Convolutional neural network (CNN): Teaching computers how to see

• Author:
• Author name

  Quantumrun Foresight
• December 1, 2023

Insight summary

Convolutional Neural Networks (CNNs) are pivotal in image classification and computer vision, transforming how machines identify and understand visual data. They mimic human vision, processing images through convolutional, pooling, and fully connected layers for feature extraction and analysis. CNNs have diverse applications, including retail for product recommendations, automotive for safety improvements, healthcare for tumor detection, and facial recognition technology. Their use extends to document analysis, genetics, and analyzing satellite imagery. With their increasing integration into various sectors, CNNs raise ethical concerns, especially regarding facial recognition technology and data privacy, highlighting the need for careful consideration of their deployment.

Convolutional neural network (CNN) context

CNNs are a deep learning model inspired by how humans and animals use their eyes to identify objects. Computers do not have this capability; when they “view” an image, it is translated into digits. Thus, CNNs are distinguished from other neural networks by their advanced capabilities for analyzing image and audio signal data. They are designed to automatically and adaptively learn spatial hierarchies of features, from low- to high-level patterns. CNNs can assist a computer in acquiring “human” eyes and provide it with computer vision, allowing it to absorb all of the pixels and numbers it sees and aid in image recognition and classification. 

ConvNets implement activation functions in a feature map to aid the machine in determining what it sees. This process is enabled by three main layers: the convolutional, the pooling, and the fully connected layers. The first two (convolutional and pooling) perform the data extraction, while the fully connected layer generates output, such as classification. The feature map is transferred from layer to layer until the computer can see the entire picture. CNNs are given as much information as possible to detect different characteristics. By telling computers to look for edges and lines, these machines learn how to rapidly and accurately identify images at rates that are impossible for humans.

Disruptive impact

While CNNs are most commonly used for image recognition and classification tasks, they can also be used for detection and segmentation. For example, in retail, CNNs can visually search to identify and recommend items that complement an existing wardrobe. In automotive, these networks can watch out for changes in road conditions like lane line detection to improve safety. In healthcare, CNNs are used to better identify cancerous tumors by segmenting these damaged cells from the healthy organs around them. Meanwhile, CNNs have improved facial recognition technology, allowing social media platforms to identify people in photos and give tagging recommendations. (However, Facebook has decided to stop this feature in 2021, citing growing ethical concerns and unclear regulatory policies on using this technology). 

Document analysis can also improve with CNNs. They can verify a handwritten work, compare it to a database of handwritten content, interpret the words, and more. They may scan handwritten papers critical for banking and finance or document classification for museums. In genetics, these networks can evaluate cell cultures for disease research by examining pictures and mapping and predictive analytics to assist medical experts in developing potential treatments. Finally, convolutional layers may assist in categorizing satellite images and rapidly identifying what they are, which can help in space exploration.

Applications of convolutional neural network (CNN)

Some applications of convolutional neural network (CNN) may include: 

• Increased use in healthcare diagnoses, including radiology, x-rays, and genetic diseases.


• The use of CNNs to classify streamed images from space shuttles and stations, and moon rovers. Defense agencies can apply CNNs to surveillance satellites and drones for autonomous identification and assessment of security or military threats.


• Improved optical character recognition technology for handwritten texts and image recognition.


• Improved robotic sorting applications in warehouses and recycling facilities.


• Their use in classifying criminals and persons of interest from urban or interior surveillance cameras. However, this method can be subject to biases.


• More companies being questioned about their use of facial recognition technology, including how they are collecting and using the data.

Questions to comment on

• How else do you think CNNs can improve computer vision and how we use it daily?


• What are the other possible benefits of better image recognition and classification?