0.24 0.03 0 0.53 0.01 0.02 0 0 0.02
C.3.13 0 0.05 0.29 0.41 0.08 0.02 0 0.12 0 0.02
C.3.14 0 0.83 0.10 0.01 0 0.02 0 0.01 0 0.02
C.3.15 0.04 0.04 0.06 0 0.11 0.02 0 0.44 0 0.29
C.3.16 0.06 0.43 0.06 0 0.11 0.07 0.01 0.06 0 0.20
C.3.17 0.21 0.10 0.04 0 0.30 0.03 0.01 0.09 0 0.23
C.3.18 0.40 0.02 0.10 0 0.29 0.02 0.14 0 0 0.02
C.3.19 0.06 0.01 0.11 0 0.79 0 0.02 0.01 0 0.01
C.3.20 0.15 0.04 0.02 0 0.57 0.01 0 0.09 0 0.12
C.3.21 0.02 0.02 0.30 0 0.02 0.37 0 0.02 0.01 0.25
C.3.22 0.01 0.22 0.09 0 0.19 0.06 0 0.02 0 0.41
C.3.23 0 0.13 0.33 0.13 0 0.09 0 0.03 0.25 0.05
C.3.24 0.17 0.20 0.02 0 0.49 0.02 0.01 0.01 0 0.09
C.3.25 0.16 0.04 0.07 0 0.65 0 0.01 0.02 0 0.04
C.3.26 0 0.42 0.17 0.09 0.03 0.08 0 0.09 0 0.12
C.3.27 0.13 0.02 0.01 0 0.23 0.01 0 0.43 0 0.17
C.3.28 0.19 0.06 0.05 0 0.62 0.02 0.01 0.01 0 0.05
C.3.29 0.17 0.07 0.07 0.01 0.58 0.01 0.02 0.03 0 0.04
C.3.30 0.07 0.07 0.06 0.01 0.11 0.01 0.01 0.43 0 0.23
C.3.31 0.08 0.04 0.07 0.01 0.42 0.02 0.04 0.22 0 0.09
C.3.32 0.04 0.18 0.05 0.04 0.08 0.15 0.02 0.15 0.02 0.27
C.3.33 0.01 0.23 0.03 0.01 0.04 0.13 0.02 0.11 0.01 0.41
C.3.34 0.17 0.04 0.08 0 0.46 0.01 0.01 0.13 0 0.11
C.3.35 0.04 0 0.17 0.01 0.15 0.03 0 0.44 0.01 0.14
C.3.36 0.13 0.10 0.16 0 0.44 0.13 0.01 0 0 0.02
C.3.37 0.06 0.02 0.22 0 0.19 0.07 0.01 0.03 0 0.40
C.3.38 0.15 0.01 0.02 0 0.80 0.01 0 0 0 0.01
C.3.39 0.07 0.13 0.30 0 0.18 0.17 0 0.07 0 0.09
C.3.40 0.03 0.01 0.03 0 0.58 0.03 0.01 0.01 0 0.02
C.3.41 0.25 0.14 0.01 0 0.34 0.01 0 0.05 0 0.19
C.3.42 0.03 0.27 0.03 0.13 0.03 0.09 0.02 0.16 0.01 0.23
C.3.43 0.14 0.05 0.08 0.01 0.60 0.01 0.08 0.02 0 0.02
C.3.44 0 0.65 0.07 0.02 0.01 0.07 0 0.07 0 0.10
C.3.45 0.04 0.01 0.25 0 0.67 0 0.01 0.01 0 0.01
Table 13
The Result of Mapping Aerial Photographs from (B.1) into Classes from (S.4)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.4.1 0.28 0.13 0.04 0.95 0 0.52 0.38 0.21 0.76 0.33
C.4.2 0 0 0.01 0 0.01 0 0 0 0 0
C.4.3 0 0.06 0.05 0 0.03 0.02 0 0.05 0 0.02
C.4.4 0.06 0.01 0.49 0 0.02 0.33 0 0.24 0.04 0.37
C.4.5 0.12 0 0.02 0 0.08 0 0 0 0 0
C.4.6 0 0.01 0 0.01 0 0 0 0.01 0 0
C.4.7 0.01 0.10 0 0 0.03 0 0 0 0 0
C.4.8 0 0.31 0 0.04 0.01 0.01 0.47 0.02 0 0.01
C.4.9 0.22 0 0.01 0 0.21 0 0 0.12 0 0.01
C.4.10 0 0.21 0.05 0 0.01 0.05 0 0.12 0.19 0.03
C.4.11 0 0.01 0 0 0.01 0 0 0 0 0
C.4.12 0.01 0.08 0 0 0.02 0 0.01 0.02 0 0
