SCRS Proceedings of International Conference of Undergraduate Students
Use of Deep Learning for Classification of Machined Surfaces
Authors: Sushaan K Attavar, Srinivasa P Pai, Santhosh Pai Hosdurg and Sandesh Rao Udupi
Publishing Date: 16-01-2023
ISBN: 978-81-95502-01-1
Abstract
The process of taking an input and converting it into a class or a likelihood that it belongs to that class is termed as image classification. For use in image processing applications, traditional machine learning models necessitate formal image processing, denoising, feature extraction, and dimensionality reduction. Deep learning eliminates the requirement for it. It is a subset of machine learning that has grown in popularity as processing units have improved, data sizes have grown larger, and research in the field has increased. In this paper, CNN is used to classify images of various machined surfaces. This study shows how to identify and classify machined surfaces such as turned, shaped, and ground surfaces using a 2D-CNN, deep learning-based machine vision inspection method. The goal of utilizing this model is for the surfaces to be detected properly the majority of the time, and for the model to run efficiently even for limited datasets. This model employs ReLU, Sigmoid, softmax activation functions, max pooling layers, and optimizers to learn unusual, unique patterns and determine what should be fed to the adjacent neuron. The proposed method is used to improve performance with a large image dataset, which comprises of various machined surfaces. When small datasets are supplied to the 2D-CNN, the model is more likely to overfit. To prevent this from happening, data augmentation can be used to produce higher, near-accurate outcomes with a smaller dataset fed to the suggested model. The proposed model produces the best results and demonstrates that without the usage of external computational resources such as a GPU, CNN can perform efficiently and produce improved and near-accurate results. In addition, when comparing the model proposed in this work to traditional machine learning approaches like ANN, it can be inferred that CNN provides better outcomes and accuracy using the same datasets as inputs to the models presented.
Keywords
Image Classification, Feature Classification, 2D-CNN, Machined Surfaces, Data Augmentation, Sigmoid
Cite as
Sushaan K Attavar, Srinivasa P Pai, Santhosh Pai Hosdurg and Sandesh Rao Udupi, "Use of Deep Learning for Classification of Machined Surfaces", In: Prashant Singh Rana, Deepak Bhatia and Himanshu Arora (eds), SCRS Proceedings of International Conference of Undergraduate Students, SCRS, India, 2023, pp. 135-144. https://doi.org/10.52458/978-81-95502-01-1-15