Research on Image Recognition Method Based on Improved Neural Network
Yang Shen · 2022 4th International Conference on Communications, Information System and Computer Engineering (CISCE) · 2022
In the 1960s, Hubel et al. proposed the concept of receptive field through the study of cat visual cortex cells[1]. In the 1980s, Fukushima [2] proposed the concept of neurocognitive machine based on the concept of receptive field, which can be regarded as the first implementation of convolutional neural network (CNN). The neurocognitive machine will decompose a visual pattern into many sub-patterns (features), and then enter the hierarchical hierarchical connected feature plane for processing. It tries to model the visual system so that it can complete recognition even when the object has displacement or slight deformation. CNN is a variant of multi-layer perceptron (MLP), developed from biologists Huber and Wessel's [3] early research on the visual cortex of cats. There is a complex structure in the cells of the visual cortex, and these cells are very sensitive to the sub-regions of the visual input space, called the receptive field.[4] CNN was proposed by Yann Lecun of New York University in 1998 [5]. In essence, CNN is widely used neural network architecture. The success lies in the local connection and weight sharing adopted by CNN. On the other hand, the complexity of the model is reduced, that is, the risk of overfitting is reduced. In this paper, using CNN as the research method, image recognition is studied by illustrating a set of comparative experiments. In the comparative experiment, I will optimize the network model of previous experiments, and finally effectively and significantly improve the efficiency of model training.