Detection and Classification of Irregularly Shaped Objects using Machine Learning Algorithms for Improved Accuracy

The aim of this research is to improve accuracy and efficiency with regard to product defect detection using a deep learning based hybrid model. It combines preprocessing, segmentation, feature extraction, and classification methods such as regionally adaptive Kuwahara filter, Intuitionistic Fuzzy C-Means segmentation, and boundary descriptors. For image classification purposes, a region-based adaptive Kuwahara filter is employed for both noise reduction and contrast enhancement. For intuitive segmentation, Intuitionistic Fuzzy C-Means (IFCM) clustering is applied to partition the images into regions of similar features for easier distinction. R-CNN and ultrasonic testing are used to discover flaws in a variety of materials, including tears, porosity, cracks, scratches, and problems with coloring. The ViT model is appropriate for scientific and industrial applications due to its increased accuracy and decreased variability.The results obtained showed a mean classification accuracy of 90.7400, which indicated performance stability over many and complex shapes. This model shows a standard deviation of 2.44193 which means that the algorithm predictions were robust and consistent with respect to a l test cases that were conducted. In this work, it is obtained that the ViT model is significantly better than CNN model where it has an increased accuracy.