Patvaļīgi sašķiebtu numura zīmju atpazīšana
Loading...
Date
Authors
Advisor
Journal Title
Journal ISSN
Volume Title
Publisher
Latvijas Universitāte
Language
lav
Abstract
Mūsdienu pasaulē neironu tīkli ir iekarojuši lielu interesi pētnieku un inženieru vidū. Ir labi zināms, ka dziļās mašīnmācīšanās modeļi labi spēj pildīt klasifikācijas uzdevumus. Tāpēc rodas jautājums, vai neironu tīkli labi spētu klasificēt Latvijā reģistrēto automašīnu numura zīmes. Ja neironu tīkls to spēj izdarīt, tad rodas sekojoši jautājumi, vai tas to dara labāk kā citi līdzīgi risinājumi un vai tas ir efektīvs risinājums. Maģistra darba mērķis ir pārliecināties, vai neironu tīkli spēj efektīvi un precīzi klasificēt numura zīmes, izmantojot satvaru TensorFlow. Lai sasniegtu darba mērķi, tika paveikti vairāki uzdevumi: izpētītas skaitļošanas resursu izmaksas, izveidots TensorFlow modelis, iegūtie rezultāti, iespēju robežās, salīdzināti ar citiem risinājumiem.
The theme of master thesis is Recognition of arbitrarily skewed number plates. In modern world, researchers and engineers have taken a great interest in field of neural networks and deep learning. It’s a general knowledge that neural networks are good tools for classification. Building on this, a question arises – is there a way to create a neural network model that could classify number plates of Latvian registered automobiles. If there is a way then few more questions arise: will this model be in any way better than other preexisting libraries or APIs, and will this neural network be an efficient solution to the proposed problem. The main goal of the Master’s thesis is to create an efficient neural network that can precisely classify number plates, using TensorFlow. To achieve the goal, author had to accomplish several tasks such as: researching the costs of computational power, creating and programming TensorFlow model, comparison between the created model and other preexisting solutions.
The theme of master thesis is Recognition of arbitrarily skewed number plates. In modern world, researchers and engineers have taken a great interest in field of neural networks and deep learning. It’s a general knowledge that neural networks are good tools for classification. Building on this, a question arises – is there a way to create a neural network model that could classify number plates of Latvian registered automobiles. If there is a way then few more questions arise: will this model be in any way better than other preexisting libraries or APIs, and will this neural network be an efficient solution to the proposed problem. The main goal of the Master’s thesis is to create an efficient neural network that can precisely classify number plates, using TensorFlow. To achieve the goal, author had to accomplish several tasks such as: researching the costs of computational power, creating and programming TensorFlow model, comparison between the created model and other preexisting solutions.