• Home
  • Textbooks
  • Fuzzy Logic with Engineering Applications
  • Fuzzy Classification and Pattern Recognition

Fuzzy Logic with Engineering Applications

Timothy Ross

Chapter 11

Fuzzy Classification and Pattern Recognition - all with Video Answers

Educators


Chapter Questions

Problem 1

A fuzzy tolerance relation, R , is reflexive and symmetric. Find the equivalence relation $\mathrm{R}_{\text {e }}$ and then classify it according to $\lambda$-cut levels $=\{0.9,0.8,0.5\}$.

$$
\underline{R}=\left[\begin{array}{ccccc}
1 & 0.8 & 0 & 0.2 & 0.1 \\
0.8 & 1 & 0.9 & 0 & 0.4 \\
0 & 0.9 & 1 & 0 & 0.3 \\
0.2 & 0 & 0 & 1 & 0.5 \\
0.1 & 0.4 & 0.3 & 0.5 & 1
\end{array}\right]
$$

Check back soon!

Problem 2

In a pattern recognition test, four unknown patterns need to be classified according to three known patterns (primitives) $a, b$, and $c$. The relationship between primitives and unknown patterns is in the following table:
If a $\lambda$-cut level is 0.5 , then into how many classes can these patterns be divided?
Hint: Use a max-min method (see Chapter 3) to first generate a fuzzy similarity relation R.

Check back soon!

Problem 3

As a first step in automatic segmentation of magnetic resonance imaging (MRI) data regarding the head, it is necessary to determine the orientation of a data set to be segmented. The standard radiological orientations are sagittal, coronal, and horizontal. One way to classify the orientation of the new data would be to compare a slice of the new data to slices of known orientation. To do the classification we will use a simple metric obtained by overlaying slice images and obtaining an area of intersection, then normalizing these, based on the largest area of intersection. This metric will be our "degree of resemblance" for the equivalence relation. From data you have the following fuzzy relation:
Sagittal
Coronal
Horizontal
New slice $\left[\begin{array}{llll}1 & \multicolumn{1}{c}{\mathrm{C}} & \mathrm{H} & \mathrm{N} \\ 0.6 & 0.4 & 0.7 \\ 0.6 & 1 & 0.5 & 0.7 \\ 0.4 & 0.5 & 1 & 0.5 \\ 0.7 & 0.7 & 0.5 & 1\end{array}\right]$
(a) What kind of relation is this?
(b) Determine the equivalence relation and conduct a classification at $\lambda$-cut levels of $0.4,0.6$, and 0.7 .

Check back soon!

Problem 4

(Note: This problem will require a computerized form of the $c$-means algorithm.) Suppose we conduct a tensile strength test of four kinds of unidentified material. We know from other sources that the materials are from two different categories. From the yield stress, $\sigma_y$, and yield strain, $\Delta_y$, data determine which materials are from the two different categories (see Fig. P11.4).
Determine which values for $m^{\prime}$ and $\varepsilon_L$ would give the following results after 25 iterations, i.e.,

$$
\mathrm{U}^{(25)}=\left[\begin{array}{llll}
0.911 & 0.824 & 0.002 & 0.906 \\
0.089 & 0.176 & 0.998 & 0.094
\end{array}\right]
$$

and final cluster centers of

$$
\mathbf{v}_1=\{8.4581 .634\} \quad \mathbf{v}_2=\{7.3461 .792\}
$$

Check back soon!

Problem 5

A problem in construction management is to allocate four different job sites to two different construction teams such that the time wasted in shuttling between the sites is minimized. Let the job sites be designated as $x_i$ and combined to give a universe, $X=\left\{\mathbf{x}_1, \mathbf{x}_2, \mathbf{x}_3, \mathbf{x}_4\right\}$. If the head office, where the construction teams start every day, has coordinates $\{0,0\}$, the following vectors give the locations of the four job sites:

$$
\begin{aligned}
& \mathbf{x}_1=\{4,5\} \\
& \mathbf{x}_2=\{3,4\} \\
& \mathbf{x}_3=\{8,10\} \\
& \mathbf{x}_4=\{9,12\}
\end{aligned}
$$

Conduct a fuzzy $c$-means calculation to determine the optimum partition, $\mathrm{U}^*$. Start with the following initial 2-partition:

$$
\mathrm{U}^{(0)}=\left\{\begin{array}{llll}
1 & 1 & 0 & 0 \\
0 & 0 & 1 & 1
\end{array}\right\}
$$

(Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!

