← All projects
AI / MLComputer Vision · CNN
Defect Detection Image Classifier
A custom CNN separating defective from non-defective manufactured parts.
01 · The problem
Manual visual inspection of parts is slow and inconsistent on the line.
02 · Approach
- 1
Generated a synthetic 400-image dataset with an 80/20 train/val split.
- 2
Built a 3-block Conv2D network from scratch in TensorFlow/Keras.
- 3
Used max-pooling, dense layers, and dropout to regularize.
03 · Outcome
Achieved perfect scores on validation for this controlled synthetic dataset.
Next projectCollaborative Filtering Recommender→