Summary: | 碩士 === 立德管理學院 === 應用資訊研究所 === 94 === Due to the rapid technical improvement, factory automation is becoming more important. Fast object recognition is one of the important issues. In the paper, we propose a neural-network-based automatic tool recognition system.
The proposed system consists of several modules: image preprocessing, feature extraction and transformation, and neural network-based recognition. The preprocessing module extracts the contour of tool image. The feature extraction and transformation module utilizes the strength value defined as the number of times of the difference variation of arc-to-chord distance between contour points in order to select the candidates of representative dominant points, and then a split-and-merge method is used to determine the number and position of dominant points. After that, the coordinates of selected dominant points are transformed into a feature vector with translation, rotation and scaling invariance by using turning function and Fourier descriptors. Finally, a neural network module performs the recognition task. From the experimental result, the system has a excellent recognition rate.
|