A learning artificial visual system and its application to orientation detection

This paper proposes a learning artificial visual system, the Learning Dendritic Model Artificial Visual System (DModel-AVS), for orientation detection inspired by biological visual mechanisms. The DModel-AVS consists of two layers: local orientation detection neurons layer and global orientation detection neurons layer. The local neurons detect local features of an image, utilizing dendrite model neurons. The global neurons are designed to implement global features of the image by summing the outputs of the local dendritic neurons. The backpropagation-based learning is performed only to the dendritic neurons. The effectiveness of the DModel-AVS is evaluated through several experiments comparing it with various convolutional neural network (CNN)-based orientation detection systems. Results show that the DModel-AVS is a more biologically plausible and effective solution to orientation detection, with higher accuracy, and lower learning costs. The proposed system has practical applications in various fields such as computer vision and robotics.