A research of granules screening on a filling production line

• Main Authors: LIN, KUAN-WEI, 林冠瑋
• Other Authors: LEE, JIANN-SHU
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/c4fap8

碩士 === 國立臺南大學 === 資訊工程學系碩士班 === 106 === Along with the progress of information and communications of technology, automatic detection technology has been replacing the traditional visual testing in Manufacturing. Our research focuses on medicine granules screening on a filling production line. When granules objects are on the production line, they would jump off the production line or flipped lead by vibration. And it’ll cause the appearance of object changed, which lead to the screening being harder. Our research’s goal is developing a system for medicine granules screening on a filling production line. With convolutional neural network base on transfer learning strategy, the last fully connected layer of a seven layer AlexNet pretrained by ImageNet will be replace by new fully connected layer which can detect whether the object is complete. Through retrain the CNN by new sample which consist of complete object and imcomplete object, speed up training and raise the performance of the system. Finally area under the Receiver Operating Characteristic of two kind of object can reach 1, and the third kind can reach 0.9243.