Dust Detection and Classification: A Machine Vision Approach for Automated Lens Inspection

In recent years, industrial vision has become increasingly important in manufacturing processes, revolutionizing quality control and automation. This multidisciplinary field integrates principles of optics, electronics, computer science and artificial intelligence to develop systems capable of acquiring, processing and interpreting images with greater speed and accuracy than the human eye. Its application enhances production efficiency and product quality while simultaneously reducing operational costs and the risk of errors. In this context, this thesis aims to develop an industrial vision system to support the cosmetic inspection of lenses, specifically to detect dust and distinguish it from actual defects. The project consists of several phases, beginning with the selection of the optimal hardware configuration, which includes choosing the camera, optics and illumination system. Particular attention was given to illumination, a crucial factor in ensuring sharp and uniform images regardless of the lenses' geometry and color. Two different types of illuminators were analyzed and compared to determine the most effective solution. The next phase focused on image processing, a key step in this process was the study, development and implementation of edge detection algorithms, which are essential for identifying object boundaries within an image to extract relevant features. The Laplacian filter and the Canny operator were explored in greater depth among the various techniques tested. Once the features on the lenses were identified, the project moved into the field of Machine Learning for classification purposes. The goal of this stage was to distinguish dust from defects, thereby improving the accuracy of the inspection system. To achieve this, the Support Vector Machine (SVM) algorithm was employed, a widely used supervised learning model for classification tasks. The model was trained on a dataset containing labeled features, enabling it to effectively differentiate between dust and defects. The results obtained demonstrate high accuracy, confirming the effectiveness of the proposed system in performing reliable cosmetic inspections of lenses.