Problem 6

A radar image of a vehicle is a mapping of the bright (most reflective) parts of it. Suppose we have a radar image that we know contains two vehicles parked close together. The threshold on the instrument has been set such that the image contains seven bright dots. We wish to classify the dots as belonging to one or the other vehicle with a fuzzy membership before we conduct a recognition of the vehicle type. The seven bright dots are arranged in a matrix $X$, and we seek to find an optimum membership matrix $\mathrm{U}^*$. The features defining each of the seven dots are given here:
Start the calculation with the following initial 2-partition:

$$
\mathbf{U}^{\circ}=\left[\begin{array}{lllllll}
0 & 0 & 0 & 0 & 0 & 0 & 1 \\
1 & 1 & 1 & 1 & 1 & 1 & 0
\end{array}\right]
$$

Find the converged optimal 2-partition. (Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!

Problem 7

In a magnetoencephalography (MEG) experiment, we attempt to partition the space of dipole model order versus reduced chi-square value for the dipole fit. This could be useful to an MEG researcher in determining any trends in his or her data-fitting procedures. Typical ranges for these parameters would be as follows:

$$
\begin{aligned}
& \text { Dipole model order }=(1,2, \ldots, 6)=x_{1 i} \\
& \text { Reduced } \chi^2 \in(1,3)=x_{2 i}
\end{aligned}
$$

Suppose we have three MEG data points, $\mathbf{x}_i=\left(x_{1 i}, x_{2 i}\right), i=1,2,3$, to classify into two classes. The data are

$$
\mathbf{x}_1=(2,1.5) \quad \mathbf{x}_2=(3,2.2) \quad \mathbf{x}_3=(4,2)
$$
Find the optimum fuzzy 2-partition using the following initial partition:
(Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!

Problem 8

Suppose we want to sample a complex signal from a demodulator circuit and classify it into two sets, $\mathrm{A}_8$ or $\mathrm{A}_1$. The sample points are $x_1=(-3,1), x_2=(-2,2), x_3=(-1,1.5)$, and $x_4=(1,2)$ as shown in Fig. P11.8. If the first row of your initial 2-partition is [ 10000 ], find the fuzzy 2-partition after three iterations, i.e., find $\mathrm{U}^{(3)}$. (Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!

Problem 9

A small number of sequential operations can effectively limit the speedup of a parallel algorithm. Let $f$ be the fraction of operations in a computation that must be performed sequentially, where $0 \leq f \leq 1$. According to Amdahl's law, the maximum speedup $s$ achievable by a parallel computer with $p$ processors is

$$
s \leq \frac{1}{f+(1-f) / p}
$$

Suppose we have three data points, each described by two features: the fraction of sequential operations $(f)$, and the maximum efficiency $(s)$. These data points and their features are given in the following table:
We want to classify these three data points into two classes $(c=2)$ according to the curves of Amdahl's law. Of the three possible hard partitions ( $n_{\mathrm{M}_2}=3$ ), the one that seems graphically plausible is

$$
\left[\begin{array}{lll}
1 & 1 & 0 \\
0 & 0 & 1
\end{array}\right]
$$

Using this partition as the initial guess, find the fuzzy 2-partition after two cycles, $\mathrm{U}^{(2)}$. (Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!

