CNN

Grayscale

An image of 4 x 4 x 1 array of matrix

RGB (3 CHANNELS)

An image of 6 x 6 x 3 array of matrix

CNN

Each input image will pass it through a series of convolution layers with filters (Kernals), Pooling, fully connected layers (FC) and apply Softmax function to classify an object with probabilistic values between 0 and 1.

Convolutional neural networks (CNNs)

In machine learning, a convolutional neural network (CNN, or ConvNet) is a class of deep, feed-forward artificial neural network that have successfully been applied to analyzing visual imagery. It's the first layer to extract features from an input img.

Spatial pooling types

Max pooling, average pooling, sum pooling

Pooling

Pooling layers section would reduce number of parameters wen img is too large.

ReLU

Rectified Linear Unit for a non-linear operation. The output is ƒ(x) = max (0, x). Purpose is to introduce non-linearity in our ConvNet.

Spatial pooling (subsampling or downsampling)

Reduces the dimensionality of each map but retains important info.

Sum pooling

Sum of all elements in the feature map.

CNN image classifications

Take an input image, process it and classify it under certain categories. Computer sees an input image as array of pixels and it depends on the image resolution. Based on the image resolution, it will see h x w x d( h = Height, w = Width, d = Dimension ).

Max pooling

Takes the largest element from the rectified feature map.

ReLU is better than tanh and sigmoid

There are other non-linear functions such as tanh or sigmoid can also be used instead of ReLU. Most of the data scientists use ReLU since performance wise ReLU is better than other two.

Neural network with many convolutional layers

a)input b) feature learning: - convo + relu - poopling - convo + relu - pooling c) classification - flatten - fully connected -softmax

Classifying img

convolutional and pooling layers output high-level features of input. Fully connected layer uses these features for classifying input image. The output is express as probability of image belonging to a particular class. The network also can be trained in backpropagation method. The weight is learned for convolutional filters and fully connected layers. The error also can be calculated by using cross- entropy loss.