TensorFlow-based deep learning image classification and application deployment method

Abstract

The invention provides a TensorFlow-based deep learning image classification and application deployment method, which comprises the following steps: 1) constructing a Tensorflow machine learning development environment; 2) data acquisition and conversion: obtaining a lot of image data having tags or having no tags form an internet through a distributed crawler system, and carrying out preprocessing on the image data; 3) model establishing and training: establishing a classification model, carrying out training, test and verification on the classification model according to the obtained image data, and keeping the trained classification model; 4) exporting the trained classification model; 5) defining a server interface, and creating a classification server for the exported classification model; 6) creating a request service and establishing connection for a mobile terminal and the classification server, creating a Web APP, the mobile terminal uploading the image data through the Web APP, and calling the classification model to realize automatic image classification; and 7) product deployment, and applying the classification server to products.

Description

technical field [0001] The invention belongs to the field of deep learning applications, and in particular relates to a TensorFlow-based deep learning image classification and application deployment method. Background technique [0002] In recent years, the sudden emergence of machine learning represented by deep learning has become a hotspot in the field of artificial intelligence research. However, the existing deep learning is mostly used for academic research, and it faces huge challenges in the field of large-scale product development and commercial applications. [0003] First, the deep neural network model is complex, with a large amount of training data and a large amount of calculation. On the one hand, the deep neural network DNN (Deep Neural Networks) used in deep learning needs to simulate the computing power of the human brain, which requires a large number of neurons in the DNN, and requires each neuron to include mathematical calculations (such as Sigmoid, Re...

AI Technical Summary

Problems solved by technology

DNN has a large number of parameters and a complex model. In order to avoid overfitting, a large amount of training data is required.

The superposition of two factors leads to an astonishingly time-consuming training of a model

[0004] Secondly, the deep neural network needs to support large models and the convergence of deep neural network training is difficult, requiring repeated experiments

[0005] Therefore, deep learning has become a direction with good results but extremely high threshold. How to achieve practical application effects has become a bottleneck problem restricting the extensive and in-depth application of deep learning.

Images