shouvikcirca / pcb_defect_classification

Machine Learning Models to classify Printed Circuit Boards

Datasets

Samples

• 298 samples of 300X300 images with labels
• 5040 samples of 300x300 images with labels
• 2669 samples of 300x300 images with labels

Labels

• 0 -> Defective
• 1 -> Non Defective

File Description

slenet.py Running the file gives inference details on the raw dataset
slenet.ipynb Contains all the experimentation done while creating slenet
slenet.th The Slenet Model
salexnet.py Running the file gives inference details on the raw dataset
salexnet.ipynb Contains all the experimentation done while creating salexnet
salexnet.th The Salexnet Model
VGG_pretrained.ipynb Experimentations for transfer learning using VGG Net

Installing Modules

$ pip3 install -r requirements.txt

Running Tensorboard Visualizations

$ tensorboard --logdir=learning_rates

should give

TensorBoard 1.14.0 at http://shouvik-Predator-PH315-51:6006/ (Press CTRL+C to quit)

Open the URL in a browser

Inference

$ python3 slenet.py

should give

RawAccuracy:0.6234544773323342
Raw Dataset Confusion Matrix
[[ 147    2]
[1003 1517]]
Raw Dataset F1-Score: 0.7218757844113998
0 misclassification: 1.342281879194631 %
1 misclassification: 39.8015873015873 %