Biomedical Engineering Reference
In-Depth Information
Chapter 10
Texture Recognition in Micromechanics
The main approaches to microdevice production are microelectromechanical sys-
tems (MEMS) [
1
,
2
] and microequipment technology (MET) [
3
-
7
]. To get the most
out of these technologies, it is important to have advanced image recognition
systems. In this chapter, we propose the Random Subspace Neural Classifier
(RSC) for metal surface texture recognition.
10.1 Metal Surface Texture Recognition
Examples of metal surfaces are presented in Fig.
10.1
. Due to changes in viewpoint
and illumination, the visual appearance of different surfaces can vary greatly,
making recognition very difficult [
8
]. Different lighting conditions and viewing
angles greatly affect the grayscale properties of an image due to such effects as
shading, shadowing, and local occlusions. The real surface images, which it is
necessary to recognize in industrial environments, have all these problems and
more, such as dust on the surface.
The task of metal surface texture recognition is important for automating the
assembly processes in micromechanics [
3
]. To assemble any device, it is necessary
to recognize the position and orientation of the workpieces to be assembled [
4
].
Identification of a workpiece surface is useful in recognizing its position and
orientation. For example, let a shaft have two polished cylinder surfaces for
bearings, one with a milled groove for a dowel joint, and the other turned by a
lathe. It will be easier to obtain the orientation of the shaft if we can recognize both
types of surface textures.
Our texture recognition system has the following structure (Fig.
10.2
). The
texture image serves as input data to the feature extractor. The extracted features
are presented to the input of the encoder. The encoder produces an output binary
vector of large dimension, which is presented to the input of the one-layer neural
classifier. The output of the classifier gives the recognized class. Later, we will
describe all these blocks in detail. To test our neural classifier RSC, we created our