Problem 10

We want to classify the performance of three computer systems based on throughput (in mips) and response time (in seconds). The data points in our sample, $X=\left\{\mathbf{x}_1, \mathbf{x}_2, \mathbf{x}_3\right\}$, are $\mathbf{x}_1=(50$, $10), \mathbf{x}_2=(40,12)$, and $\mathbf{x}_3=(20,5)$. Using the initial 2-partition,

$$
\underline{U}^{(0)}=\left[\begin{array}{lll}
1 & 1 & 0 \\
0 & 0 & 1
\end{array}\right]
$$

verify that the optimum fuzzy 2-partition after two cycles is

$$
\underline{\mathrm{U}}^{(2)}=\left[\begin{array}{lll}
0.974 & 0.9418 & 0 \\
0.026 & 0.0582 & 1
\end{array}\right]
$$

(Use $m^{\prime}=2.0$ and $\varepsilon_L \leq 0.01$.)

Check back soon!
04:54

Problem 11

There are many different grades of naphtha, which is a mixture of hydrocarbons characterized by a boiling point range between $80^{\circ} \mathrm{C}$ and $250^{\circ} \mathrm{C}$. There are four types of naphtha $(n=4)$ that are characterized based on density, average molecular weight, and hydrogen-to-carbon $(\mathrm{H} / \mathrm{C})$ molar ratio $(m=3)$ :
There are several studies that predict the products of the naphtha pyrolysis based on light and medium naphtha. It would be useful to classify the above four types into either light or medium classes $(c=2)$.

Using $m^{\prime}=2$ and $\varepsilon=0.01$ conduct the following:
(a) Hard $c$-means.
(b) Fuzzy c-means.
(c) Find the classification metric.
(d) Find the similarity relation that results from the U-partition found in part (b).

Ronald Prasad
Ronald Prasad
Numerade Educator
04:22

Problem 12

In gas systems there are two basic properties of gas (temperature and pressure) that can be used to determine whether calculations for the system can be done using the ideal gas law or if a more robust property package is required. The ideal gas law applies to systems with high temperature $(T)$ and low pressure $(P)$. The drawback of employing a more robust approach is that the computational requirements increase tremendously.
Classify the given the set of data for $P$ and $T$ into two classes to determine which systems require a robust property package for thermodynamic calculations. Use $m^{\prime}=2$ and $\varepsilon=0.01$ for the following:
(a) Hard $c$-means.
(b) Fuzzy c-means.
(c) Find the classification metric.
(d) Find the similarity relation using the U-partition from part (b).

Eric Mockensturm
Eric Mockensturm
Numerade Educator

Problem 13

The following data points describe the temperature, density and mono-ethanol-amine (MEA) weight fraction for an MEA/water solution. Classify the data into three classes using $m^{\prime}=2$ and $\varepsilon=0.001$.
Conduct the following:
(a) Hard $c$-means.
(b) Fuzzy c-means.
(c) Discuss the classification and how it relates to the original data and how the partitions from parts (a) and (b) are related.
(d) Find the classification metric.

Check back soon!
01:10

Problem 14

The biomechanical department of a prominent university is conducting research in bone structure. One study involves developing a relationship between the wrist joint angle and the sarcomere length in the lower arm. In this study the following data were obtained:
(a) Classify these data, in one cycle, into two classes using the hard $c$-means method.
(b) Classify these data into two classes using the fuzzy $c$-means method; use $m^{\prime}=2$ and $\varepsilon=$ 0.01 and conduct two cycles. What is the value of the accuracy at the end of two cycles?
(c) Find the classification metric.
(d) Find the similarity relation for the U-partition that results from part (b).

Manik Pulyani
Manik Pulyani
Numerade Educator

Problem 15

Show that when two separate fuzzy vectors are identical, i.e., $\mathbf{a}=\mathbf{b}$, the inner product $\mathbf{a} \cdot \mathbf{b}^{\mathrm{T}}$ reaches a maximum as the outer product $\mathbf{a} \oplus \mathbf{b}^{\mathrm{T}}$ reaches a minimum.

Check back soon!

