saann's picture
Fix weights filename
b5b4dbb
from fastapi import FastAPI, File, UploadFile, HTTPException
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input
from fastapi.staticfiles import StaticFiles
from fastapi.templating import Jinja2Templates
from fastapi.requests import Request
from fastapi.responses import JSONResponse
import numpy as np
from PIL import Image
import io
import os
import json
import tensorflow as tf
app = FastAPI(title="PCB Defect Detection API")
# Mount static files and templates
app.mount("/static", StaticFiles(directory="static"), name="static")
templates = Jinja2Templates(directory="templates")
# ── Load model on startup ─────────────────────────────────────────────────────
MODEL_PATH = "pcb_model.keras"
CLASS_PATH = "class_names.json"
IMG_SIZE = (224, 224)
model = None
class_names = {}
def build_model():
from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.layers import Dense, GlobalAveragePooling2D, Dropout, BatchNormalization
from tensorflow.keras.models import Model
base = MobileNetV2(input_shape=(224, 224, 3), include_top=False, weights=None)
x = GlobalAveragePooling2D()(base.output)
x = BatchNormalization()(x)
x = Dense(512, activation="relu")(x)
x = Dropout(0.4)(x)
x = Dense(256, activation="relu")(x)
x = Dropout(0.3)(x)
out = Dense(6, activation="softmax")(x)
return Model(base.input, out)
@app.on_event("startup")
async def load_model():
global model, class_names
weights_path = "pcb_weights.weights.h5"
if os.path.exists(weights_path):
model = build_model()
model.load_weights(weights_path)
print("✅ Model loaded successfully")
else:
print("⚠️ Model not found — using demo mode")
if os.path.exists(CLASS_PATH):
with open(CLASS_PATH) as f:
class_names = json.load(f)
else:
class_names = {
"0": "missing_hole",
"1": "mouse_bite",
"2": "open_circuit",
"3": "short",
"4": "spur",
"5": "spurious_copper"
}
# ── Defect descriptions ───────────────────────────────────────────────────────
DEFECT_INFO = {
"missing_hole": {
"label": "Missing Hole",
"description": "A required drill hole is absent from the PCB. This prevents component mounting and causes assembly failure.",
"severity": "High",
"color": "#FF4444"
},
"mouse_bite": {
"label": "Mouse Bite",
"description": "Small notches or indentations along the PCB edge, resembling bite marks. Usually caused by routing errors.",
"severity": "Medium",
"color": "#FF8C00"
},
"open_circuit": {
"label": "Open Circuit",
"description": "A broken trace or gap in the copper path that interrupts electrical continuity.",
"severity": "High",
"color": "#FF4444"
},
"short": {
"label": "Short Circuit",
"description": "Unintended connection between two copper traces, causing electrical short that can damage components.",
"severity": "Critical",
"color": "#CC0000"
},
"spur": {
"label": "Spur",
"description": "An unwanted protrusion on a copper trace. Can cause unintended connections with adjacent traces.",
"severity": "Low",
"color": "#28A745"
},
"spurious_copper": {
"label": "Spurious Copper",
"description": "Extra copper remaining on the board after etching. Can cause short circuits if near other traces.",
"severity": "Medium",
"color": "#FF8C00"
}
}
def preprocess_image(image_bytes: bytes) -> np.ndarray:
img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
img = img.resize(IMG_SIZE)
arr = np.array(img, dtype=np.float32)
arr = preprocess_input(arr)
return np.expand_dims(arr, axis=0)
@app.get("/")
async def home(request: Request):
return templates.TemplateResponse("index.html", {"request": request})
@app.post("/predict")
async def predict(file: UploadFile = File(...)):
# Validate file type
if not file.content_type.startswith("image/"):
raise HTTPException(status_code=400, detail="File must be an image.")
contents = await file.read()
if len(contents) > 10 * 1024 * 1024:
raise HTTPException(status_code=400, detail="Image too large. Max 10MB.")
try:
img_array = preprocess_image(contents)
except Exception:
raise HTTPException(status_code=400, detail="Could not process image.")
if model is None:
# Demo mode — return mock prediction
import random
classes = list(DEFECT_INFO.keys())
detected = random.choice(classes)
confidence = round(random.uniform(0.75, 0.97), 4)
top3 = random.sample(classes, 3)
top3_scores = sorted([round(random.uniform(0.01, 0.3), 4) for _ in top3], reverse=True)
top3_results = [{"class": c, "confidence": round(s * 100, 2)} for c, s in zip(top3, top3_scores)]
top3_results[0] = {"class": detected, "confidence": round(confidence * 100, 2)}
else:
preds = model.predict(img_array)[0]
top_idx = int(np.argmax(preds))
confidence = float(preds[top_idx])
detected = class_names.get(str(top_idx), "unknown")
top3_idx = np.argsort(preds)[::-1][:3]
top3_results = [
{"class": class_names.get(str(i), "unknown"), "confidence": round(float(preds[i]) * 100, 2)}
for i in top3_idx
]
info = DEFECT_INFO.get(detected, {
"label": detected.replace("_", " ").title(),
"description": "Defect detected on the PCB surface.",
"severity": "Unknown",
"color": "#666"
})
return JSONResponse({
"success": True,
"defect": detected,
"label": info["label"],
"confidence": round(confidence * 100, 2),
"severity": info["severity"],
"color": info["color"],
"description":info["description"],
"top3": top3_results
})
@app.get("/health")
async def health():
return {"status": "ok", "model_loaded": model is not None}