Using Convolutional Neural Networks for Label Character Verification Applications

• Main Authors: LIN,YI-CHENG, 林奕丞
• Other Authors: LIN,HSUEH-YI
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/ua56a5

碩士 === 國立勤益科技大學 === 資訊工程系 === 107 === On the label defect detection, the traditional methods require a complete image to be a standard sample. Then get the difference value by comparing to sample. Also, the difference value and threshold which is defined by engineer determine whether it is defect. This method can get high accuracy. But when selecting the location of text can’t offset or rotate. Otherwise, it will be defined defect. In recent years, convolutional neural networks are widely used in image processing. The combination of image feature extraction and optimization algorithm by convolutional neural networks has solved many image problems. Due to the improvement of hardware, calculation is a burden no longer. Base on those reasons, this thesis proposes the model of defect detection of convolutional neural networks. Segmentation of features through convolutional neural networks and classification by fully connected networks. Finally, the defect is predicted by using the score of classification results.