Curved and cylindrical samples are not flat; when illuminated with ordinary lighting, they often end up looking bright on one side and dark on the other, with a bunch of highlights scattered across the surface.

In many on-site debugging sessions, the bottleneck isn't the algorithm itself—it's the very first image. If the target doesn't appear in the image, subsequent adjustments to thresholds, filters, and models will become extremely difficult.

This article won’t delve into complicated formulas; instead, let’s approach it the way we’d look at it on an actual construction site: Where exactly was the original image problematic, how did things stabilize after the lighting change, and what are the limitations of this approach?

Capacitive Character Detection

This case is testing...Capacitive Character DetectionA sample can be simply understood as...Cylindrical capacitor, material/surface properties can be categorized asElectronic components, cylindrical or curved structures.

The most troublesome point on site is:Reflection interferenceThe effective solution is:Diffuse shadowless lighting.

The issue of reflections cannot be solved simply by increasing exposure; the key lies in controlling the proportion of direct reflections that enter the lens.

Diffuse, shadowless light wraps the sample in softer illumination, reducing localized highlights and shadows.

The last thing to look at isn't how beautiful the parameters are, but whether the results have become more stable: After separating the characters from the background, the recognition process becomes much more stable.

Such schemes also have limitations: diffuse illumination may obscure subtle height differences, so it needs to be verified in comparison with dark-field or directional lighting.

Copper Pipe Head Scratch Detection

This case is testing...Copper Pipe Head Scratch DetectionA sample can be simply understood as...Copper pipe, material/surface properties can be categorized asMetal/high-reflection material, cylindrical or curved surface structure.

The most troublesome point on site is:Reflection interferenceThe effective solution is:High angleBrightfield illumination.

The issue of reflections cannot be solved simply by increasing exposure; the key lies in controlling the proportion of direct reflections that enter the lens.

High-angle brightfield illumination can stably illuminate the main surface, but be mindful of specular reflections from highly reflective materials.

The last thing to look at isn't how beautiful the parameters are, but whether the results have become more stable: highlight the scratches and avoid interference from surface textures.

Such solutions also have limitations: they require real-world verification, taking into account the sample material, target size, and camera angle.

Cylinder Dimension Measurement and Inspection

This case is testing...Cylinder Dimension Measurement and InspectionA sample can be simply understood as...Cylinder dimensions, material/surface properties can be categorized asMetal/high-reflection material, cylindrical or curved surface structure.

The most troublesome point on site is:The edge outline is unclear.The effective solution is:Parallel backlight contour imaging, backlightContour lighting.

When the edges are unclear, first consider converting the problem into contour imaging rather than continuing to focus on the surface texture.

Parallel backlighting results in narrower edge transitions, making it more user-friendly for dimension measurement.

The last thing to look at isn't how pretty the parameters are, but whether the results have become more stable: sharp edge transitions with high contrast and no halo effects.

Such solutions also have limitations: backlighting is more suitable for contour-based tasks but less appropriate for tasks that require observing surface textures or color differences.

How do you judge it on the spot?

If you encounter a similar issue, I don’t recommend immediately asking, “What’s the best light source to use?” A more practical way to ask would be:

1. Is the target now separated from the background?

2. After the light is changed, does it enhance only the target, or does it also enhance irrelevant textures?

3. Can this image be consistently reproduced, rather than looking good only on a specific sample?

The value of a lighting setup isn't to make the images look prettier—it's to reduce the amount of guesswork the algorithm has to do.

Why are curved-surface and cylindrical samples prone to uneven illumination? The key isn't simply boosting brightness—it's about creating stable imaging contrasts. It’s all too common on-site to see images that are bright yet exhibit unstable detection. What really needs to be addressed is this: Are the target features clearly highlighted, and have interfering signals been effectively suppressed?