Photosensing systems have been used extensively for detection, gaging, identification and characterization. They are used in high-speed defect detection and gaging, characterization, pattern recognition and identification. These systems are called on to perform with high reliability and performance in increasingly difficult circumstances.
Many noncontact optical sensing systems are integrated into real-time automated control systems. To improve the reliability of any automated sensing system, it becomes necessary to understand the signals generated, how decisions are made by the system, and what causes reliable or unreliable operation. How can the system be fooled? Manufacturers want the systems to detect only the features or targets of interest, but not other image details.
Reliable vs. Unreliable Systems
Unreliable systems look just like reliable systems. They have illumination and optics, sensors, CCD cameras and photodiodes, and electronics and computers. Illumination, both passive and active, and optics provide the basic optical signal within the system. The sensors collect this signal, the electronics convert the optical signals to electronic signals for the computer to analyze and make decisions. In addition to these basic features, vision systems sometimes have displays to guide human use, collect data for analysis to better understand the process and provide output signals to automatically control electromechanical switches.
The information and decisions provided by the system may not always be correct, leading to errors in decisions. A system in which the signal contains nondefect-related information and results in sensing errors can be termed an unreliable system. Unreliable system performance results when an optical signal contains degraded information from irrelevant variations in the target, background or optical system.
Reliable optical sensing system performance results when the optical sensing system responds only to the physical features it is intended to sense and ignores other features or changes.
In operation, unreliable inspection systems provide false defect detection-both positive and negative-inconsistent gaging results, incorrect sorting decisions or inconsistent characterizations. The systems appear to be working satisfactorily and efficiently but produce too many false positives or miss too many targets of interest. In short, they make an unacceptable number of mistakes for practical use.
Practical Background or Noise Signals
The most common causes of signal degradation are:
Acceptable changes in physical properties not of interest.
Complex optical images or fields.
In online product inspection systems, acceptable changes in physical properties often rise from product variations in color, shape and surface roughness. Theses variations do not affect the performance of the product but can affect optical sensing in the optical field and confuse the sensing system.
Complex optical fields or images are present in defect detection on integrated circuit production. This requires the identification of small, sub-micron defects within the image containing complex circuitry that covers thousands of microns. Verifying the integrity of complex mechanical assemblies provides other challenges.
In addition, background variations occur for many reasons, including changes in the sun's position through factory windows, moving crane lights and variations in light source intensity. The system performance also can be degraded by targets moving out of the optical depth-of-focus or optical field-of-view.
These same concerns apply to sensing systems in a large variety of detection situations with rapidly moving targets from detection of enemy vehicles and aircraft to detection of golf balls, baseballs and cancer cells.
As mentioned earlier, unreliability can be determined by the frequency of false negatives and false positives: How often is a defect missed when there actually is a defect present? How often is a defect detected when there
is no defect?
Therefore, the basic questions are:
How can the sensing and decision operations be confused?
How does the light carry information and how can this information be distorted?
How can the sensing system be chosen or changed to provide practical, acceptable performance?
What are the changes in the operating requirement, such as electronic response time, when a different operating requirement, such as optical resolution, is changed?
Techniques for Discriminating a Feature from Background Noise
In any type of detection system, the most reliable and simplest system is usually obtained by a high signal-to-noise (S/N) ratio in the first stage of the detection system.
In optical sensing and detection, this improvement in S/N ratios can be generated by using one of four general optical properties to discriminate a feature from background noise: geometry, spectral properties, polarization properties and induced optical properties.
Any less-than-optimal performance in the first stage of any multistage detection system must be dealt with in succeeding stages at a probable increase of cost, complexity, time and reliability.
An example of each general optical property includes:
Geometry: In inspection of integrated circuits and photomasks used in their production, the use of optical spatial filtering can virtually eliminate the imaging of the highly geometrical rectangular circuit information and provide images containing only defect information.
Spectral Properties: In inspection of plastic surgical thread (or sutures), the use of suitable infrared filters eliminated signals from the thread and increased S/N to a value of about six with filter from a value of about one without the filter.
Polarization Properties: In inspection of non-metals, the use of polarization reduces unwanted reflections from surface and subsurface interfaces.
Induced Optical Properties: In detection of subsurface lamination defects, induced stresses can produce optical signals that image as bulls-eyes, at the surface, centered around the defect.
Discriminating Between Background / Noise Features and Brass Ball
The optical field may contain practical background or noise signals that can interfere with the detection. After optical improvements in the S/N ratio have been made, then analog and digital techniques can be applied.
Using the example of a brass ball moving across a screen can provide a simple, helpful model for considering instrumental alternatives and trade-offs in optics, photo-detectors, electronics and signal processing.
These include using time-discrimination techniques, including subtracting an image containing slowly varying background features from an image containing both a rapidly moving target and the background; optical collection and sensors shaped and positioned to compensate for some classes of slow variations; optical and digital shape/size discrimination; Fourier and wavelet analysis; and specialized analytic techniques for specific situations. Q
A system in which the signal contains nondefect-related information and results in sensing errors can be termed an unreliable system.
In online product inspection systems, acceptable changes in physical properties often rise from product variations in color, shape and surface roughness. These variations do not affect the performance of the product but can affect optical sensing in the optical field and confuse the sensing system.
Unreliability can be determined by the frequency of false negatives and false positives.