C.4.13 0.02 0 0 0 0.18 0 0 0.03 0 0
C.4.14 0.01 0 0 0 0.02 0 0 0 0 0
C.4.15 0.08 0 0 0 0.02 0 0 0 0 0
C.4.16 0 0 0 0 0.04 0 0 0 0 0
C.4.17 0.01 0.01 0.01 0 0.02 0.02 0 0.03 0 0.19
C.4.18 0.03 0 0.01 0 0.01 0 0 0.07 0 0
C.4.19 0.01 0.06 0.01 0 0.04 0.01 0 0.01 0.01 0
C.4.20 0.12 0 0.24 0 0.14 0.02 0.13 0.02 0 0
C.4.21 0.03 0.02 0.07 0 0.10 0.01 0 0.03 0 0.03
Table 14
The Result of Mapping Satellite Images from (S.4) into Classes from (B.1)
B.1.1 B.1.2 B.1.3 B.1.4 B.1.5 B.1.6 B.1.7 B.1.8 B.1.9 B.1.10
C.4.1 0 0.01 0.03 0.52 0 0 0 0 0.44 0
C.4.2 0.10 0 0.38 0 0.48 0.03 0 0 0 0.01
C.4.3 0.10 0.01 0.08 0 0.38 0.06 0 0.17 0.05 0.15
C.4.4 0.04 0 0.08 0 0.12 0 0 0.30 0.14 0.32
C.4.5 0.19 0 0 0 0.81 0 0 0 0 0
C.4.6 0 0.10 0.07 0.74 0.02 0 0 0 0.07 0
C.4.7 0.01 0 0 0 0 0.99 0 0 0 0
C.4.8 0 0.99 0 0.01 0 0 0 0 0 0
C.4.9 0.18 0 0.02 0.01 0.28 0 0 0.38 0 0.13
C.4.10 0.03 0.06 0.19 0 0.28 0.03 0 0.18 0.16 0.07
C.4.11 0.07 0 0.25 0 0.45 0 0.12 0 0 0.11
C.4.12 0.04 0 0 0 0.78 0.01 0 0.13 0 0.04
C.4.13 0 0 0 0 1.00 0 0 0 0 0
C.4.14 0 0 0.06 0 0.94 0 0 0 0 0
C.4.15 0.18 0 0.01 0 0.54 0 0 0.19 0 0.08
C.4.16 0.02 0 0.02 0 0.95 0 0.01 0 0 0
C.4.17 0 0 0 0 0 0 0 0 0 1.00
C.4.18 0.03 0 0.08 0.03 0.21 0 0 0.36 0 0.29
C.4.19 0 0.03 0.10 0 0.72 0.02 0 0.08 0 0.05
C.4.20 0.04 0 0.26 0 0.67 0 0 0.02 0 0.01
C.4.21 0.27 0.03 0.03 0 0.61 0 0 0 0 0.06
From the given data can be drawn the following conclusions. First of all, the classes reflecting
natural formations (forests, fields of various kinds, reservoirs and rivers) have great compatibility
from the point of view of displaying satellite images in UAVs - analogues. For example, classes C.4.8
and C.3.14 appeared in B.1.2 almost in full and all of them are annotated as forests. Classes C.4.17
(rivers) and C.2.16 (ponds) are also strongly reflected in class B.1.10 (bodies in general). Indeed, it
might seem obvious, but it is worth noting that when reflected in reverse, such ambiguity disappears.
For example, the same class of reservoirs B.1.10 when displayed in satellite classes relatively is
equally distributed between C.4.1 and C.4.4 or with an ambiguous advantage (0.32) is reflected in the
class of ships C.3.17. This shows that, from the point of view of natural formations, it is the data
obtained with the help of satellite imaging that can serve to expand the UAV dataset.
