Classification of Defects in Bushes Using Deep Learning Approaches

Abstract

Bush manufacturing industries generate thousands of bushes every day. The quality of the product is very crucial; if not it often leads to a huge loss for the manufacturer. The bushes are made up of steel material, highly reflective, and cylindrical shapes. Typically, the bushes have several types of defects that include cracks on the surface, unfinished surfaces, dent, etc. The manual inspection is carried out to check the product quality which is very tedious and time-consuming. Hence, we address the necessity for automated inspection of these bushes for classification of defect and non-defect. The conventional methods often fail to automate the process because the size of the crack is minute, or the algorithms are not efficient. In this paper, we aim to solve the problem using a deep learning approach. A combination of multiple cameras are employed to collect the sample dataset, and convolutional neural network is employed for binary and multi-class classification of the defect types. The proposed method performed better with the accuracy of 99.85% for binary classification and 89.32% for multiclass classification for the test data. In addition, we compare the accuracy with state-of-the-art deep learning techniques.