Inquiry About the Most Suitable AI Model for Inspecting Product Packaging Damage

(3) Share

Report

Posted on by CU16040450-0

We would like to know the most suitable AI model for inspecting product packaging damage.

In our logistics process, we want to determine whether product packaging should be classified as acceptable or defective based on images of the package exterior.

More specifically, we are considering a system in which a user uploads an image, and the image is analyzed using a model trained on previously accumulated image data to produce a result.

We plan to continuously increase the amount of training data and expect the model accuracy to improve over time.

There are two main types of damage we want to detect: crushing and dents. As we build the dataset, we plan to label each image by damage type and inspection result.

We believe this would be difficult to achieve using Copilot Studio alone. Instead, we think it may be possible by integrating Copilot Studio with a trainable AI service, such as AI Builder Object Detection or Azure AI Custom Vision.

If anyone has recommendations on the most suitable AI model or approach for this use case, we would greatly appreciate your advice.

Thank you in advance.

Categories:

General topics

I have the same question (0)

All responses (4)

Answers (0)

Sort by

Suggested answer

11manish 2,281 on at

Like (1)

Report
Copy link

Link copied!

The most suitable approach is to use Azure AI Custom Vision to build an image classification (acceptable vs defective) and optionally object detection model (dent, crush).

This allows continuous retraining as your dataset grows and integrates well with Power Platform.

Copilot Studio can be used as a front-end, but not for model training itself.

Was this reply helpful? Yes No
Suggested answer

Valantis 5,139 on at

Like (2)

Report
Copy link

Link copied!

Hi @CU16040450-0,

Your thinking is on the right track Copilot Studio alone won't cut it for this, you do need a trainable vision model behind it.

One thing worth flagging before you go down the Custom Vision path: Azure Custom Vision is being retired by Microsoft and they are directing new projects away from it. Their own docs also explicitly state it is not optimal for detecting subtle differences like dents in quality assurance scenarios. Crushing would likely be fine since it's a more obvious deformation, but dents could be a problem depending on your packaging material and lighting.

Here is how I'd approach it:

If you want to stay inside the Power Platform ecosystem, start with AI Builder custom image classification. You label your images directly in the Power Platform maker portal (acceptable / defective, and by damage type), train the model there, and call it from a canvas app or Power Automate flow. It retrains automatically as you add more labeled images, which fits your plan to grow the dataset over time. Integration with Copilot Studio is also straightforward from there. The trade-off is it's less powerful than dedicated Azure services for very subtle damage.

If accuracy on dent detection turns out to be a problem as your dataset grows, the current Microsoft-recommended path is Azure Machine Learning AutoML Vision. More setup required since it lives in Azure rather than Power Platform, but significantly better accuracy potential for fine-grained defects and no retirement risk.

I'd recommend starting with AI Builder to get a proof of concept running quickly with your existing labeled images, measure the accuracy on real packaging photos, and move to Azure ML AutoML if you hit a ceiling on dent detection quality.

Ref: https://learn.microsoft.com/en-us/ai-builder/overview

Ref: https://learn.microsoft.com/en-us/azure/ai-services/custom-vision-service/overview

Best regards,

Valantis

✅ If this helped solve your issue, please Accept as Solution so others can find it quickly.

❤️ If it didn’t fully solve it but was still useful, please click “Yes” on “Was this reply helpful?” or leave a Like :).

🏷️ For follow-ups @Valantis.

📝 https://valantisond365.com/

💼 LinkedIn

▶️ YouTube

Was this reply helpful? Yes No
Suggested answer

Haque 2,944 on at

Like (1)

Report
Copy link

Link copied!
Hi @CU16040450-0,

Seems the requirement needs a pure and complete computer vision (deep learning) based solution. Before going to the best AI model selection I want to bring up some points that may help you to select the best selection.

Common AI Model Approaches

Convolutional Neural Networks (CNNs): CNNs-used for image classification and defect detection. They can be trained to identify various types of packaging damage such as dents, tears, misalignments, and contamination.

