A deep learning model to identify resistor bands.
This is my first attempt at a deep learning project and it didn't really go as I hoped.
The original idea was to build a model to identify both 4 and 5 band resistor bands. Once bands are identified, the resistor value could be calculated using band color and position.
In hindsight, this resistor calculation is difficult because of resistors in different orientations. For example, given a 4-band resistor with bands yellow, violet, orange, gold; What is the first band? Yellow or gold ? How would you be able to know given any arbitrary resistor?
The model also seemed to have trouble with training on both resistor types and performed much better when focused on either 4 or 5 band resistors.
I also probably didn't have enough data or variety in data to train the model very well. My dataset is pretty imbalanced as seen in notebooks/explore.ipynb.
train model - train.py
see notebooks/test-model.ipynb
Images of 4 and 5 band resistors taken with USB microscope.
- 67 values of 4 band resistors
- 60 values of 5 band resistors
- 1270 images in total
Dataset prepped and bundled using dataset.py.
Dataset should be put in ./data/ to use for training model in train.py.
Finalized dataset hosted at https://www.kaggle.com/datasets/barrettotte/resistors
# init environment
make env_init
conda activate what-the-ohm