Glass Bottle Bottom Inspection Based on Image Processing and Deep Learning

Abstract

This paper presents a novel glass bottle bottom inspection system based on image processing and deep learning. An image processing technique is applied to locate bottle bottom, utilizing both median filter and high pass filter to remove noises. Moreover, Hough circle transform is used for detecting region of interest (ROI) of a bottle bottom. Subsequently, a cropped image can be obtained to produce a square masked ROI image by masking unnecessary regions. A masked image is then resized and fed into a pre-trained predictive model to distinguish between normal and defectives bottle. A predictor is constructed by deep convolution neural network (CNN), consisting of three convolutional layers and two fully connected layers. The proposed model is programmed using Python supported by OpenCV and Keras. Experiment results show that the accuracies of both bottom location and defects detection are 99.00% and 98.50%, respectively. The computation time of bottom location process is equal to 22ms and it spends 48ms for classifying the defectives bottle. Hence, the proposed model not only obtains high accuracy, but also achieves real time inspection ability.