We have all heard about artificial intelligence and its capabilities: news about innovative developments,
Briefly about AI
Artificial intelligence is a set of technologies andalgorithms that are able to mimic some of the cognitive functions inherent in humans. However, it is important to understand that AI technology is far from the “super brain” with which it is often associated. It is still just a technology that has no consciousness, cannot think and reason like a human.
However, there are a number of tasks that, in terms of cognitive features, are close to human thinking. It is them that artificial intelligence successfully solves, and it is customary to call them “AI tasks”.
These tasks include:
- Computer vision and object recognition: An algorithm can be shown a photo or video stream from which the program will select data and classify.
- Speech synthesis recognition: algorithms convert speech signals into digital information, which the program also classifies.
- Working with a variety of information flow, including natural language data: applicable when there is a large database.
- Decision Support: Algorithms generate a decision function.
All these tasks are the main directions inimplementation of AI elements. And all of them are already being actively integrated into our daily lives: from automated assistants on websites to "smart" cameras on city streets.
AI in ecology
AI is relevant and effective for many industries, the development of education and even culture. But it also significantly affects the transformation of the sphere of ecology.
Already now at the waste processing enterprisesthere are robots that help sort waste. The technology is being introduced into systems for monitoring and analyzing air, water bodies and soils. And each of us can meet "environmentally friendly" AI, for example, in automatic recycling points.
In general, it is impossible to single out any specifics of AIspecifically in ecology. With its help, you can significantly reduce costs, which is important for any business. So, when a real person is replaced by a “machine” in the recycling system, the whole process is automated, and the cost of servicing the devices is reduced.
How to train AI to recognize recyclables
Pattern recognition is one of the mostcommon AI tasks. For its solution, convolutional neural networks are most suitable - a model of a computer program that is closest to how a person recognizes objects in reality. The "layers" of such a network are similar to the layers of the retina.
A neural network is a simplified model of workhuman brain. Its basic elements - neurons - have a large number of connections, interconnections, which are usually grouped into layers. Each connection of neurons is assigned a certain impact force - weight. The input data to the network is fed to the first layer, then they are distributed to the next layers in accordance with the current weight of the relationships. The final result can be obtained from the last layer of the neural network.
Training a convolutional neural network isselecting the weight of the interconnections of neurons in order to obtain the correct result as a result of its work on the last layer of the network. In the case of recognition of recycled materials, two tasks are solved: segmentation - determining the area in the photo with the object and classification, understanding what kind of object it is. Therefore, in this case, two sequentially operating neural networks are used: the first receives an image as input and outputs the contours of the found objects, and the second sequentially processes the found contours and returns the belonging of each contour to a certain class of objects.
Submission of a set of examples (images) "for input"called "supervised learning". This process requires a large number of photographs, on which the necessary objects are circled and signed. When learning the technology in the recycling machine, it will be necessary to collect more than 50,000 images of objects.
Showing a large number of images "at the input" andby measuring the quality of their "output", it is possible to build and select specific neurons in the network. If the hypotheses on the choice of neurons turn out to be true, the network is trained, then the error is gradually minimizing. Ideally, the network, based on the results of training, should accurately recognize the images that were loaded into it and identify similar images.
Nuances of recognition
Crumpled plastic bottles, mangled aluminum cans, wet waste paper - how can AI understand what recyclable materials can be recycled and what fractions to determine it?
When teaching AI technology, it is important to considerthe human factor, because it will be the people who will load the recyclables, who for the most part will not care about the quality of the waste delivered. Let us clarify that quality here means purified recyclable materials that are suitable for recycling.
To keep different scenarios in mind and preparetechnology, the developers among the downloaded sample images include those very “spoiled” objects. So, AI can learn to recognize the same plastic bottles in any of their forms. For example, a bottle has a characteristic cap or a certain texture that is fixed by a net.
According to external forms, standards, textures, the fractions of the delivered raw materials are determined. And according to the data of the weight categories of the fraction, it is possible to calculate, for example, wet waste paper.
In the future, the technology learns already in the process of work: when it sees real objects handed over by people. Operators process incoming new data, select the necessary images and correct the network.
AI becomes universal over timea tool that helps to optimize various areas of production and our lives. In ecology, this is the ability to respond in a timely manner to certain circumstances, reduce costs and minimize errors that can be made due to the human factor in work.
However, like any technology, AI requirescontinuous improvement. So, in the field of recycling, additional training of smart devices occurs regularly. How far AI can improve ecological processes and improve the global environment, only time will tell. But it is already clear that the use of artificial intelligence is one of the effective steps towards our green future.
Blazar, which was found 20 years ago, turned out to be an extreme object
TESS discovered a "new Earth": a stone planet with water is in the habitable zone
Look at the consequences of the collision in 1181 of two stars