Automated Dimensional Verification of Textile Cuts Using Improved U-Net and ResUnet Models

In the world of making clothes, the difference between a garment that fits perfectly and one that feels awkward often comes down to a few millimeters. When a factory cuts fabric to create a shirt, a sleeve, or a collar, those pieces must match a precise blueprint. If a piece is even slightly too large or too small, the problem does not stay isolated; it travels through the assembly line, causing the final product to look wrong, shrink unevenly, or require expensive rework. For decades, checking these dimensions has been a slow, manual task. Skilled workers have had to lay out each piece of fabric, flatten it by hand, and measure it with rulers or templates. This process is not only time-consuming but also prone to human error, as the soft, stretchy nature of fabric makes it difficult to hold perfectly still, and different workers might measure the same piece slightly differently.

To solve this, researchers have turned to computer vision, a field where cameras and software work together to "see" and measure objects. However, measuring a flat piece of cloth is much harder than measuring a rigid metal part. Fabric wrinkles, shifts, and deforms, confusing standard measurement tools that expect straight lines and fixed shapes. The challenge is to build a system that can look at a photograph of a cut piece of fabric, trace its exact outline, find specific points on that outline, and calculate its real-world size with millimeter precision, all without a human needing to touch it.

A team of researchers from the Peruvian University of Applied Sciences has developed a new automated system designed to tackle this exact problem. They created a specialized setup to photograph textile cuts and then used advanced artificial intelligence to analyze the images. Instead of relying on a single, rigid method, their system adapts to the specific shape of the fabric piece it is examining. For some parts of a shirt, like a small collar, one type of analysis works best; for larger sections like the back panel, a different approach yields better results. The researchers tested their system on four fundamental parts of a T-shirt: the front panel, the back panel, the collar, and the sleeves. They found that by combining different mathematical strategies within their software, they could achieve a level of accuracy that rivals human experts but with the speed and consistency of a machine.

The heart of their solution is a custom-built camera chamber. Imagine a small, light-proof box where the lighting is perfectly even and the camera is locked in place, so it never moves or changes its focus. Inside this box, a worker places a piece of cut fabric on a flat, white surface. The camera takes a high-resolution picture, and the system immediately begins its work. The software first cleans up the image and then uses two different types of deep learning models, which are essentially highly trained digital brains, to trace the outline of the fabric. These models are not just looking for edges; they are trained to understand the complex curves and soft boundaries of cloth. To make sure these digital brains learn correctly, the researchers taught them using a special combination of three different learning rules. One rule helps the model understand the overall shape, another focuses on the precise edges, and a third helps it pay extra attention to tricky areas where the fabric might blend into the background. By mixing these rules together, the system learns to draw the outline of the fabric with extreme care, avoiding the jagged or fuzzy edges that often plague automated systems.

Once the software has traced the outline, it identifies key points along that edge, such as the corners of a collar or the ends of a sleeve. This is a critical step because the system needs to measure the distance between these specific points to determine if the piece is the right size. The researchers found that the best way to find these points depended on the shape of the fabric. For smaller, more complex pieces like collars, a specific configuration of their software worked best. For larger, simpler pieces like the front and back of the shirt, a different configuration performed more reliably. The system is smart enough to choose the right tool for the job, automatically selecting the best setting based on what it sees in the image. After identifying the points, the system converts the distance between them from pixels on a screen into real-world millimeters. They calibrated this conversion using a known reference length, ensuring that a measurement of 100 pixels always equals the same physical distance, regardless of the fabric's texture or slight variations in lighting.

The results of this study show that the system works with remarkable precision. When tested on the different parts of the T-shirt, the software achieved an accuracy where the measurements were off by only about one millimeter. This level of precision is significant because it meets the strict requirements of industrial manufacturing, where even small errors can ruin a batch of clothing. The system performed better than older, fixed methods that used the same settings for every type of fabric. By adapting its approach to the specific geometry of each piece, the new system reduced errors and provided consistent results across all the different components tested. The researchers also noted that the system could generate a final report for each piece, telling the factory worker exactly which dimensions were correct and which ones fell outside the acceptable range. If a piece was too small to be fixed, the system flagged it for rejection; if it was slightly off but could be saved, it provided that information as well.

This work demonstrates that automated quality control for textiles is not only possible but can be highly effective when the technology is tailored to the unique challenges of fabric. The researchers did not just build a faster ruler; they built a flexible system that understands the nuances of cloth. While the study focused on a single model of T-shirt, the approach suggests a path forward for the entire industry. By moving away from manual measurement and toward adaptive, computer-based verification, factories can reduce waste, speed up production, and ensure that every garment that leaves the line meets the same high standard of fit and quality. The system stands as a practical, scalable solution that bridges the gap between the soft, unpredictable nature of fabric and the rigid demands of modern manufacturing.