• Computer vision is currently being deployed for a variety of new tasks, as it is rarely used for tasks other than facial recognition.
•Recent scientific advances pave the way for lightweight, reliable AI models that are better suited to field conditions.
• Researchers looking to design more efficient convolutional neural networks are looking for inspiration in the human brain.
By opening new ground in the field of cooking robots, US-based startup Posha has developed a countertop machine that uses computer vision to identify details such as vegetable cuts and the extent of meat. While it is unlikely to launch a revolution, this latest innovation highlights the growing democratization of technology that allows machines to interpret images and videos. It also emphasizes that it is a major goal of current computer vision research. At the moment, many of today's models are not suitable for new applications in robotics and edge computing, and therefore hardware resources are still too demanding. A situation that has added a driving force to develop more effective and reliable solutions for deployment in sectors such as agriculture, health, robotics, and automotive industries.
This method demonstrated significant robustness to corrupted data. This has been a long-standing challenge for computer vision.
Inspired by the human brain
In a recent breakthrough, a team from the Institute of Basic Sciences (IBS), Uhara University, and Max Planck Institute developed a new AI (AI) technology called LP melting. Inspired by human brain function, researchers sought to explore the possibility of mimicking the selective processing of information through visual cortex. CNNS uses filters (or convolutional kernels), small blocks of small blocks that scan images to extract the most prominent visual features. And these blocks are usually square in shape and are of fixed size. Departing from this paradigm, methods developed by researchers allow AI models to adapt the shapes and dimensions of these filters, stretching them horizontally and vertically for various tasks in ways that mimic the way the brain selectively focuses on specific details. Tests demonstrated that the new method not only improves the accuracy of interpretations, but also significantly robustness to corrupted data. This has been a long-standing challenge for the application of these systems.
Agriculture and biodiversity
In agriculture, computer vision is increasingly used in planting projects such as remote sensing and those managed by French startup Morfo, which collects and analyzes data on vegetation coverage and health. Technology can also be deployed to optimize the production of food crops, especially in greenhouses. Greenhouses in particular can deploy solutions that combine computer vision and AI provided by Tel Aviv-based startup Fermata to detect disease and pest presence, significantly reduce grower workload and minimize potential exposure.
However, the wider use of computer vision for biodiversity and agricultural applications is still hampered by the extensive work involved in training models, which requires important resources. To address this issue, researchers at the University of Illinois have developed a machine learning system that demonstrates the ability to train to distinguish between aerial images of flowering and non-flowering grasses with little human intervention. To take on the task of accurately identifying this crop characteristic under various conditions, the system utilized an Esgan (efficiently monitored generative and adversarial network) architecture in which competing models train each other and fundamentally reduce the need for human-resolved data. Approaches that facilitate the task of adapting AI to a specific purpose may be used immediately for other crops and applications in other sectors.