Problem 16

For two fuzzy vectors $\underset{\sim}{\mathbf{a}}$ and $\underset{\sim}{\mathbf{b}}$ and the particular case where $\hat{a}=\hat{b}=1$ and $\mathbf{a}=\mathbf{b}=0$, show that when $\underset{\sim}{\mathbf{a}}=\underset{\sim}{\mathbf{b}}$, then the inner product $\underset{\sim}{\mathbf{a}} \cdot{\underset{\sim}{\mathbf{b}}}^{\mathrm{T}}=1$ and the outer product $\underset{\sim}{\mathbf{a}} \oplus \underset{\sim}{\mathbf{b}^{\mathrm{T}}}=0$.

Check back soon!

Problem 17

For two fuzzy vectors a and $\mathbf{b}$, prove the following expressions (transpose on the second vector in each operation is presumed):
(a) $\overline{\mathbf{a}} \bullet \underline{\mathbf{b}}=\mathbf{\overline { \mathbf { a } }} \oplus \overline{\mathbf{b}}$
(b) $a \cdot a \leq 0.5$

Check back soon!

Problem 18

Prove the following:
(a) For any $\underset{\sim}{\mathrm{A}} \in \mathrm{P}^*(\mathrm{X})$, prove that $(\mathrm{A}, \underset{\sim}{\mathrm{A}})_{1 \propto 2}=1$.
(b) For any A on X , prove that
$(\mathrm{A}, \overline{\mathrm{A}})_1 \leq \frac{1}{2}$
$(\mathrm{A}, \overline{\mathrm{A}})_2 \leq \frac{1}{2}$

Check back soon!

Problem 19

Show that the metric in Eq. (11.53) always gives a value less than or equal to the metric in Eq. (11.54) for any pair of fuzzy sets.

Check back soon!

Problem 20

In signal processing the properties of an electrical signal can be important. The most sought-after properties of continuous time signals are their magnitude, phase, and frequency exponents. Three of these properties together determine one sinusoidal component of a signal where a sinusoid can be represented by the following voltage:

$$
V(t)=A \sin \left(f_0 t-\theta\right)
$$

where $A=$ magnitude (or amplitude) of the sinusoidal component
$f_0=$ fundamental frequency of the sinusoidal component
$\theta=$ phase of the sinusoidal component
With each of these properties representing a "feature" of the electrical signal, it is possible to model a fuzzy pattern recognition system to detect what type of sinusoidal components are present. Let us define the prototypical values for pattems of magnitude, frequency, and phase that we are interested in:
Draw the resulting three-feature membership graphs.
Now let the input sinusoidal signal vector $\mathbf{B}$ comprise three crisp singletons, i.e., B $=\left\{5 \mathrm{~V}, 45 \mathrm{~Hz}, 45^{\circ}\right\}$, with weights of $0.6,0.2$, and 0.2 assigned to each of the corresponding features. Determine which pattern vector $\mathbf{B}$ most closely resembles.

Check back soon!

Problem 21

Using the same patterns as in Problem 11.20, but with a new input fuzzy pattern $\underset{\sim}{B}$ and features given by

$$
\begin{aligned}
B_{\text {voltage }} & =\left\{\frac{0}{1}+\frac{0.2}{2}+\frac{0.7}{3}+\frac{1.0}{5}+\frac{0}{6}\right\} \\
B_{\text {freqpency }} & =\left\{\frac{0}{20}+\frac{0.5}{30}+\frac{1.0}{40}+\frac{0.4}{50}+\frac{0}{60}\right\} \\
\mathrm{B}_{\text {pthase }} & =\left\{\frac{0}{50}+\frac{0.3}{70}+\frac{0.7}{90}+\frac{1}{110}+\frac{0.7}{120}+\frac{0}{130}\right\}
\end{aligned}
$$

determine the pattern that most closely matches the input pattern.

Check back soon!

