The general guiding idea of this thesis is to develop the ability to use TensorFlow for practical programming to solve image processing related problems on the basis of mastering the basic knowledge and characteristics of deep learning. The whole paper tries to be simple and profound. Through easy to understand language and detailed program analysis, it introduces the basic usage, high-level model design and corresponding programming of TensorFlow.
This paper emphasizes the combination of theory and practice, focusing on the application of TensorFlow programming to solve image recognition, provides a large number of data sets, and implements deep learning model in the form of code for readers’ reference.
This article is intended for programmers learning artificial neural networks, deep learning TensorFlow programming, and image processing.
This article consists of 22 chapters, covering the installation and use of the Python class library, the basic data structure and use of TensorFlow, and TensorFlow Data set creation and reading, artificial neural network, feedback neural network, theoretical basis of full convolutional neural network, deep learning model creation, model characteristics, algorithm, ResNet, Slim, GAN, etc.
Due to the details of the content is too much, so xiaobian only part of the knowledge point screenshots out of the rough introduction, each section has more detailed content, I hope we can get the true meaning!
Chapter 1 introduces the basic content of deep learning, introduces the application of deep learning to computer vision and its development direction, and introduces the application prospect of using deep learning to solve computer vision problems, aiming to explain that computer vision is the future development direction and inevitable trend of using deep learning and artificial intelligence.
Chapter 2 covers Python installation and the most commonly used class libraries. Python is a very user-friendly language that makes it easy to express formulas and visions in code without having to learn much programming. The Python-specific library threading is not very common, but is used to lay the foundation for subsequent data reads and tensorflow-specific format generation.
Chapter 3 comprehensively introduces the basic classification, algorithm and theoretical basis of machine learning, and introduces different algorithms, such as regression algorithm and decision tree algorithm concrete implementation and application. These are the basic theoretical parts of deep learning, through which the structure and application of deep learning are shown to readers thoroughly and accurately, laying a solid foundation for further mastering the application of deep learning in computer vision.
Chapter 4 focuses on using the Python language. The introduction and implementation of different Python class libraries will help readers strengthen their Python programming ability and learn the corresponding class libraries. These are things that will be used again and again later in this article. At the same time, I use the knowledge I have mastered to learn the visual presentation skills of data. Although this skill is a basic one in data analysis, it plays a very important role.
Chapter 5-6 is an introduction to the use of the OpenCV library. This book focuses on image processing, and the reading, editing and processing of image data is the most important part of this book. OpenCV is a Python library dedicated to image processing, through the basic explanation and advanced introduction to enable readers to master the use of this important library. Learn to image cutting, transformation and translation of the code.
Chapter 5 introduces the basic content of convolution kernel in the form of examples, and uses Python language to realize the function of convolution kernel. Convolution kernel is a very important basic part of this book, and also a very important part of image processing. It is of great help to write corresponding programs to realize the processing, mastery and understanding of convolution checking image: convolutional neural network.
Chapter 7~8 is the basic introduction of TensorFlow. Through an entertaining website, it shows the basic application of TensorFlow to readers, demonstrates the fitting process of neural network classification by means of graphics and images, and understands the content behind it while having fun.
Chapter 9 is the focus of the book and the foundation of neural networks. The feedback algorithm in this chapter is a milestone algorithm to solve the problem of excessive computation in neural networks. The author introduces this algorithm in plain language through detailed and careful explanation, and realizes the most important algorithm content of this neural network for readers through independent coding form. The content of this chapter may seem small, but it is very important.
Chapter 10 provides a detailed description of TensorFlow data input and output. From reading CSV files, to teaching readers to create a dedicated TensorFlow data format TFRecord, which is rarely covered in the current market books. Using TensorFlow framework for programming, data preparation and standardization is a top priority, so this chapter is also a relatively important chapter.
Chapter 11~12 is a basic tutorial of applying convolutional neural network to study on TensorFlow framework. After the preparation and introduction of the previous chapters, it is the most basic skill of deep learning to use basic theory volume integral neural network to identify handwriting, which is also a very important learning foundation. And in the process of programming, the author shows the reader that parameter adjustment plays an important role in model test results. This is not covered in relevant books on the market at present, which is very important.
Chapter 13~14 is the introduction and application of convolutional neural network algorithm. In these two chapters, the author introduces in detail the application of convolutional neural network, especially in the application of image recognition, from simple handwritten numerical recognition to display object recognition. With the help of the data set of the image recognition contest and the convolutional neural network model which won prizes in the contest, the readers can master the variation of the convolutional neural network. The theoretical basis of convolutional neural network is the forward and reverse process of convolution. Generally, the forward process is easy to understand and learn, but the reverse operation is basically not involved, and if there is, it is only the copy and extract of the formula. In Chapter 14, the operation and calculation methods of the reverse process of convolutional neural network are introduced in detail. Through a large number of examples, the reverse operation of convolutional neural network is described in detail for the first time. This is something that is missing from relevant books.
Chapter 15 illustrates the process of image recognition using convolutional neural network through a complete example. The example comes from ImageNet image recognition competition, and the model used is also the model with the highest accuracy in the competition. Through the detailed analysis of each step of the project, the reader is taught how to use convolutional neural network for image recognition.
Chapter 16 introduces the composition of VGGNet, focusing on VGGNet’s network callback and its ability to perform Finetuning afterwards. This chapter will reuse the examples in Chapter 15 to VGG16 implementation, to provide readers with a different perspective and a different model approach to solve the problem.
Chapter 17 provides answers to some of the interview questions given by current deep learning employers. These questions can help employers analyze who is a high quality candidate, and can also help employers refine their technical concepts and knowledge, identify their own positioning, and pave the way for future promotion and salary increase.
Chapter 18 introduces the ResNet model of deep learning network, which uses a large number of residual modules as the basic component of the network. Its main function is to make the network increase with the change of depth, without weight attenuation, gradient attenuation or disappearance and other problems. In addition to the ResNet model, emerging convolutional neural models, including SqueezeNet and Xception, are introduced.
Chapter 19-20 begins the advanced phase of TensorFlow learning, focusing on an API – Slim, which is a tool for defining A lightweight development library for training and evaluating more complex models. The two chapters not only introduce the method of using it, but also make a multi-layer perceptron MLP, a convolutional neural network CNN, and finally use Slim pretraining model to Finetuning.
Chapter 21 introduces full-convolutional neural network image segmentation, first explains the theoretical basis and implementation method of segmentation, and then gives the step-by-step process and programming basis of full-convolutional neural network for image segmentation, and finally gives the actual practice of using VGG16 full-convolutional neural network for image segmentation.
Chapter 22 is about gAN-adversarial generation network. Although the theory in this chapter seems boring, the author uses a “generator” and a “discriminator” together to carry out “confrontation” in a network to reduce the difficulty of reading. Finally, through the use of GAN to generate handwritten numbers of the case so that readers really learn the application of GAN.
In addition, this paper introduces the most popular and best deep learning models for image recognition, which are the hot spots and research focuses of deep learning at present.
Due to the article space limitation, so small make up here to do too much introduction, need this document 👇
Through TensorFlow image processing, we can fully master the deep learning model and its application
- Fully and deeply explain feedback neural network and convolutional neural network theory system
- Combined with the implementation of deep learning cases, master TensorFlow programming methods and skills
- Emphasis on deep learning practical application development ability and problem solving ability