Bio-inspired interactive feedback neural networks for edge detection
Chuan Lin,Yakun Qiao,Yongcai Pan
DOI: https://doi.org/10.1007/s10489-022-04316-3
IF: 5.3
2022-12-03
Applied Intelligence
Abstract:In recent years, deep learning technology has significantly improved the performance of various computer vision tasks. Convolutional neural networks for edge detection tasks are usually composed of two parts: encoding and decoding. Encoding is mainly used for feature extraction and characterization; while decoding is mainly used to effectively integrate the local, detailed features extracted from encoding into global information. The research suggests that convolutional neural networks are a technical replication of biological neural networks in a simplified sense, and the model building of convolutional neural networks relies on the intricate connections and cellular functions of biological neural networks. For visual tasks, the visual "ventral stream" neural pathway plays an essential role. And inspired by the interactive feedback mechanism between cells and tissues in the ventral stream visual pathway, this paper proposes a novel convolutional neural network model for edge detection. We use Visual Geometry Group Net to simulate the visual signal perceptron retina, the relay lateral geniculate nucleus, and the ventral stream neural pathway, and based on the visual neural information transmission mechanism, we build a lateral interactive feedback coding network. Also by using the bottom-up feedback parsing and parallel processing of feature information in the Inferior Temporal layer, the feedforward information, and feedback information are integrated in a longitudinal feedback interactive manner in the decoding network. We conducted experiments on the publicly available datasets BSDS500, NYUD, Multicue-Boundary, and Multicue- Edge, achieving Score 0.824, 0.773, 0.839, and 0.892, respectively. The results show that our method achieves good performance and is highly competitive.
computer science, artificial intelligence