Problem 22

Transuranic waste will be stored at a southeastern New Mexico site known as WIPP. The site has underlying strata of rock salt, which is well-known for its healing and creeping properties. Healing is the tendency of a material to close fractures or other openings, and creep is the capacity of the material to deform under constant load. The radioactive wastes are stored in rooms excavated deep underground. Because of the creep of the ceiling, these rooms will eventually collapse, thus permanently sealing the wastes in place. The creep properties of salt depend on the depth, moisture content, and clay content of the salt at the location being considered. Rock salt from specified depths was studied through numerous tests conducted at various labs nationwide. These data comprise the known patterns. Hence, each pattern has three features. Now, the possibility of locating a room at a certain depth is being investigated.

We wish to determine the creep properties at some depth of salt with a certain clay and moisture content. Membership functions for each of the patterns are shown in Fig. P11.22.
Find which known pattern the unknown pattern matches the best. The features for the unknown pattern are given by the crisp singletons

$$
\mathrm{B}=\{\text { depth }=1750 \mathrm{ft}, \text { clay content }=6.13 \%, \text { moisture content }=12.5 \%\}
$$

The weights given to the features are $W=\{0.5,0.3,0.2\}$.

Check back soon!

Problem 23

Using the same known pattern as in Problem 11.22, and using fuzzy features for the new pattern, find which known pattern matches the new pattern most closely. Features for the new pattern are as follows:

$$
\begin{aligned}
\mathrm{B}_{\text {depth }} & =\left\{\frac{0}{1700}+\frac{0.5}{1725}+\frac{1}{1750}+\frac{0.5}{1775}+\frac{0}{1800}\right\} \\
\mathrm{B}_{\text {clay coesent }} & =\left\{\frac{0}{5.5}+\frac{0.5}{5.813}+\frac{1.0}{6.13}+\frac{0.5}{6.44}+\frac{0}{6.75}\right\} \\
\mathrm{B}_{\text {moisture coencent }} & =\left\{\frac{0}{11.0}+\frac{0.5}{11.75}+\frac{1}{12.5}+\frac{0.5}{13.25}+\frac{0}{14.0}\right\}
\end{aligned}
$$

Check back soon!
01:03

Problem 24

A member of the police bomb squad has to be able to assess the type of bomb used in a terrorist activity in order to gain some knowledge that might lead to the capture of the culprit. The most commonly used explosive device is the pipe bomb. Pipe bombs can be made from a variety of explosives ranging from large-grain black powder and gunpowder to more sophisticated compounds, such as PETN or RDX. Identification of the explosive material used in the pipe bomb (after detonation) will tell a bomb squad investigator where the materials might have been purchased, the level of sophistication of the terrorist, and other important identifiers about the criminal.

Four basic types of energetic materials are used in making pipe bombs, each with its own distinctive pattern of post-mortem damage.
1. Explosives. Those that detonate at a velocity equal to the compressional sound speed of the explosive material itself. These materials are by far the most energetic (also the most difficult to acquire) and are characterized (after explosion) by very small pipe fragments, highly discolored (usually bluish in tint) fragments, and extreme collateral damage (especially close to the detonation).
2. Propellants. Usually formed from some compound based on nitrocellulose. These materials do not detonate, but burn very rapidly. Usually propellants are formed in special geometric shapes that allow their surface area to remain constant or increase as they burn, thus causing the burning rate to increase until the compound has been completely exhausted. The destructive force of propellant-based pipe bombs is somewhat less than that of true explosives and is characterized by medium fragment size, little discoloration, and moderate collateral damage.
3. Large-grain black powder. Has been around since about 600 BC , when the Chinese discovered the carbon-sulfur-potassium nitrate mixture. The size of black powder grains can vary tremendously, but the geometry is such that the powder always burns down (the burn rate always decreases once the entire surface of the mixture is burning). Although still very deadly, the damage from these types of pipe bombs is less than that of the other two. The residual fragment size is larger, and the discoloration of the fragments is slight.
4. Gunpowder. A subclass of black powder, usually considered to be homemade. It is characterized by very large fragments (one or two in number), almost no discoloration, and little collateral damage. Black powder is still very common among terrorists.
We can form a table of patterns for each feature:
Assume that the membership space for each feature can be partitioned similarly into four sections on a normalized abscissa, as shown in Fig. P11.24. The other two graphs (for discoloration and collateral damage) would look identical to the one in Fig. P11.24 (with different labels). The weights assigned to each feature are $0.5,0.3$, and 0.2 , respectively. Now, say a new bombing has taken place and the aftereffects measured over three features are denoted as singletons, given as

