Pre-IdentifyNet: An Improved Neural Network for Image Recognition

Abstract

With the rise and development of artificial intelligence, image recognition and classification technology has received more and more attention as an important branch of its research field. Among them, the introduction of deep learning networks and the construction of neural network structures not only avoid a lot of the tedious work of manual extraction, but also improve the accuracy of image recognition. Convolutional neural networks have many advantages that conventional neural networks do not have. Therefore, image classification systems based on convolutional neural networks emerge in endlessly, but there is still much room for improvement in terms of recognition accuracy and recognition speed. Based on this, this paper proposes an improved deep convolutional neural network to improve the accuracy of the network by changing a series of parameters such as the number of channels of the convolution layer, the size of the convolution kernel, the learning rate, the number of iterations, and the size of the small batch with speed. In this paper, three data sets were selected, namely sewage, animals and the Simpson Family. Comparing the improved convolutional neural network network with the existing SqueezeNet and GoogleNet. It is found that the accuracy of the network is maintained while maintaining a similar speed. Both F1-score and F1-score have been improved with a higher recognition rate and better recognition effect in image recognition classification.