1. Prepare your equipment

Use a Raspberry Pi or computer, USB/HD camera, Python, an internet connection, and access to FAVORIOT. Prepare labeled images of acceptable and defective products for training.

2. Install dependencies

                    python3 -m pip install opencv-python numpy requests tensorflow
                

On a Raspberry Pi, choose TensorFlow or TensorFlow Lite packages compatible with your operating system and hardware.

3. Train a defect classification model

Start with a pretrained MobileNetV2 model, then train a binary classifier using labeled product images. Split your data into training, validation, and held-out test sets. Verify which class corresponds to a defect before deploying the model.

                    import tensorflow as tf
                    from tensorflow.keras import layers

                    base = tf.keras.applications.MobileNetV2(
                        input_shape=(224, 224, 3),
                        include_top=False,
                        weights="imagenet"
                    )
                    base.trainable = False
                    model = tf.keras.Sequential([
                        base,
                        layers.GlobalAveragePooling2D(),
                        layers.Dense(1, activation="sigmoid")
                    ])
                    model.compile(optimizer="adam",
                                  loss="binary_crossentropy",
                                  metrics=["accuracy"])
                    # Train with your labeled image dataset before use.
                

4. Inspect camera frames

Read frames with OpenCV, resize them to 224 × 224, apply the exact preprocessing used during training, and run predictions. Translate the score into “Product OK” or “Defect Detected” using the verified class mapping and a tested threshold.

                    import cv2
                    cap = cv2.VideoCapture(0)
                    ret, frame = cap.read()
                    if ret:
                        resized = cv2.resize(frame, (224, 224))
                        # Apply model-specific preprocessing and predict.
                    cap.release()
                

5. Publish results to FAVORIOT

Create a device in FAVORIOT and securely store its device developer ID and API key. The example below sends an inspection event to the FAVORIOT V2 streams endpoint.

                    import os
                    import requests

                    endpoint = "https://apiv2.favoriot.com/v2/streams"
                    payload = {
                        "device_developer_id": os.environ["FAVORIOT_DEVICE_ID"],
                        "data": {"status": "Defect Detected!"}
                    }
                    headers = {
                        "Content-Type": "application/json",
                        "Apikey": os.environ["FAVORIOT_API_KEY"]
                    }
                    response = requests.post(
                        endpoint, json=payload, headers=headers, timeout=15
                    )
                    response.raise_for_status()
                    print("Inspection result sent")
                

Run this function after classification. For real deployments, include event timestamps, confidence scores, product IDs, retries, and duplicate protection.

6. Create your inspection dashboard

Open FAVORIOT Streams to check incoming data, then build a dashboard that tracks pass/fail counts, defect rates, and inspection history. Add notifications or other response workflows after validating the end-to-end system.