Artificial intelligence against forest fires
The deterioration of the environment because of the human being is a problem that has been present during the length of history. In the past one of the main factors that produced the extinction of species was the massive indiscriminate hunting, either because of food or fun need. However, currently one of the main factors that extinction have produced is environmental pollution, characterized by both global warming and garbage accumulation. These facts have increased from the industrial revolution and the technological boom that occurred more than a century ago, where it has been shown that both are able to destroy complete ecosystems, alerting the scientific community.
However, industrial and technological development has given us advanced tools that, together with a society aware of the environmental damage, can help counteract the damage caused. Within the area of computer science, in recent decades, progress has been made important in artificial intelligence and machine learning, which allows you to teach computers to "think", so that they manage to perform tasks that were thought to be thought only, only the I could do a person. Based on this, various algorithms and computer techniques have been developed that support human work against natural disasters, destruction of ecosystems and climate change, as is the case of Bee2fire, an artificial intelligence capable of analyzing the environment and detecting early spotlights of forest fires.
The first computational programs that laid the basis of artificial intelligence were developed by scientists Rossenblatt and Turing in the 1950s, looking for machines to behave intelligently, based on the ideas that Isaac Asimov had raised in his science books fiction. After this computing area was founded, it was promoted exponentially together with technological development, especially during the last two decades. Currently artificial intelligence is defined as the "[…] branch of computational sciences concerned with the automation of intelligent behavior" (Ponce 13). This implies that it is a series of algorithmic steps performed by a computer, which simulate or reproduce intelligent behavior, similar to that observed in biological organisms.
And in the same way, it was tried. It is thanks to this that the concept of neural networks is born. According to Matich (2001) neural networks are processing units that exchange each other information and have two essential characteristics: these are used to recognize patterns (images, temporary sequences or others) and also have the ability to learn and with it, improve their own performance (5). This last characteristic is one of the most important qualities of intelligence, since it allows "learning from errors" and optimally performing the entrusted task.
In a biological organism, neurons are the functional basic unit of the nervous system. When the body receives a stimulus from the outside the sensory neurons send a nerve impulse to the brain. Once the impulse has reached the brain, it processes and interprets the signal through complex neuronal connections, and finally an answer is issued that can be a sensation or reflex arc. This process can be summarized in the fact that there is an input information that is then processed, and then emit new information, all through connections between one cell and another. This reasoning is used as a guide in the creation of artificial intelligence.
Within an artificial neuronal network, the basic information unit is the neuron, which stores a specific type of data determined by the manufacturer. These neurons are organized to form three layers interconnected to each other. The first is the input layer or input, which collects a specific type of data with which it will work, the second is the hidden layer or hidden layer that processes the data and finally the output or output, which delivers the desired result. Generally a neuronal network must be trained to learn to effectively perform the task for which it was manufactured. This is achieved by entering tests with known results, and "guiding" the network to increase the chances of giving a desired result according to the entry delivered.
As an application of these neuronal networks to the protection of the environment, Bee2fire has been developed, which mainly detects smoke columns both in natural environments and natural forests and reserves, as in mining areas or industrial facilities. This network has a neuronal entry layer that receives the images that capture the surveillance chamber of the sector of interest. The captured image decomposes into pixels, and these are admitted to the entrance where they are classified by their tone and intensity. They are then transferred to the hidden layers where they are prosecuted to distinguish lines, and then form edges and end up identifying objects. Finally, the results are classified in the output layer as a chance that the image entered is one of three states: Clean (clean skies), clouds (cloudy skies) or smoke (smoke columns).
The neuronal network was trained using a file of 2378 images classified with one of the three possible states. After training the network showed 98% success in the recognition and classification of the images delivered. The images that are introduced into the program once the training is applied are processed and then entered to a first filter called Pytorch Routine. If the probability that the image is classified as "smoke" is greater than 0.8 The image is reviewed by a second filter called IBM Watson Classifier, where if the probability is also greater than 0.8 The fire alarm is emitted. On the contrary, if in any of the filters the probability is less than 0.8 The program ends and does not emit any alarm.
However, when the network is implemented, a precision result close to 82% was obtained. This gives the opportunity for artificial intelligence to be improved so that it has a greater degree of efficacy, in order to use it as an ally against the combat of forest fires with an important degree of security. In addition, this achievement serves as an incentive to use and investigate this type of technologies to deal with environmental problems, such as Rainforest Connection that detects sounds of chainsaws and timber trucks in protected forests or AI for Earth, Microsoft project that Finance artificial intelligence applications that promote a sustainable future, mainly related to agriculture, water, biodiversity and climate.
In short, artificial intelligence can be a very useful tool and can also be applied to various fields and areas, such as environmental protection against problems caused by the same human being. The development of a project such as Bee2fire helps monitor and prevent possible events that end up destroying complete ecosystems in forest plantations and natural parks, also allows you to demonstrate new and innovative methodologies of practical application of intelligent neural networks of intelligent neural networks.
Leave feedback