Table 15
Examples of digital images of classes "Forest"
Class label An example of a central office
B.1.2
P.3.14
P.4.8
The most effective class from the point of view of mapping satellite images into UAV classes was
class B.1.5 – non-industrial buildings. Classes C.4.5, C.4.13, C.4.14, C.2.9, by more than 80 percent,
were reflected precisely in class B.1.5, while they themselves represent different types of buildings,
which indicates that they you can supplement this class. On the other hand, the dataset as a whole
can be supplemented with classes that have a not so high level of reflection in B.1.5 (from 0.3 to 0.75),
but at the same time it is the largest. Examples of such are P.4.21, P.4.20, P.4.15, P.4.12, P.4.11, P.3.43,
P.3.43, P.3.36, P.3.34. and other. Also, classes C.4.16, C.2.14 (parking lots) showed a rather high, but
at the same time, false level of display - such indicators can also be considered a marker for selecting
a new dataset class or expanding an already existing one (for example, motor vehicle class B.1.3).
Table 16
Examples of Images of Classes Related to Water Bodies
Class label An example of a central office
B.1.10
P.2.16
P.4.17
Table 17
Examples of images of building classes and strongly reflected in it
Class label An example of a central office
B.1.5
P.2.9
P.2.14
P.4.5
P.4.13
P.4.14
P.4.16
In addition, the division of the UAV construction class – the dataset into classes B.1.1, B.1.5, B.1.7
– showed its impracticality, because in B.1.1 and B.1.7 a significant share of satellite data was not
displayed, and these classes themselves were displayed in meaningless, such as C.4.1, C.4.8 (fields
and forests), C.3.1 (aircraft), C.2.4 (beaches).
3. Conclusions
In general, the study confirmed the feasibility of creating mixed-type datasets and the possibility of
supplementing the UAV dataset with images of satellite data, or creating new classes based on them.
In the perspective of future research, it is possible to highlight the creation of a universal
"framework" set that could be easily modified to meet the needs of various tasks.
Declaration on Generative AI
The author(s) have not employed any Generative AI tools.
References
[1] P. Helber, B. Bischke, A. Dengel, D. Borth, EuroSAT: A Novel dataset and deep learning
benchmark for land use and land cover classification, IEEE Journal of Selected Topics in Applied
Earth Observations and Remote Sensing (2019). doi: 10.1109/JSTARS.2019.2918242.
[2] G.-S. Xia, J. Hu, F. Hu, B. Shi, X. Bai, Y. Zhong, L. Zhang, AID: A benchmark dataset for
performance evaluation of aerial scene classification, IEEE Trans. Geoscience and Remote
Sensing 55(7) (2017) 395–398.
[3] A. Raza, H. Huo, S. Sirajuddin, T. Fang, Diverse capsules network combining multi-
convolutional layers for remote sensing image scene classification, IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing 99 (2020). doi:
10.1109/JSTARS.2020.3021045.
[4] Y. Yang, S. Newsam, Bag-of-visual-words and spatial extensions for land-use classification, in:
Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic
Information Systems, pp. 270–279. doi: 10.1145/1869790.1869829.
[5] O. Cholyshkina, P. Prystavka, O. Kozachuk, V. Zivakin, Training set AERIAL SURVEY for data
recognition systems from aerial surveillance cameras, in: Proceedings of IX INTERNATIONAL
CONFERENCE Information Technology and Implementation, Kyiv, 2022, paper 79.
[6] A. A. Komarov, Intelligent on-board forest fire search system. Qualifying work for obtaining a
bachelor's degree in the specialty "Automation and computer-integrated technologies". National
Aviation University, Kyiv, 2023.
[7] N. I. Boyko, Y. I. Bidak, Effectiveness methods definition emotions of people by language on
cross-sectional datasets, in: Proceedings of International conference 'Modern research in world
science', publishing house SPC "Sci-conf.com.ua", Lviv, Ukraine, 2023.
[8] P. Prystavka, S. Dolgikh, O. Kozachuk, Terrain image recognition with unsupervised generative
representations: Effect of anomalies, in: Proceedings of 12th International Conference on
Advanced Computer Information Technologies, IEEE, Ruzomberok, Slovakia, 2022, pp. 485–488,
doi: 10.1109/ACIT54803.2022.9913178.
[9] P. Prystavka, S. Dolgikh, O. Cholyshkina, O. Kozachuk, Latent representations of terrain in
aerial image classification, in: Proceedings of International Conference on ICT in Education,
Research and Industrial Applications. Integration, Harmonization and Knowledge Transfer,
volume 3013, 2021, pp. 86–95
[10] P. Prystavka, O. Cholyshkina, O. Kozachuk, D. Yaremenko, Neural network automation of filling
the data set of aerial digital images, Information technology and society 2(4) (2022) 88–99. doi:
10.32689/maup.it.2022.2.12.
[11] TensorFlow, 2024. URL: https://www.tensorflow.org/.