Image Recognition Based on Convolution Neural Network with Small Data Set

Abstract

With the development of artificial intelligence, deep learning is more and more widely used in image recognition. As a representative deep learning algorithm, convolutional neural network has been widely used due to its excellent performance in image processing. It is widely used in various fields and has made great achievements. Image recognition is the most important problem in the development of computer vision. Since computers can recognize ob jects, computer vision can be used in many fields. How to identify vehicles and people in driverless driving and public safety is very important and meaningful. By identifying cars or pedestrians in front of you on autopilot, computers can predict danger by taking precautions in advance. Therefore, this experiment USES convolutional neural network to classify cat and dog images. Neural network generally needs a large number of data sets, but the number of our data sets often cannot meet our needs, so it is very meaningful to build a better model in the case of small data. A nine-layer artificial convolutional neural network is established and the error iteration is tested by the optimal value and test set of batch gradient descent method and local demand model. Experimental results show that after 30 iterations, the accuracy of training set is close to 100%, while the accuracy of test set is only about 70%. Therefore, this paper adopts the following three methods to optimize the model step by step. First, we use the migration learning method to call the VGG16 model in the keras library to conduct training prediction for this experiment. Data enhancement was then performed to further optimize the model. Finally, the model is optimized and improved by fine-tuning. The experimental results show that the experimental results are good and can meet our experimental requirements.