$$
\mathrm{B}=\left\{\text { fragment size }=\frac{1}{0.7}, \text { fragment color }=\frac{1}{0.6}, \text { damage }=\frac{1}{0.5}\right\}
$$

Determine the composition of the bomb used in the bombing.

Michael Sorola
Michael Sorola
Numerade Educator

Problem 25

Using the information in Problem 11.24, but selecting a fuzzy input, perform a pattern recognition. The fuzzy input patterns in the form of triangular fuzzy numbers based on the three features are as follows:

$$
\begin{aligned}
\mathrm{B}_{\text {frage size }} & =\left\{\frac{0}{0.43}+\frac{1}{0.5}+\frac{0}{0.57}\right\} \\
\mathrm{B}_{\text {trige. color }} & =\left\{\frac{0}{0.53}+\frac{1}{0.6}+\frac{0}{0.67}\right\} \\
\mathrm{B}_{\text {dumage }} & =\left\{\frac{0}{0.63}+\frac{1}{0.7}+\frac{0}{0.77}\right\}
\end{aligned}
$$

Check back soon!

Problem 26

We intend to recognize preliminary data coming off a satellite. Each of the five data packets has a unique packet header identifier, as follows:

$$
\begin{aligned}
& A_1=\text { satellite performance metrics } \\
& A_2=\text { ground positioning system }
\end{aligned}
$$

$$
\begin{aligned}
& \mathrm{A}_3=\mathrm{IR} \text { sensor } \\
& \mathrm{A}_4=\text { visible camera } \\
& \mathrm{A}_5=\text { star mapper }
\end{aligned}
$$

The three header values each set will look for are (1) signal type, (2) terminal number, and (3) data identifier. The weights assigned to each of the headers are $0.3,0.3$, and 0.4 , respectively. Let us define the fuzzy pattern as

$$
\begin{aligned}
& \mathrm{A}_1=\left\{\frac{0.2}{x_1}+\frac{0.2}{x_2}+\frac{0.6}{x_3}\right\} \\
& \mathrm{A}_2=\left\{\frac{0.3}{x_1}+\frac{0.4}{x_2}+\frac{0.7}{x_3}\right\} \\
& \mathrm{A}_3=\left\{\frac{0.4}{x_1}+\frac{0.6}{x_2}+\frac{0.8}{x_3}\right\} \\
& \mathrm{A}_4=\left\{\frac{0.5}{x_1}+\frac{0.8}{x_2}+\frac{0.9}{x_3}\right\} \\
& \mathrm{A}_5=\left\{\frac{0.6}{x_1}+\frac{1.0}{x_2}+\frac{1.0}{x_3}\right\}
\end{aligned}
$$

A data stream given by the crisp singleton

$$
\mathrm{B}=\left\{\frac{1.0}{x_1}+\frac{1.0}{x_2}+\frac{1.0}{x_3}\right\}
$$
is received. Determine which of the five different packets we are receiving at the present time.

Check back soon!

Problem 27

Signals are investigated from the following four digital signal processing plants: $A_1=$ least mean squares, $\mathrm{A}_2=$ root-mean square, $\mathrm{A}_3=$ Newton's method, and $\mathrm{A}_4=$ steepest descent method. The three ( $m=3$ ) important parameters that will be considered in each $c$-space are convergence rate, tracking, and stability. The weights assigned to each of the features are 0.4 , 0.4 , and 0.2 , respectively. The data patterns corresponding to the features are membership triangles

$$
\begin{aligned}
& \mathrm{A}_1=\{0.2,0.3,0.8\} \\
& \mathrm{A}_2=\{0.4,0.4,0.6\} \\
& \mathrm{A}_3=\{0.6,0.2,0.4\} \\
& \mathrm{A}_4=\{0.8,0.5,0.2\}
\end{aligned}
$$

The sample data set has the following vector pattern as a membership triangle:

$$
\mathrm{B}=\{0.5,0.5,0.5\}
$$

Determine the pattern most closely represented by the sample data set.

