Supervised Defect Segmentation: Fine Defect Detection

In this case, we aim to detect very small defects within the images.

../_images/%E7%BB%86%E5%B0%8F%E7%BC%BA%E9%99%B7.png

Accurate Supervised Defect Segmentation Model Test

../_images/%E7%BB%86%E5%B0%8F%E5%87%86%E7%A1%AE%E8%AE%AD%E7%BB%83.png

  • The accurate supervised defect segmentation model was trained, validated, and tested on a dataset of 53 images.

Test results for the accurate supervised defect segmentation model:

../_images/%E5%87%86%E7%A1%AE%E6%A3%80%E6%B5%8BOK.png ../_images/%E5%87%86%E7%A1%AE%E6%A3%80%E6%B5%8BNG.png

From the test results, we can see that the accurate supervised defect segmentation model struggles to detect very small defects.

Fast Supervised Defect Segmentation Model Test

../_images/%E7%BB%86%E5%B0%8F%E5%BF%AB%E9%80%9F%E8%AE%AD%E7%BB%83.png ../_images/%E7%BB%86%E5%B0%8F%E5%BF%AB%E9%80%9F%E8%AE%AD%E7%BB%831.png

  • The fast supervised defect segmentation model was also trained, validated, and tested on the same 53-image dataset. Additionally, a preprocessing step was applied to resize images to 1536 pixels.

Test results for the fast supervised defect segmentation model:

../_images/%E5%BF%AB%E9%80%9F%E6%A3%80%E6%B5%8BOK.png

From the test results, the fast supervised defect segmentation model is capable of detecting even very small defects.

Conclusion

  • In this case, the fast supervised defect segmentation model demonstrated significantly better performance in detecting fine defects by using a preprocessing step to resize images.

  • In practical deployment, the model automatically resizes the input image to the preprocessed resolution to ensure accuracy and consistency.

  • Therefore, preprocessing is a key step in improving the performance of supervised defect segmentation models, especially for fine defect detection.