Twenty-five million attempts – that is how many were needed for Eng. Georgi Harbaliyev – a master's graduate in the specialty "Computer Systems and Technologies" at the Technical University – Plovdiv branch – to train his created virtual drone to detect and extinguish forest fires. Eng. Harbaliyev does not claim sole credit for the idea to develop a virtual flying machine and give it the reasoning and experience to extinguish fires. "I needed courage and scientific support to do something that had not been done before," shares the young specialist. Here, the support and even the idea came from Assoc. Prof. Dr. Eng. Dilyana Budakova, who is a specialist in artificial intelligence and head of the Department of "Computer Systems and Technologies" at the Technical University – Plovdiv branch. "It is not enough for a student to simply be an honor student, but they must have the desire to attempt scientific developments for which not everything can be read in textbooks," explains Assoc. Prof. Budakova. That is exactly why she became the scientific supervisor of his diploma thesis "Detection and Extinguishing of Fire by Using Machine Learning Algorithms via Simulation." Thus, the graduate student and his scientific supervisor decided that the most effective way to extinguish a fire in the mountains is with a drone, because the machine can fly high where it is safe for it. The drone has another advantage – the "firefighting" drone flying from above precisely defines the boundaries of the fire. Thus, the fire can be extinguished faster. And if there is a "swarm" of drones, even larger fires will certainly be extinguished even faster. "In my diploma thesis, I developed only the theoretical formulation for extinguishing forest fires. The created model, which includes artificial intelligence methods, serves to train the flying machine and is a means for rapid extinguishing. In fact, the firefighting machine can be not only a drone, but also a helicopter or ground equipment. However, I have not created models for the others." – explains Eng. Georgi Harbaliyev. How the virtual drone "reasons." The "reasoning" of the drone created by Georgi Harbaliyev follows human logic: trial/error. The young graduate is the person who says what is correct and what is not. For correct behavior, the "reasoning" of the virtual drone receives praise, and when it makes a mistake, it receives punishment. In this way, the thinking machine learns when it is flying at the correct altitude, when it maintains the correct tilt so as not to fall, when to turn on the water stream, and how to direct it towards the burning tree in the forest with the precision of a sniper. Assoc. Prof. Budakova explains: "The drone model is actually a large neural network with many layers and many neurons in each layer. The training of the neural network is repeated millions of times until the effective detection, positioning, and extinguishing of the fire begins." And yet, how does the virtual drone recognize the fire? Eng. Harbaliyev's thinking drone recognizes which tree is burning through a computer vision system. However, there is a possibility that this could be done by the difference in temperature, but such an approach requires further training. A very important question is whether the virtual firefighting drone created at the Technical University – Plovdiv branch can become real. Especially since the university has a special scientific and educational direction "Aviation Technology and Technologies," where students and scientists develop real machines. Here is the opinion of Assoc. Prof. Dr. Dilyana Budakova: "Yes, a real drone for extinguishing forest fires can be created. But it will be more complex due to the computational requirements of the networks. For now, drones are not autonomous. The electronics inside it must be specially designed so that it can autonomously execute missions. Such a drone does not exist at the moment, but it could. And it could use the intelligence developed by my graduate student."
plovdiv24.bg, 13.02.2024
