What is a feature map?

In a convolutional neural network (CNN), feature maps are the output of the convolutional layerA convolutional layer is an component of CNN's that contains filters that are used to extract important features from input data. after the convolution operation has occurred on an image passed to the convolutional layer.

How are feature maps formed?

In the illustration above, we can see how feature maps are formed from a provided input image.

The image to be sent to the convolutional layer is a multi-dimensional array containing pixel values. The convolutional layer includes filters that are matrices that will be multiplied with the image, and the result will be a feature map(s).

If we have multiple filters, each filter is individually multiplied with the input image and returns a different feature map for each filter.

How to tune feature maps

To tune feature maps, we can use the three ways listed below.

1. Stride: This refers to the number of jumps we want the filter to move after each multiplication operation on the input image. A stride of N means N pixel jumps. A large stride value reduces the size of the feature map.

2. Depth: It refers to the number of filters we want to use for the convolution operation. Each filter gives an individual feature map. N filters will give N feature maps.

3. Padding: We can also provide padding to the original input image by appending additional pixels to the borders of the input image.

In the diagram below, we can see how these three methods work collectively to form feature maps.

Uses of feature maps

Feature maps are used in convolutional neural networks for several reasons, some of which are mentioned below.

• Detect important features: Initially, the feature maps capture low-level patterns, but as they propagate to successive layers, they detect new patterns and combine them to form high-level features.

• Feature sharing: A feature map is passed through multiple layers in a neural network for image processing; hence, the features found by the previous layers are propagated to each successive layer.

• Object recognition: Feature maps can also be passed to an artificial neural network which can be trained to predict the object in the image.

• Image segmentation: Feature maps can divide an image into different segments, each representing a meaningful part of the unsegmented image.

Conclusion

Feature maps are a core component of convolutional neural networks as they capture essential features from the input image to aid the network in decision-making. Moreover, they allow CNNs to process large datasets faster by discarding unwanted data and retaining important information.