When it comes time to plan the Thanksgiving feast, some cooks turn to timeworn family recipes. More adventurous chefs, however, take the holiday as an opportunity to try something new, searching for inspiration online or in the pages of cookbooks.
Whether you’re handling the cooking or just enjoying the meal this holiday season, we’ve compiled some festive applications for computer vision that you’ll be thankful for.
Deep Learning vs. The New York Times
Four New York Times food columnists recently hit the kitchen to determine if deep learning solutions could help plan a tasty Thanksgiving meal. For their part in the experiment, Priya Krishna prompted GPT-3, a neural network trained on texts including cookbooks, with notes and suggestions based on her food preferences.
The results of GPT-3’s Thanksgiving menu was strange—to say the least. The solution advised Krishna to cook a 12-pound turkey without additional oil and cooked up a naan stuffing with dozens of ingredients. In general, GPT-3’s recipes were light on specifics and even the best-tasting dishes were lacking.
Though it’s clear you won’t see AI-powered menu plans en masse any time soon, various solutions play a role in helping make Thanksgiving festivities possible. Enterprises including retailers, farmers, manufacturers, and restaurants all use vision AI every day to unlock the value of visual data.
AI-Enhanced Crop Management for Cranberries, Potatoes, Pumpkins, and More
Cranberry sauce is perhaps the most polarizing of Thanksgiving staples, but that’s not stopping researchers from leveraging deep learning in cranberry bogs to eliminate environmental risks and optimize yields. The latest volume of Computers and Electronics in Agriculture includes a report from Rutgers University researchers who’ve developed methods for counting cranberries, identifying fruit that’s exposed to excessive sunlight, and for predicting the internal temperature of individual berries. Trained with the help of both drone- and ground-based imagery, their solution offers quick and accurate results to help farmers avoid crop loss and optimize yields. They hope to contribute to intelligent irrigation systems for additional crops in the future.
Whatever the crop, computer vision models have the potential to enhance agricultural techniques and contribute to better harvests. Across the globe, visual data analytics solutions help to automate hazard detection, inventory plant populations, assist in genetic selection, and more. Below, you’ll see an example of a ripeness detection model for analyzing shape, texture, and color to ensure fruit is harvested at the peak of its ripeness.
AI’s role continues long after crops leave the farm too. Once products have reached store shelves, computer vision supports retailers in automating planogram compliance, spotting hazards throughout the store, and generating insights based on customer behavior.
More Than Turkey: Precision Livestock Counting and Health Monitoring
Manual livestock inventory counts leave agribusinesses vulnerable to costly human errors everyday. What’s more, even experienced livestock production and intake professionals are often thwarted by environmental conditions. As the weather gets colder, they’re often forced to contend with steam and condensation in addition to visible snow and ice. This makes it all the more challenging to distinguish the difference between nearly identical turkeys, pigs, steer, or other animals moving rapidly in close quarters. Plainsight has worked with one of the world’s largest food processors to address these challenges and introduce models for counting animals with near-total accuracy.
Beyond counting, models for livestock health monitoring offer greater visibility into each animal’s well being. Gait analysis models, for example, can spot subtle signs of trouble in the way a turkey trots. A 2021 study published in Frontiers in Animal Science details efforts to improve poultry pose estimation efforts with deep learning and automated gait analysis. Manual gait scoring is a standard practice in poultry production, but data collection and annotation typically proves time consuming. Leveraging deep learning and computer vision should help enterprises overcome persistent challenges and potentially address the issue of limited data availability.
Computer vision models for detection, tracking, counting, and monitoring have applications for all types of livestock – from huge steers to tiny honeybees. Applied in apiaries, for example, computer vision models can support nearly the entire global supply chain by helping beekeepers recognize subtle changes in hive behavior and even spot signs of colony collapse. Whether or not honey makes it way into any Thanksgiving dishes, you can thank bees for their role in fertilizing holiday staples like apples, pumpkins, squash, and green beans.