Check back soon!

Problem 28

Lube oils are classified by three features: color, viscosity, and flash point. Depending on the values of these features, the lube oil is classified as 100 neutral $(100 \mathrm{~N}), 150$ neutral $(150 \mathrm{~N})$, heavy solvent neutral (HSN), and 500 neutral ( 500 N ). Among the features, color is the most important, followed by viscosity, then flash point. The reason for this ordering is that it is easier to blend lube oils to obtain correct viscosity and flash point than it is to blend to obtain proper color. Any material not falling into one of these lube oil categories is downgraded to catalyst cracker feed (PGO), where it is converted to gasoline.

Fuzzy patterns for each of these features are shown in Fig. P11.28. The weights for these features are 0.5 for color, 0.3 for viscosity, and 0.2 for flash point. You receive a lab analysis for a sample described by the crisp singleton

$$
B=\left\{\text { color }=6.5, \text { viscosity }=825 \mathrm{~m}^2 / \mathrm{s}, \text { flash point }=750^{\circ} \mathrm{C}\right\}
$$

Under what category do you classify this sample?

Check back soon!
00:47

Problem 29

Over several years a satellite tracking facility has classified several objects on the universe of signal to noise ratio (SNR), total signal (TS), and radius (R). The fuzzy sets are shown here for four satellites:

ASTEX

DMSP

$$
\begin{aligned}
& \mathrm{A}_1=\left\{\frac{0.1}{\mathrm{SNR}}+\frac{0.15}{\mathrm{TS}}+\frac{0.2}{\mathrm{R}}\right\} \\
& \mathrm{A}_2=\left\{\frac{0.2}{\mathrm{SNR}}+\frac{0.2}{\mathrm{TS}}+\frac{0.3}{\mathrm{R}}\right\}
\end{aligned}
$$
SEASAT
MIR $\quad \mathrm{A}_4=\left\{\frac{0.9}{\mathrm{SNR}}+\frac{0.9}{\mathrm{TS}}+\frac{0.9}{\mathrm{R}}\right\}$
with weights $w_1=0.4, w_2=0.4, w_3=0.2$. One night an unknown object is tracked, and the following observation is made:

$$
\mathrm{B}=\left\{\frac{0.3}{\mathrm{SNR}}+\frac{0.3}{\mathrm{TS}}+\frac{0.3}{\mathrm{R}}\right\}
$$

Which satellite does the object most closely resemble?

Sarah Wharton
Sarah Wharton
Numerade Educator

Problem 30

A set of patterns indicating the performance of an aluminum smelting cell is developed. The important features are bath temperature $T\left({ }^{\circ} \mathrm{C}\right.$ ), cell voltage $V$, and noise $N$ (standard deviation of the cell resistance). The cell conditions (pattems) are described as follows:
The fuzzy sets are represented by Gaussian membership functions:

$$
\begin{aligned}
& \mathrm{A}_1=\left\{\exp \left[-\frac{(T-945)^2}{4^2}\right], \exp \left[-\frac{(V-4.2)^2}{(0.1)^2}\right], \exp \left[-\frac{(N-26)^2}{5^2}\right]\right\} \\
& \mathrm{A}_2=\left\{\exp \left[-\frac{(T-950)^2}{4^2}\right], \exp \left[-\frac{(V-4.4)^2}{(0.1)^2}\right], \exp \left[-\frac{(N-6)^2}{2^2}\right]\right\} \\
& \mathrm{A}_3=\left\{\exp \left[-\frac{(T-970)^2}{8^2}\right], \exp \left[-\frac{(V-4.8)^2}{(0.1)^2}\right], \exp \left[-\frac{(N-6)^2}{2^2}\right]\right\} \\
& A_4=\left\{\exp \left[-\frac{(T-965)^2}{6^2}\right], \exp \left[-\frac{(V-4.7)^2}{(0.1)^2}\right], \exp \left[-\frac{(N-20)^2}{5^2}\right]\right\}
\end{aligned}
$$

