Overview / Usage

Soap bar products mostly have so many unique details such as engravings, letters, numbers, or logos that represented the label of the product. But the problem that often occurs in the production process is that the details on the bar soap are often damaged due to scratches, dust, dirt, etc. As a result, traditional quality control cannot prevent product defects from reaching consumers.

Therefore, AI bar soap smallest defects detection has been deployed to help manufacturers rapidly to identify defects in real-time. Leveraging deep learning techniques successfully to automated detections without human interactions. Based on a small training set images, the product defects similar and previously unseen defects on its own.

Methodology / Approach

The project is broken down into three phases:

• Phase 1: Preparing data for training process. In this phase, the data consists the images that collected to build a model by the Intel OpenVino toolkits. Images were taken by a web camera. For the best result, the images kindly were taken up to 500 frames.
• Phase 2: Deploying inference. After the models were build and have been trained, move to the next phase, deploying inference. In this phase, the models converted to the inference engine format using thhe model optimizer.
• Phase 3: Implementation into a real application. The last phase, deploy an application. It's so many features that I've created, both hardware and software. It is currently can be deployed on production lines equipped and integrated with an overhead camera and a conveyor.

Technologies Used

This project has been tested on a desktop PC with:

• NVDIA GeForce GTX-1080Ti
• Ubuntu 16.04.5 LTS (x86_64)
• CUDA 9.2
• cuDNN 7.1.4
• TensorFlow 1.10.0