Streszczenie:
Here a system capable of recognizing medicines in the form of pills is described. This model is supposed to solve the problem of identifying pills by image. This tool should be helpful for a big amount of people since according to statistics there has been an increase in pill use and one of the challenges of that is seen in the 44% increase in calls to U.S. poison control centers from 2003 to 2007, with the majority of those calls being about "pill identification".
Such problems are particularly important especially for the older population. According to the information published by National Library of Medicine’s [5] 9 out of 10 American citizens that are older than 65 and have more than one type of a prescribed pill are prone to misidentify those prescribed pills. Depending on the exact situation, such mistakes can lead to improper treatment, cause serious health risks or in some cases even death.
A pill identification system may also be used in industrial applications. One of the most obvious applications would be as an additional security measure. For instance, it can be used at the factories that produce the medicine, another application would be the hospitals, to make sure that patients are getting the exact medicine that was prescribed to them and to completely eliminate human error.
Another use of such a model could be to help the environment by recognizing the discarded pills and thus give an opportunity to optimize their utilization process.
Even though the focus of this paper is mostly on pill identification, it basically represents a model capable of identifying objects with rather few subtle differences, which in and of itself can be a valuable tool and has a lot of potential applications in numerous fields.
To do the classification convolutional neural networks will be used.
