opencv- Image Preprocessing

.dockerignore

@@ -0,0 +1,13 @@
+.git
+.github
+__pycache__
+*.pyc
+*.pyo
+.pytest_cache
+.mypy_cache
+.ruff_cache
+logs/
+.env
+*.ipynb
+Image Processing/
+image/

.env.example

@@ -0,0 +1,16 @@
+# PCB Defect Detection - environment configuration.
+# Copy to .env and edit what you need. All keys are prefixed with PCB_.
+# Requires python-dotenv for automatic loading.
+
+# Paths (leave blank to use the project-relative defaults).
+PCB_MODEL_PATH=
+PCB_GOLDEN_DIR=
+PCB_LOG_DIR=logs
+
+# API
+PCB_API_HOST=0.0.0.0
+PCB_API_PORT=8000
+PCB_MAX_UPLOAD_MB=10
+
+# Logging: DEBUG, INFO, WARNING, ERROR, CRITICAL
+PCB_LOG_LEVEL=INFO

.github/workflows/ci.yml

@@ -0,0 +1,46 @@
+name: CI
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  lint:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: pip install ruff
+      - run: ruff check .
+      - run: ruff format --check .
+
+  typecheck:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: pip install mypy torch torchvision Pillow numpy imagehash scikit-image matplotlib streamlit fastapi pydantic uvicorn python-dotenv
+      - run: mypy inference_new.py app.py api.py config.py
+
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+      - run: pip install torch torchvision Pillow numpy imagehash scikit-image matplotlib streamlit pytest fastapi uvicorn python-multipart python-dotenv httpx --index-url https://download.pytorch.org/whl/cpu --extra-index-url https://pypi.org/simple
+      - run: pytest -v
+
+  docker:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+      - uses: docker/setup-buildx-action@v3
+      - run: docker build --tag pcb-defect-detection:ci .

.gitignore

@@ -0,0 +1,34 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+dist/
+build/
+
+# Tool caches
+.pytest_cache/
+.ruff_cache/
+.mypy_cache/
+
+# Virtual envs
+my_virtual_env/
+.venv/
+venv/
+
+# Model weights (large binaries)
+model/*.pth
+
+# Logs
+logs/
+*.log
+
+# Env files
+.env
+
+# IDE
+.vscode/
+.idea/
+
+# OS
+Thumbs.db
+.DS_Store

Dockerfile

@@ -0,0 +1,24 @@
+# syntax=docker/dockerfile:1
+FROM python:3.11-slim
+
+# System libs needed by scikit-image / Pillow, plus curl for the healthcheck.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libglib2.0-0 \
+        curl \
+    && rm -rf /var/lib/apt/lists/*
+
+WORKDIR /app
+
+# Deps first so they stay cached across source changes.
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+RUN mkdir -p logs
+
+# 8000 = REST API, 8501 = Streamlit.
+EXPOSE 8000
+EXPOSE 8501
+
+CMD ["uvicorn", "api:app", "--host", "0.0.0.0", "--port", "8000"]

README.md

@@ -1 +1,157 @@
-# AI_PCB_Defect_Detection_Classification_System
+# AI PCB Defect Detection and Classification
+A small system for finding and labeling defects on printed circuit boards. It
+compares a test image against a defect-free reference, flags the regions that
+look wrong, and classifies each one with a CNN.
+
+Built during my internship and then cleaned up afterwards so it can actually be
+run, tested, and deployed.
+
+
+## What it does
+
+Given a PCB image, the pipeline:
+
+1. Picks the closest golden (defect-free) reference from a folder of known
+   boards, using perceptual hashing.
+2. Slides a window across the image and compares each patch to the matching
+   patch on the reference using SSIM.
+3. Sends the patches that differ through a ResNet-50 classifier trained on the
+   DeepPCB dataset.
+4. Runs Non-Max Suppression so overlapping detections collapse into one.
+
+Output is a list of boxes with labels and confidences, and an annotated copy of
+the input image.
+
+
+## Defect classes
+
+From the DeepPCB dataset:
+
+- missing_hole
+- mouse_bite
+- open_circuit
+- short
+- spur
+- spurious_copper
+
+
+## Stack
+
+- PyTorch / torchvision (ResNet-50)
+- scikit-image (SSIM), ImageHash (pHash), NumPy, Pillow
+- Streamlit for the UI, FastAPI for the API
+- pytest, ruff, mypy
+- Docker + docker-compose
+- GitHub Actions for CI
+
+
+## Layout
+
+```
+app.py              Streamlit UI
+api.py              FastAPI endpoints
+inference_new.py    Detection pipeline
+config.py           Paths, ports, logging, env vars
+model/              Trained weights (.pth, gitignored)
+PCB_USED/           Golden reference images
+tests/              pytest suite
+```
+
+
+## Running it
+
+Clone and install:
+
+```bash
+git clone https://github.com/AradhyaStuti/AI_PCB_Defect_Detection_Classification_System.git
+cd AI_PCB_Defect_Detection_Classification_System
+python -m venv venv
+venv\Scripts\activate          # on Windows
+pip install -r requirements.txt
+```
+
+Then either the UI:
+
+```bash
+streamlit run app.py
+```
+
+or the API:
+
+```bash
+uvicorn api:app --reload
+```
+
+You need the trained weights at `model/best_resnet50_pcb_defects_50epochs.pth`
+and some golden images in `PCB_USED/`. Paths can be overridden via env vars
+(see `.env.example`).
+
+
+### Docker
+
+```bash
+docker-compose up --build
+```
+
+Brings up the API on `:8000` and the Streamlit UI on `:8501`.
+
+
+## API
+
+```
+GET  /health     health check
+POST /detect     upload a PCB image, get back detections
+```
+
+`/detect` takes a multipart file upload and returns JSON with the defect
+count, inference time, and a list of `{label, confidence, box}` entries.
+
+
+## Tests
+
+```bash
+pytest -v
+```
+
+The tests mock the model, so you don't need the weights to run them.
+
+
+## Config
+
+All runtime settings are environment variables prefixed with `PCB_`. Copy
+`.env.example` to `.env` and edit what you need. See `config.py` for the full
+list.
+
+
+## What I did during the internship
+
+- Got familiar with the DeepPCB dataset and the defect categories
+- Wrote the preprocessing and region-extraction logic
+- Trained the ResNet-50 classifier (`trained.ipynb`)
+- Put together the detection pipeline (`inference.ipynb`)
+- Built a Streamlit UI to test it interactively
+
+
+## What I added afterwards
+
+- Moved config into a single file driven by env vars
+- Rewrote the inference pipeline as a class so it loads the model once
+- Added batching so classification isn't one-patch-at-a-time
+- Added a FastAPI layer and a Dockerfile / compose setup
+- Wrote a pytest suite and a GitHub Actions workflow
+- Added logging to a rotating file
+
+
+## Things I'd still like to try
+
+- Transformer-based backbones (ViT, Swin)
+- Multi-scale windowing for defects of different sizes
+- Real PCB images instead of only the DeepPCB ones
+
+
+## Author
+
+Aradhya Stuti
+
+- https://github.com/AradhyaStuti
+- https://www.linkedin.com/in/aradhya-stuti-9b2b9529a