To reflect the relative importance of the features, select $w_1=0.3, w_2=0.5$, and $w_3=0.2$. Now a new data sample (measurements from a smelting cell) yields temperature $=953^{\circ} \mathrm{C}$, voltage $=4.5 \mathrm{~V}$, and noise $=12$ (a data singleton). Classify the operating conditions of the cell.

Check back soon!

Problem 31

Use the same patterns as in Problem 11.30. But now use a sample comprising fuzzy sets. This is appropriate because measurements such as temperature are subject to substantial error, and electrical signals fluctuate over time as disturbances affect the system. The new sample is represented as the following:

$$
\mathrm{B}=\left\{\exp \left[-\frac{(T-957)^2}{3^2}\right], \exp \left[-\frac{(V-4.6)^2}{(0.2)^2}\right], \exp \left[-\frac{(N-16)^2}{3^2}\right]\right\}
$$

Classify the operating conditions of the cell based on this information.

Check back soon!

Problem 32

Skis are classified on the basis of three features: weight, performance stiffiness, and response times in turns. There are four different types of skis: freestyle, giant slalom (GS), slalom, and all-around. These skis have Gaussian distributions on each of the features and the parameters for the distribution are given in following table. A Gaussian distribution has the form

$$
\mu_{\mathrm{A}_j}(x)=\exp \left[-\left(\frac{x_j-a_{i j}}{\sigma_{\alpha_{i j}}^2}\right)^2\right]
$$

The fuzzy patterns are defined on a normalized scale as follows:
The weights given to the features are

$$
\begin{aligned}
w_{\text {weight }} & =0.3 \\
w_{\text {stifhess }} & =0.4 \\
w_{\text {response }} & =0.3
\end{aligned}
$$

A new ski whose features are given on a normalized scale by a crisp singleton,

$$
\mathrm{B}=\{\text { weight }=45, \text { stiffness }=60 \text {, response time }=65\}
$$

is introduced into the market. Determine what type of ski the new ski should be labeled.

Check back soon!

Problem 33

In Problem 11.32 the new ski introduced into the market was given by a crisp singleton. However, given the uncertainty in measurements, it is more appropriate to define a ski by fuzzy parameters. For the same problem, and with the same weights assigned to each of the features, classify the new ski if it is given by a fuzzy set whose membership functions are given by a Gaussian distribution whose parameters (mean and standard deviation) are given in the following table:

Check back soon!
03:18

Problem 34

Generate a fuzzy grammar for the syntactic pattern recognition of an isosceles trapezoid.

Sheryl Stone
Sheryl Stone
Numerade Educator

Problem 35

Generate a fuzzy grammar for the syntactic pattern recognition of an equilateral triangle.

Check back soon!

Problem 36

Continue Example 11.28 by developing fuzzy grammars for the pattern recognition of the two electric sources, symbols $\ominus$ and $k$.

Check back soon!
14:33

Problem 37

The accompanying table shows the intensity values (for an 8 -bit image) associated with an array of 25 pixels. Use the image enhancement algorithm on these intensity values to enhance the image. Do you recognize the pattern in the image?

Geena Pullo
Geena Pullo
Numerade Educator
01:08

Problem 38

The following table shows the intensity values (for an 8-bit image) associated with an array of 25 pixels. Use the image-softening algorithm on these intensity values to remove the *"salt and pepper" noise (shown as shaded pixels) from the image of the alphabetic character M.

James Kiss
James Kiss
Numerade Educator