A Method for 3D Morphological Measurement and Cutting Point Localization of Green Asparagus Based on Improved YOLOv8n-seg and RGB-D Data

In the quiet rows of a vegetable field, a simple yet stubborn problem has long frustrated farmers: how to harvest a crop that grows fast, grows in a crowd, and refuses to stay still. Asparagus is a perennial plant that sends up tender shoots every spring, but these shoots are not uniform. They curve, they twist, and they often hide their roots deep in the soil, making it difficult for a machine to know exactly where to cut. For decades, this has meant that the harvest relies entirely on human hands, a labor-intensive process that is becoming increasingly expensive and difficult to staff. To solve this, scientists have turned to the field of computer vision, a branch of technology that teaches machines to "see" and understand the world through cameras. The goal is to build a robot that can look at a patch of asparagus, identify which stalks are ready to eat, measure their size, and pinpoint the exact spot to sever them from the earth. However, teaching a machine to do this is not as simple as taking a photograph; the plant's irregular shape and the complex lighting of a farm create a visual puzzle that standard software often fails to solve.

A team of researchers from the Jiangxi Academy of Agricultural Sciences and the Chinese Academy of Agricultural Sciences has recently developed a new way to help a robot see green asparagus clearly. They created a specialized computer program, a type of artificial intelligence designed to recognize and outline individual plants in a photograph. The researchers started with an existing, well-known system for finding objects in images but realized it was too heavy and slow for a robot moving through a field. The original system was like a powerful but bulky engine that consumed too much fuel to be practical for a small, battery-powered machine. To fix this, the team redesigned the core of the program, stripping away unnecessary weight while making it sharper at spotting the specific, slender shape of an asparagus stalk. They replaced the standard "brain" of the system with a more efficient structure that could process information faster without losing accuracy. They also added special tools to help the computer understand the long, thin nature of the plant and to ignore the noise of weeds or shadows that might confuse it.

The result of this redesign is a new model called SES-YOLO, which acts as the eyes for a potential harvesting robot. When this model looks at an image of an asparagus field, it does not just find the plant; it draws a precise outline around every single stalk, separating it from the soil and the leaves. This is a crucial step because, unlike fruits that have a clear skin or a distinct shape, asparagus is a long, continuous stem that can bend in any direction. The researchers tested their new model against several other popular systems and found that it was not only faster but also more accurate. In their tests, the new model correctly identified and outlined the asparagus stalks with a precision of 98.6 percent, meaning it rarely missed a plant or confused a weed for a crop. It also ran significantly faster than the older versions, processing images quickly enough to keep up with a moving robot. This speed is vital because a harvesting robot cannot wait for a slow computer to think; it must make decisions in real-time as it moves down the row.

Once the computer has successfully outlined the plant, the next challenge is to measure it and decide where to cut. The researchers combined the visual outline with data from a depth camera, a special type of lens that can tell how far away every part of the image is. By merging the flat picture with this distance information, the system can calculate the true length and thickness of the asparagus stalk, even if it is curved. They found that by breaking the stalk into small segments and measuring each one, they could determine the total length with an average error of less than 2 percent. This level of accuracy is important because farmers need to know the size of the crop to determine if it is ready to be harvested. The system also locates the "cutting point," the specific spot on the stem where the edible part begins and the tough, woody root ends. Since the transition between the tender green top and the white underground root is often hidden just below the soil surface, the researchers programmed the system to estimate this point based on the geometry of the stalk. In their field tests, the system located this cutting point with an average error of only 0.34 centimeters, a margin of error so small that a mechanical arm could easily adjust to it.

To prove that this system could work in the real world, the researchers installed their new program onto a small, portable computer designed for robots, known as the Jetson Orin NX. This device is powerful enough to run complex software but small and efficient enough to be carried by a robot in the field. When they tested the system on this hardware, it processed images at a speed of 38.69 frames per second. This means the robot could "see" and make decisions more than 38 times every second, a rate that is fast enough to handle the rapid movements required during a harvest. The researchers noted that this speed was nearly 60 percent faster than the original, unmodified system, demonstrating that their lightweight design successfully balanced the need for speed with the need for accuracy. The study concludes that this approach provides a reliable technical foundation for building automated asparagus harvesters, offering a way to replace the back-breaking labor of manual picking with a machine that can see, measure, and cut with human-like precision.