TMPA-2021: Investigation of the capabilities of artificial neural networks in the problem of classifying objects with dynamic features

Nikita Laptev, Vladislav Laptev, Gerget Olga, Dmitrii Kolpashchikov and Andrey Kravchenko

Investigation of the capabilities of artificial neural networks in the problem of classifying objects with dynamic features.

TMPA is an annual International Conference on Software Testing, Machine Learning and Complex Process Analysis. The conference will focus on the application of modern methods of data science to the analysis of software quality.

To learn more about Exactpro, visit our website https://exactpro.com/

Follow us on
LinkedIn https://www.linkedin.com/company/exactpro-systems-llc
Twitter https://twitter.com/exactpro

Transcript

1. Speaker: Dmitrii Y. Kolpashchikov
   Investigation of the capabilities of
   artificial neural networks in the
   problem of classifying objects with
   dynamic
   features
   

   View Slide

2. 1
   1.1 Billion Ha
   20 000
   Average
   number of fires
   per year
   2 Million Ha
   5 Billion rub.
   The amount of
   damage to forestry
   Relevance
   Average area of forest
   fires per year
   Forest area of
   the Russian
   Federation
   

   View Slide

3. 2
   It is important to the early detect fire source, precisely localize them and take timely measures to extinguish
   them. Timely detection and appropriate action are crucial to prevent disasters, which can save lives and
   property.
   Patrolling
   Satellite monitoring
   Aerial
   monitoring
   Video monitoring
   Monitoring types
   

   View Slide

4. Low flying clouds
   No constant shape and color
   Data quality is dependent on the distance
   between the camera and fire source,
   times of the day and weather conditions
   Detection problems
   3
   

   View Slide

5. Video classification
   Selecting an area of smoke
   Localization of the source of fire
   Monitoring system tasks
   4
   

   View Slide

6. Algorithm of the system
   5
   Video
   Algorithm for
   extracting
   dynamic
   features
   Object
   detection
   LST
   M
   Fire source
   localization
   algorithm
   

   View Slide

7. Frame preprocessing
   6
   

   View Slide

8. Smoke cloud detection
   7
   

   View Slide

9. Classification
   9
   

   View Slide

10. Classification
    9
    RNN CNN
    GRU Network LSTM Network
    Fully connected neural
    networks
    Accuracy,% 74,8 77,3 69,3
    Processing time, s 0,19 0,18 0,06
    

    View Slide

11. 9
    Results
    

    View Slide

12. Results
    9
    Number of processing frames Frames size Accuracy, % Time, s
    3
    68,98 0,04
    72,12 0,07
    75,05 0,11
    77,33 0,21
    5
    79,97 0,06
    83,21 0,08
    85,4 0,12
    85,7 0,25
    7
    73,74 0,18
    75,32 0,30
    78,5 0,41
    77,79 0,75
    11
    65,74 0,27
    68,54 0,49
    70,66 0,98
    71,03 1,12
    TP FP TN FN Precision Recall Accuracy
    238 31 42 15 0,885 0,94 0,854
    

    View Slide

13. 9
    Results
    

    View Slide

14. Thank you for
    your attention
    Team:
    Laptev N.V.,
    Laptev V.V.,
    Gerget O.M.,
    Kolpashchikov.
    D.Y.,
    Kravchenko А.А.,
    

    View Slide