SPICEnet

Convolutional neural network for recognition of electrical components on schematics.

About

SPICEnet is a transfer-learnig model based on VGG16 which is able to classify the following electrical components:

• Resistor
• Capacitor
• Inductor
• Diode
• Corner
• T-Junction
• Cross
• Ground

This model is part of the PNG2SPICE project, which needs information about the lines as well, which is why the classes Corner, T-Junction and Cross exist. We recommend checking out PNG2SPICE for moreinformation.

Train it yourself

Download this repo, navigate to it and create a virtual environment:

python3 -m venv .env

Activate it:

source .env/bin/activate    # Linux
.env/Scripts/activate       # Windows

Install the requirements with

pip install -r requirements.txt

You can now add/remove data from the img_src folder, which holds a few instances of the listed components in the form of screenshots. You should adhere to the naming convention in order for the training process to go smoothly.

1. Add/remove data from img_src
2. Name the data like "T_info_N" where T is the type number (1 (Resistor) to 8 (Ground)) and N is the number of samples of that part. info can be whatever you want. For example, if you add two capacitor images, be sure to name one 2_mycap_1 and the other 2_mycap_2.
3. Modify SPICENET_PARAMS in SPICEnet.ipynb to your liking (see the documentation in section "SPICEnet parameters")
4. Run the notebook. Tensorflow will utilize your GPU if you have it set up properly.