We have talked to some potential customers from the manufacturing and logistics industries. All share problems with their current method for inbound goods inspection. Manual inspection requires heavy human efforts and cost while traditional expensive image recognition system still fails to recognize apparent faults that are not pre-defined in the system.   The system developed by FPT uses AI ( with cognitive visual recognition) to support product checking before product items are processed to the warehouse. This system can also be used to inspect incoming products on the production line.
  • FPT develops the AI algorithms, executes the training process, and implements the application that supports the goods inspection procedure.
  • The system helps to identify objects in abnormal conditions (e.g. dented, torn, crushed) automatically and promptly during the round check process.
Model    
  • First, cameras are installed along train tracks to gather images of train wagons as they drove by.
  • The images are then automatically uploaded to the AI processing system (using Tensor flow) on cloud. After analyzing, the system can successfully recognize abnormal or damaged products. As more training data are gathered, system‚Äôs visual recognition capabilities are improved to an accuracy rate of over 90% in just a short period of time.
  • Photos of anomalies and damages are saved as evidence.
Advantages
  • High flexibility: Applicable for various products (mixed of different shapes, size or categories). Customers can easily train the model with objects that are more appropriate for their purposes.
  • Self-improved system: AI allows the Inspection system to be self-improve gradually.
  The solution has shown noticeable effects: Reduce 30-40% process time and around 70% of human resources used for goods inspection. Video and live demo are available upon request.   Cloud computing - Machine learning - IoT