| 摘要: | 本研究提出一個新的二階神經網路，我們適當安排此網路的結構，使其能達到旋轉不變的特性，所以此神經網路可以用來辨識任意角度旋轉的物件，最特別是我們在訓練過程並不像大多數其它學者的方式需要同一物件許多旋轉樣本，此網路的訓練集中每一物件只需一版本，實驗顯示我們的方法能成功用來辨識任意角度旋轉的光碟片。
A second-order neural network is designed to be invariant to changes in rotation. Rotation invariance is achieved through a special arrangement of the network structure. The training set only requires one view of each target object. We describe the weight sharing strategy and present a compact disk recognition neural network illustrating its usefulness. The simulation results show that the proposed neural network can distinguish between the target compact disks independent of the transformation in rotation. |
