|
SMART FOOD DISPENSER DRONE FOR STRAY DOGS AND CATS
|
|
|
P. Balazy; P. Gut; P. Knap; B. Marczynski
|
|
|
||
|
|
|
|
1314-2704
|
|
|
||
|
English
|
|
|
21
|
|
|
6.2
|
|
|
• Prof. DSc. Oleksandr Trofymchuk, UKRAINE
• Prof. Dr. hab. oec. Baiba Rivza, LATVIA |
|
|
||
|
A severe problem affecting urban areas is a large number of stray animals,
either r eject- ed by their owners or born in the wild. These animals face the problem of malnutrition, as it is difficult for them to obtain food without human help in urban conditions. The fact that shelters are overcrowded is not suitable for these animals as well well. In o rder to protect them from the tragic death of starvation, the authors decided to de- velop an intelligent system for feeding stray animals. The system makes it possible to recognize an animal and dispense the appropriate amount of food for it, taking into ac- count the species and size. This task is accomplished using convolutional artificial neu- ral networks to analyze camera images. A Raspberry Pi minicomputer was used as the computing platform. The device is energy independent. Powered by an in tegrated pho- tov oltaic module, it does not require a permanent power source. This makes it easy to station it in places where electricity is not directly available, such as parks. An addition- al advantage of this device is its mobility – the food dispenser i s a dron e that is be able to move around the area where it is currently located, as it is moving on wheels. This allows the drone to change its position and operate over a larger area. This paper presents the classification performance of four neural netwo rks. The transf er learning method was used to train the models. It allows adapting a pre pre-trained algorithm in a completely new application. The Matlab environment was used to train the model, from which the finished models were then exported to the general format for tra nsfer- ring neural networks - ONNX. The trained algorithm was then implemented on a Rasp- berry Pi single single-board computer equipped with an AI hardware accelerator - Neural Compute Stick 2. The presented application fits into the idea of signal pr ocessing at the edge (Edge AI). Brief description of the study and used methods. Brief description of the study and used methods. Brief description of the study and used methods. Brief description of the study and used methods. Brief description of the stu dy and u sed met methods. |
|
|
conference
|
|
|
||
|
||
|
21st International Multidisciplinary Scientific GeoConference SGEM 2021
|
|
|
21st International Multidisciplinary Scientific GeoConference SGEM 2021, 7 - 10 December, 2021
|
|
|
Proceedings Paper
|
|
|
STEF92 Technology
|
|
|
SGEM International Multidisciplinary Scientific GeoConference
|
|
|
SWS Scholarly Society; Acad Sci Czech Republ; Latvian Acad Sci; Polish Acad Sci; Serbian Acad Sci & Arts; Natl Acad Sci Ukraine; Natl Acad Sci Armenia; Sci Council Japan; European Acad Sci, Arts & Letters; Acad Fine Arts Zagreb Croatia; Croatian Acad Sci
|
|
|
47-54
|
|
|
7 - 10 December, 2021
|
|
|
website
|
|
|
cdrom
|
|
|
8359
|
|
|
smart; dispenser; stray; animals; networks
|
|