PickAMoo: LIDAR-enhanced mask R-CNN segmentation for precision weight estimation in dairy cattle using smartphone imaging

Why weigh cows without touching them

For dairy farmers, knowing how much each cow weighs is vital for setting feed rations, monitoring health, and making breeding decisions. Yet putting cows on a scale takes time, equipment, and often some stressful pushing and prodding. This study explores whether an ordinary smartphone, helped by its depth sensor and modern image analysis, can estimate a cow’s weight from a photo taken in the barn, turning a routine chore into a quick, contact free snapshot.

The problem with tapes and heavy equipment

On many farms, workers still judge weight by eye, by wrapping a measuring tape around the chest, or by guiding animals over fixed scales. These methods can work, but they demand skill, time, and close handling of large animals. They can also be inaccurate for certain breeds or very heavy cows, because body shape varies. Recent research has shown that cameras and computer vision can read body size and shape from images, but most existing systems rely on fixed 3D cameras and carefully controlled walkways that are expensive and hard to install in ordinary barns.

Turning a smartphone into a barn side tool

The researchers designed a workflow built around a handheld smartphone that many farmers already own. They focused on iPhone models with a built in depth sensor (LiDAR), which helps the phone judge how far away the cow is. A custom app called PickAMoo guides the user to stand at the right distance and angle so that the entire cow is in view. During field tests on two Swedish research farms, the team collected nearly 3,000 images from over 500 cows, along with true weights from an automatic scale and a set of traditional tape measurements for comparison.

Teaching the phone to see the cow’s shape

The next challenge was teaching the computer to pick out the cow’s outline from each busy barn scene. The team manually drew precise borders around cows in more than 500 photos and used these as examples to train a modern image segmentation model known as Mask R CNN. Among several tested versions, one based on a ResNet 101 backbone did best, correctly separating the cow from the background in almost all cases. From these outlines, the system calculated simple shape features such as height, length, and projected body area, then adjusted them using the depth reading so that cows photographed at slightly different distances could still be compared fairly.

From outline to weight band

Instead of predicting exact weight down to the kilogram, the researchers grouped cows into nine weight bands, similar to how a farmer might think in practical ranges. They trained a machine learning model to link the scale corrected shape features to these bands, while carefully avoiding shortcuts that could leak information from the test data back into training. When checked on a separate group of cows not used to build the model, the smartphone based system placed animals into the correct weight band about 90 to 96 percent of the time. Most mistakes were only one band off, meaning the estimate was close to the true weight even when not perfect. Compared with standard tape based formulas, the new method avoided some of the breed specific errors seen in heavier Swedish Red cows.

What this means for everyday farming

The study shows that, under realistic barn conditions, a phone in a worker’s hand can provide useful, non contact weight estimates for dairy cows by combining depth sensing, smart image segmentation, and a tailored learning algorithm. While it is not a replacement for a calibrated scale when precise weights are needed, it could support frequent, low effort checks that help adjust feed, track weight changes over time, and flag animals that may need closer attention. With further testing on more breeds and farm types, and streamlining of the software to run fully on device, such tools could make accurate cattle weighing more accessible and less stressful for both cows and people.

Citation: Guzhva, O., Ternman, E., Lindberg, M. et al. PickAMoo: LIDAR-enhanced mask R-CNN segmentation for precision weight estimation in dairy cattle using smartphone imaging. Sci Rep 16, 16020 (2026). https://doi.org/10.1038/s41598-026-54742-3

Keywords: dairy cattle, smartphone imaging, LiDAR, computer vision, weight estimation