Intoduction to Convolutional Neural Network
CNN stands for Convolution neural network. CNN was initially developed within the neural network image processing community. A CNN involves basically two operations named as convolution and pooling as feature extractors. The output of this sequence of operations are connected to completely connected layer same as Multi- layer perceptron. There are basically two quite pooling used like max-pooling and average-pooling. Max- pooling selects the utmost number of values in input feature map region of every step and average-pooling selects the typical number of values in the region. CNN works on both structured and unstructured data.
When CNN is applied in any image dataset then following steps have been followed:
? Consider the image as a two dimesional matrix which conatins features of image say intensity values of pixels involved in image.
? Apply convolutional layer in which suitable filters are used to extract the useful features of image.
? Pooling layer is used to reduce dimensionality of the feature map.In pooling layer a filter is choosen and apply on the image, Select max values of pooling window in case of max pooling and select minimum values of the pooling window in case of min pooling. To shift the pooling window strides are used. If stride is 2 that mean each time pooling window has to shift 2 steps.
? After that Flatten Layer is used to convert two dimensional data into one dimensional array.
? Now the one dimensional input is given to the fully connected dense layers in which ‘Relu’ activation function is used.
? In the output layer we can use any suitable activation function like ‘sigmoid’, ‘softmax’ etc. Acording to our dataset.
? Finally we get the output from the output of fully connected dense layers.