Semantic Segmentation Models: Classifies each pixel in an image, allowing precise localization of defects. Popular architectures are U-Net, DeepLab, and Mask R-CNN.

Object Detection Models: Models like YOLO (You Only Look Once), SSD (Single Shot MultiBox Detector), and Faster R-CNN detect and localize defects as bounding boxes. Useful for identifying multiple defect types in a single image.

Multi-Modal Fusion Models: Combine visual data with other sensor inputs (e.g., infrared, X-ray) for enhanced defect detection.

Inspectoin neeeded by Logistics Process

For inspecting product packaging damage, deep learning models based on CNNs, semantic segmentation, and object detection are the most suitable. In a logistics process where the goal is to classify product packaging as acceptable or defective based on exterior images, these models can be trained to perform binary classification or defect detection. Key considerations for your logistics use case:

Binary Classification: CNN-based classifiers - categorize images into "acceptable" or "defective".

Defect Localization: Object detection models - highlight specific damaged areas, aiding in quality control decisions.

Image Quality: High-resolution images of the package exterior improve model accuracy.

Real-Time Processing: Lightweight models like YOLO can enable fast, on-the-fly inspection in logistics workflows.

Automation Integration: These AI models can be integrated into conveyor belt camera systems or mobile inspection devices for seamless detection.

Dataset Considertions: Key considerations for dataset and damage type labeling:

Multi-Label Annotation: Images may have multiple damage types; ensure your labeling system supports multi-label classification.

Consistent Labeling Guidelines: Define clear criteria for what constitutes crushing versus dents to maintain labeling consistency.

Balanced Dataset: Aim for a balanced number of examples for each damage type to avoid model bias.

Use of Annotation Tools: Employ annotation tools that facilitate efficient and accurate labeling of damage types.

Managed AI Services for Model Training and Deployment and Integration

Given the complexity of training and continuously improving AI models for packaging damage detection, integrating Copilot Studio with managed AI services is a practical approach. Two prominent Microsoft-managed services suitable for this use case are AI Builder Object Detection and Azure AI Custom Vision.

AI Builder Object Detection

Part of the Power Platform, AI Builder offers a low-code/no-code interface for training custom object detection models.

Supports labeling images by damage type (e.g., crushing, dents) and inspection result.

Automatically retrains models as new labeled data is added, aligning with your plan for continuous improvement.

Easily integrates with Power Apps, Power Automate, and Copilot Studio for seamless workflow automation.

Best suited for teams preferring rapid prototyping and integration within the Power Platform ecosystem.

Azure AI Custom Vision

A more advanced, dedicated AI service for training and deploying custom vision models.

Provides greater flexibility and control over model training, including fine-tuning and export options.

Supports detailed object detection and classification tasks, suitable for subtle damage types like dents.

Integrates with Azure Functions and can be connected to Copilot Studio via Power Automate flows or custom connectors.

Recommended for scenarios requiring higher accuracy and scalability beyond Power Platform capabilities.

Which one is the best and how to use:

I would suggesgt Start with AI Builder Object Detection to quickly build a proof of concept within the Power Platform, leveraging its ease of use and automatic retraining. In other side Evaluate Azure AI Custom Vision if your accuracy requirements increase or if you need more advanced customization and scalability.

NOTE: Both services can be integrated with Copilot Studio, where Copilot handles user interaction and workflow orchestration, while the managed AI service performs the model inference and training.

References:

AI Builder Object Detection Lab for Power Platfrom World Tour.

AI Builder vs Custom Vision in Azure.

I am sure some clues I tried to give. If these clues help to resolve the issue brought you by here, please don't forget to check the box Does this answer your question? At the same time, I am pretty sure you have liked the response!

Was this reply helpful? Yes No
Valantis 5,139 on at

Like (1)

Report
Copy link

Link copied!

Hi @CU16040450-0,

Just wanted to check in and see if everything is working now. If you still need any help, feel free to let me know.

Also, if the issue is resolved, it would be great if you could mark the answer as solved so others with the same question can find it easily.

Thanks and have a great day!

Was this reply helpful? Yes No