Detection and Tracking Method for Walnuts in a Walnut Sorting Device Based on Improved YOLOv8n+DeepSORT

In the world of agriculture, sorting nuts is a task that demands both speed and a keen eye. For centuries, farmers and processors have relied on human hands or simple mechanical sieves to separate good walnuts from bad. These traditional methods work, but they are slow and often miss subtle flaws hidden deep inside the shell. To see what lies beneath the hard exterior without cracking it open, engineers turn to X-ray imaging. This technology, familiar to anyone who has visited a hospital, uses invisible radiation to pass through solid objects, revealing their internal density and structure on a screen. When combined with modern computer vision, these X-ray images can be analyzed by software that learns to recognize patterns, distinguishing a plump, healthy kernel from a shriveled or empty one. The challenge, however, lies in doing this quickly enough for a moving conveyor belt, where walnuts stream past in a continuous, chaotic flow. The computer must not only identify each nut but also track its movement from one moment to the next, ensuring it is sorted correctly even as it jostles against its neighbors.

Researchers at Xinjiang University have tackled this specific problem by developing a new system designed to watch and sort walnuts as they travel through an industrial sorting machine. Their goal was to create a digital eye that is both sharp enough to see fine internal details and fast enough to keep up with the pace of a factory line. They started with a powerful existing tool for finding objects in images, a type of artificial intelligence known as a neural network, and then refined it specifically for the unique challenges of walnut X-rays. In their experiments, they fed the system thousands of X-ray images of walnuts, ranging from perfectly full to completely empty, teaching the software to recognize the subtle differences in how the radiation passes through the shell and the kernel inside.

The core of their solution involves two working parts that function in tandem. The first part is a detector that scans the X-ray images to find every walnut and determine its quality. The second part is a tracker that follows each walnut as it moves along the belt, assigning it a unique identity so the machine knows which bin to drop it into. The researchers found that the standard versions of these tools were not quite good enough. The X-ray images of walnuts often have low contrast, making the internal details look blurry, and the standard software sometimes lost track of a nut when it moved too quickly or got too close to another one. To fix this, the team made three specific improvements to the detection software. They added a new attention mechanism that helps the computer focus on the most important, scattered details of the walnut's interior, ignoring the distracting background. They replaced a standard step that enlarges the image with a more dynamic method that preserves the smooth edges and texture of the nut, preventing the image from becoming blocky or distorted. Finally, they streamlined the part of the system that combines different layers of information, removing unnecessary calculations to make the whole process faster without losing accuracy.

For the tracking component, the researchers changed how the software decides that two images of a walnut in different frames belong to the same object. Instead of relying solely on how much the shapes overlap, they introduced a method that also considers the distance between the centers of the shapes. This small adjustment proved vital when walnuts were crowded together or partially hidden, as it helped the system maintain a steady lock on each individual nut, preventing it from accidentally swapping identities or losing track entirely. When they tested their improved system against the original versions, the results were clear. The new detector identified walnuts with significantly higher accuracy, catching more of the subtle defects and fewer false alarms. It processed images at a speed of 412 frames per second, which is fast enough to handle the rapid flow of a real production line. The tracking system also improved, reducing errors where the computer might lose a target or confuse one nut for another.

The study demonstrates that by carefully tuning these digital tools to the specific physics of X-ray imaging and the behavior of walnuts on a conveyor belt, it is possible to automate a task that has long relied on human judgment. The improved system does not just find the nuts; it understands their movement and their internal quality with a level of precision that traditional machines cannot match. This work suggests a path forward for the nut processing industry, where speed and quality are equally critical. By proving that these advanced algorithms can be adapted to handle the messy, real-world conditions of a factory floor, the researchers have provided a blueprint for more efficient, reliable, and automated sorting of walnuts and potentially other similar crops. The findings indicate that the future of agricultural sorting lies not in faster machines alone, but in smarter software that can see what the human eye might miss and remember what the human hand might forget.