Image Classification Applied to the Detection of Leather Defects for Smart Manufacturing

In the shoe production workshops, animal leather is used as the main raw material. Generally, an operator manually checks the surface of the leather, making sure that it does not present defects that compromise the quality of the final product. This type of inspection is subject to human error and uncontrollable factors, which represents an opportunity for the automation of the process through a system of artificial vision. A data set was developed consisting of images of animal leather, in good coordination and with defects. The digitized samples were subjected to image processing using OpenCV and Scikit-Learn, and then used in a convolutional neural network interfacing, using TensorFlow’s Keras library in Python. Finally, the trained model is capable of classifying new images into two possible groups: “Defective Leather” and “Defect-free Leather”. The trained model offers 80% predictive accuracy and 85% reliability. Although the result can be considered satisfactory, it is expected to raise the mentioned percentage with a more robust data set than the one used for the project.