Design of Convolutional Neural Network Based on Reticulated Convolution Module

Convolutional neural networks use convolutional layers for feature extraction. Due to the limited feature extraction capabilities of traditional convolutional layers, the performance of convolutional neural network models is constrained. This paper presents an easy-to-port convolution module called Reticulated Convolution Module, which is dedicated to improving the ability of convolutional neural networks to extract key features. It adopts a funnel-like mesh structure, which first refines the features and then performs grouping, combination and fusion operations. We validate our Reticulated Convolution Module through extensive experiments on different types of datasets such as 101_food, Caltech-256 and GTSRB. The experimental results show that the convolutional neural network designed by Reticulated Convolution Module has excellent performance.