keras 卷积神经网络_使用keras的卷积神经网络初学者指南

本文是针对初学者的keras卷积神经网络指南,介绍了如何利用keras搭建和应用卷积神经网络,涉及深度学习的基础知识。
摘要由CSDN通过智能技术生成

keras 卷积神经网络

Deep Learning is becoming a very popular subset of machine learning due to his high performance results achieved on a wide range of datasets types. One known way to use this ML technique in particular, is to classify images and photos. The algorithm is known as Convolutional Neural Network (CNN) and is available on the Keras library in Python with all its simplicity for building neural networks.

深度学习由于在各种数据集类型上获得的高性能结果而成为机器学习中非常受欢迎的子集。 特别地,使用这种ML技术的一种已知方法是对图像和照片进行分类。 该算法被称为卷积神经网络(CNN),可在Python的Keras库中使用,因为它非常容易构建神经网络。

Image for post
Figure 2: Example of computer vision tasks ( 图2:计算机视觉任务示例( Source: 资料来源Fei-Fei Li, Andrej Karpathy & Justin Johnson (2016)) 李飞飞,安德烈·卡帕蒂和贾斯汀·约翰逊(2016) )

The computer uses pixels to view images. The pixels in an image are usually related. For example, a certain set of pixels can represent edges in an image or some other pattern. Convolution Neural Networks can be used to help identify patterns, objects and even people in images and photos.

计算机使用像素查看图像。 图像中的像素通常是相关的。 例如,一组特定的像素可以代表图像或其他图案中的边缘。 卷积神经网络可用于帮助识别图像,照片中的图案,物体甚至人物。

Convolution multiplies the pixel matrix with the filter matrix or “kernel” and then adds the multiplied values. Then, the process of convolution slides to the next pixel and repeats the same process until all image pixels are covered. This process is shown further bellow.

卷积将像素矩阵与滤波器矩阵或“内核”相乘,然后将相乘后的值相加。 然后,卷积过程会滑到下一个像素,并重复相同的过程,直到所有图像像素都被覆盖为止。 进一步显示此过程。

The purpose of computer vision is to enable machines, computers or programs to observe the world like humans and apply knowledge to specific tasks, such as image and video recognition and image classification. With the development of deep learning algorithms (more specifically, convolutional neural networks), this has become possible nowadays. And affordable to everyone.

计算机视觉的目的是使机器,计算机或程序能够像人类一样观察世界,并将知识应用于特定任务,例如图像和视频识别以及图像分类。 随着深度学习算法(更具体地讲,卷积神经网络)的发展,如今这已成为可能。 而且大家都负担得起。

The architecture of the Convolutional networks are the connectivity pattern of neurons in the animal brain, and are inspired by biological processes, the connectivity pattern between neurons resembles the organization of the animal visual cortex.

卷积网络的体系结构是动物大脑中神经元的连接模式,并受生物过程的启发,神经元之间的连接模式类似于动物 视觉皮层 的组织

Convolutional neural networks have the ability to capture spatial and temporal correlations in input images by applying correlation filters. Due to the reduced number of parameters involved and the reusability of weights, it performs better on image data sets.

卷积神经网络具有通过应用相关性过滤器捕获输入图像中的空间和时间相关性的能力。 由于所涉及参数的数量减少以及权重的可重用性,它在图像数据集上的性能更好。

1.如何运作? (1. How does it work?)

Convolutional neural networks (CNNs) do not view images like we humans. We treat the image as a flat canvas with colors. We may not care about the width and height of the image, but we can perceive these parameters.

卷积神经网络(CNN)不能像人类一样查看图像。 我们将图像视为带有颜色的平坦画布。 我们可能不在乎图像的宽度和高度,但是我们可以感知这些参数。

Image for post
Figure 3: RGB color composition
图3:RGB颜色组成

However, CNNs will view the image into different three-dimensional objects, such as three-dimensional objects where the three-dimensions are the color coding (or color channel), mainly red, green and blue (RGB), when combined, produce the colors we see.

但是,CNN会将图像查看到不同的三维对象中,例如三维对象(其中三维是颜色编码(或颜色通道),主要是红色,绿色和蓝色(RGB))在组合时会产生图像。我们看到的颜色。

  • 0
    点赞
  • 1
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值