大量的标注图片为AlexNet的出现打下坚实基础,为了完成大量样本下上千个种类图像的判断,需要一种学习能力更强的网络。但使用现有的全连接神经网络处理如此大的数据量会导致计算量大的离谱,而卷积神经网络(CNN)比全连接神经网络(FC)理论效果略差,参数少、连接少、容易训练。

关键词:

  • Dropout
  • Rectified Linear Units(ReLU激活函数)
  • Local Response Normalization(LRN局部响应归一化)
  • Overlapping Pooling(重叠池化)
  • PCA(主成份分析算法)

笔记:

ImageNet Classification with Deep Convolutional P1.png
ImageNet Classification with Deep Convolutional P2.png
ImageNet Classification with Deep Convolutional P3.png
ImageNet Classification with Deep Convolutional P4.png
ImageNet Classification with Deep Convolutional P5.png
ImageNet Classification with Deep Convolutional P6.png
ImageNet Classification with Deep Convolutional P7.png
ImageNet Classification with Deep Convolutional P8.png
ImageNet Classification with Deep Convolutional P9.png

标签: CNN, 神经网络, ImageNet, Convolution

评论已关闭