Convolutional neural networks

Posted on September 2, 2021

Neville Dubash headshot

Convolutional neural networks (ConvNets) are routinely used in image recognition, with state-of-the-art performance for object detection and image classification tasks, and have proven successful in many applications.

A ConvNet takes an input, such as an image, passes it through a series of hidden layers, many of which perform convolution operations with filters, and then outputs a prediction. In contrast to classical image processing techniques that use predefined filters, the ConvNet learns what types of filters are most useful to the task at hand, and at what layer of the network to use them. For example, in early layers of the networks the filters might be simpler, such as edge or texture filters. Deeper in the network the filter might identify more complex structures, such as faces or cars. In this manner, a ConvNet can be trained to take an image and output useful information about its content.

Our droplet detection algorithm used a fast ConvNet for object detection and help locate droplets suspended in a water-in-oil emulsion in under a second.

View original post on LinkedIn