Image Recognition Algorithm Based on Deep Learning

Image Recognition (IR) is one of the important research topics in the field of computer vision and pattern recognition, and it is also an important tool for the realization of an intelligent society in the future. Deep learning is a process of simulating the human visual system and the cognitive link of the human brain, extracting data features in a structured manner, and finally obtaining the deep information of the image. The whole training process can get a good IR effect without manual participation. The purpose of this article is to research IR algorithms based on deep learning. This paper firstly summarizes the algorithms of IR and deep learning. Aiming at the problems of deep learning objective cost function that is easy to fall into local minimum, slow convergence speed and over-fitting, it proposes the improvement of the algorithm on the deep learning model. At the same time, analyze and design the relevant parameters that affect the model to optimize the model. This paper proposes the weight threshold parameter variable analysis method, the original golden section method, the network structure determination method based on the sensitivity of network nodes, and the pyramid model structure method based on the convolutional network. The experimental results show that the algorithm based on the node sensitivity structure can recognize 1000 images in 3.14s, and the recognition accuracy rate is 98.6%, which is better than other methods.