Artificial Intelligence (AI) algorithms

This is because AI algorithms can learn to recognize patterns in the data that humans may not be able to see. AI algorithms can also process images much faster than humans can. AI algorithms can be scaled to work with massive datasets of images. This is critical in weed detection applications, which typically collect a vast number of images at a single site.

Artificial intelligence is the overarching term used to describe the use of machines and programs to make decisions. There are subsets of AI, including  machine learning (ML) and deep learning (DL), that are typically used to make the most complex decisions. These can be used in weed detection to identify target and non-target species within large numbers of images using specific techniques (called Computer Vision), making weed detection using AI much faster. The terminology can become confusing to those who are new to AI and so it is helpful to learn about these via an internet search, or the published literature so that we can understand the terms, uses and applications.

Below are some links that provide some terminology descriptions,  from the basic to the more technical level:

1. Drone Image Processing: How Artificial Intelligence Helps (geonadir.com)
2. Machine Learning Drones: 3 valuable concepts you need to know (geonadir.com)
3. 7 Terms I Found Confusing in Computer Vision for Beginners (Tutorial) (geonadir.com)
4. Deep Learning vs. Machine Learning: A Beginner’s Guide | Coursera
5. A Review of Machine Learning and Deep Learning Applications | IEEE Conference Publication | IEEE Xplore

The most commonly used DL algorithm for processing image data are convolutional neural networks (CNNs). CNNs are inspired by the way that the visual cortex of the human brain works. The basic unit of a CNN is called a convolutional layer. A convolutional layer takes an input image and applies a series of filters to it. The filters are small, square arrays of numbers that are designed to detect specific features in the image. For example, one filter might be designed to detect edges, while another filter might be designed to detect corners. The output of a convolutional layer is a feature map.

A feature map is a two-dimensional array of numbers that represents the activations of the filters in the layer. The feature maps from different layers are then combined to form a representation of the image that can be used for classification or other tasks.

CNNs are a powerful tool for image processing and computer vision tasks. They are able to learn to recognise patterns in images that are difficult for humans to see. CNNs are also able to scale to work with large datasets of images, which is important for many applications. Commonly used CNN frameworks include VGGNet, ResNet, InceptionNet, U-Net and YOLO.

While deep learning models like convolutional neural networks (CNNs) dominate the object detection landscape, other traditional machine learning algorithms can also be used for this purpose. Some examples of non-deep learning models include support vector machines (SVMs), random forests, and Adaboost.

Non-deep learning models are typically less accurate than deep learning models, but they are also less complex and require less training data. The lower computational requirements of non-deep learning models make them more accessible to a wider range of scientists.