NeuronalNet is going to be a small neural network that recognizes symbols uploaded by a user. These users can interact with it on a small website which is based on Microsoft's Blazor framework. Beside Blazor for the frontend, we will use gRPC on .NET for the server application. By using protocol buffers, the client will be able to send its data to the server to enhance the neural net. The goal is to create a consistent neural network over time to precisely recognize different types of symbols.
Users can upload pictures of different kinds of traffic signs. Once uploaded, the server responds with the best working neural net at that time. Thus our database expands and the corresponding neural net improves, while the client is able to predict what image has been uploaded.
The website also includes a simple dashboard which contains information about all processed traffic signs. You can check how many signs of specific sign types have been uploaded so far and what the latest neural net is capable of.
The server mainly consists of three parts: a gRPC server to handle client requests, a database to store traffic signs, entire traffic images as we call them and neural net states, and the calculation process to improve the neural net.
Any client requests are made via gRPC which seemed more straightforward than creating a dedicated REST-service or something similar. It processes incoming traffic signs/images and is able to upload neural net states. Therefore clients can easily predict images locally and do not rely on an addtional server.
The database (MariaDB) stores all the data we need on one hand, and works as a connection between the gRPC service and calculation process on the other hand. First one inserts new images and extracts neural net states while second one extracts all images available and inserts enhanced version of the neural net.
As already mentioned, the calculation process impvoes the neural net while using all images which are held by the database. It's basically the heart of the image classification.
| Key | Value |
|---|---|
| MARIADB_PASSWORD | database user password |
| NEURO__DB_CONNECTION_STRING | credentials for database connection |