Machine vision image recognition utilizes robots to analyze and process images to achieve various goals. Most industrial quality problem research requires one-to-one analysis; here we can only offer a brief overview of automated quality inspection. Solving practical problems requires specific analysis and selection of appropriate recognition methods.
Automated quality inspection can effectively reduce the cost of quality inspection in modern industrial automated production, while increasing the quality and speed of inspection. Compared with human vision, machine vision has obvious advantages.
1) High accuracy: Machine vision-based automated quality inspection can significantly improve grayscale levels while also being able to observe targets at the micrometer level;
2) High speed: The machine can perform quality inspection at a specified frequency, even down to the microsecond level;
3) High stability: The quality of manual quality inspection largely depends on the inspector's professional skills and individual work status, avoiding missed or incorrect inspections.
4) Information storage: The results of machine vision inspection, whether product status or inspection result description, can be easily and automatically saved and archived, preparing data for the next step of analysis.
Common appearance quality issues include:
Product surface defects: scratches, marks, roller marks, pits, roughness, ripples
Inclusions, damage, and stains on the surface of non-metallic products
Assembly quality inspection: Inspection of part assembly integrity, inspection of assembly dimensional accuracy, and measurement of position/angle.
Print inspection: Incomplete printing, skewed printing, unclear edges
These problems can be summarized as follows:
Determine if the target has unplanned edge features, such as additional textures or indentations.
Determine planned feature matches, such as Brute-Force and FLANN.
Most feature detection algorithms involve the identification of corners, edges, and blobs in an image. The most commonly used feature detection and extraction algorithms include:
For corner detection: Harris, FAST
Used for spot detection: SIFT (Blob), SURF, BRIEF
ORB: Represents the directional FAST algorithm and the rotation-invariant BRIEF algorithm.
Matches: Brute-Force, FLANN
Visual positioning in industrial production has been applied in many industries, such as semiconductor packaging and industrial manufacturing.
I. Semiconductor Packaging Field
In the semiconductor field, equipment needs to retrieve the pick-up head based on the chip position information obtained by machine vision, and accurately pick up the chip for bonding. This is a widespread application of vision positioning in the semiconductor packaging field.
II. Robotics Industrial Manufacturing Field
With the development of robot vision positioning technology, the accuracy and stability of robots in operation are improved, which greatly increases the efficiency of enterprises in production, saves labor costs, and enhances market competitiveness.
When discussing visual applications, some might ask about the difference between image processing and video processing. Essentially, video processing is also image processing—that is, processing frames within a video—but it includes an additional video decoding process. Furthermore, video processing, besides processing images within a single frame, may also require analyzing the relationships between frames.
Below, we'll use a simple demo to help readers understand the general implementation steps:
Open the camera and read the frames.
Preprocess the frame image (grayscale conversion, filtering, binarization. In actual quality inspection, image correction and ecological processing may also be required to achieve better processing results).
Defect identification of frame images
Save images that meet the requirements.
In practical applications, the industrial control system is typically invoked based on the quality inspection results to perform corresponding product diversion, sending unqualified products to the re-inspection area and qualified products to the next process.
While automated quality inspection offers many advantages, machine vision still faces numerous challenges in completely replacing manual visual inspection. For instance, the reflection and refraction of different materials can affect the extraction of features from the object being inspected. Therefore, the brightness and angle of the light source, as well as the quality of image acquisition, directly impact the inspection results. For example, in detecting scratches on glass and reflective surfaces, the problem often lies in the integrated imaging of different defects.
In actual production, we often use machine vision to identify specific defect patterns given in the program, determining whether they have occurred and the degree of quality deviation compared to standards. However, some defects may not have been encountered in our previous experience, leading to missed detections. With manual quality inspection, inspectors would generally recognize the new defects and make further quality checks. Of course, these problems will be optimized and resolved with technological advancements.
