-2-
FIELD OF INVENTION
The invention relates to an expert system for ore sorting and ore classification to
improve the quality of ferruginous manganese ore feed to a ferromanganese
plant.
NOVELTY OF INNOVATION:
* Developed an innovative methodology for on-line ore sorting and ore
classification using image processing and radial bases neural network
techniques. This was used to develop an expert system for on line ore
sorting and ore classification for ferruginous manganese ores feed to a
ferromanganese plant.
* Developed a cost effective process for on line ore sorting and ore
classification.
BACKGROUND OF THE INVENTION
Most of Ferro-manganese plant in India utilizes lumpy (-75, + 10 mm)
ferruginous manganese ore of higher grade (Mn : > 46 %; Mn/Fe ratio : > 6;
Al2O3 ; < 3 %). Indian manganese ore deposits occur mainly as bedded
sedimentary deposit and shows significant difference in color due to variation in

-3-
ore composition, Ore lumps of high manganese (> 46 %) and high Mn / Fe ratio
(> 6) shows steel gray color. Ore containing low manganese and high iron with
low Mn / Fe ratio shows reddish brown color. Ores having higher alumina with
low Mn / Fe ratio shows white color To improve the ore grade and to maintain
the Mn / Fe ratio for Ferromanganese making ore sorting and ore blending is
most common method. Most common existing method to Improve the one
quality is manual we sorting and ore blending. X-ray based on line ore sorting
systems needs highly complicated system with high investment. So, any system
far on line ore sorting and ore classificatfon far better blending can be of
immense significance.
SUMMARY OF THE INVENTION
Image processing and artificial neural network are two most emerging
technologies to develop the machine vision systems. Digital image
(photographs) is a representation of a two-dimensional image as a finite set of
digital valves, called pixels. This representation of image is processed by various
operations to extract useful features from an image. An artificial neural network
(ANN) is an interconnected group of artificial neurons that uses a computational
model for information processing based on a connectionist approach to
computation. It can team from examples and a trained network can recognize
the unknown examples. These two techniques were used to develop this
methodology and process steps are explained and described here after.

-4-
Process Steps : There are four major steps involved in this methodology and
high speed computing is required in the system.
Step 1: Image actuisiton In this step various kind of ores and ore blends
are selected from feed material of the ferromanganese plant to produce digital
images. Digital images of these ores were produced by using digital camera
(Example: SANYO O.B Mega-pixel). The illumination at the site is to be sufficient
to support the image producing instrument for desired operation.
Step 2: Image processing :In this step various image produced were
categorized as high grade ore (Images of manganese enriched ores), ferruginous
ore (Images of Iron enriched ores) and tow grade ores (Images of alumina
enriched ores). These images are pre-processed to remove the noise and to
improve the contrast for better visibility of objects. It can be done by using
computer and high level computer language (C, C ++, Java etc.), A digital
image is a matrix representing the pixel intensities. The value of any matrix
element represents the brightness or that point in most of method minimum
brightness is called black and stands for zero (o) and maximum intensity is white
and stands for 225.
Step 2.1: RGB Color Model : Image data represents physical quantities such
as chromatics and luminance. Chromaticity is the color quantity defined by its
wave length white luminance is the amount of light. This information can be
expressed such as attributes like color and brightness. The most common color

-5-
models are RGB (red, green and blue), HSV (hue, saturation, value), and CMYK
(cyan, magenta, yellow, black). A color model is a method to represent color
and the if relationship to each other RGB color model was selected in our system
because of its simplicity and ore quality can be easily distinguished in these
colors. It was confirmed by chemical analysis that the color difference In the
ores is depending on the chemical composition and reddish color is due to iron
minerals and white color is due to alumina minerals and steel gray color is due to
manganese minerals. In RGB color (Red, Green, and Blue) analysis significant
difference was found in ail three kinds of ore images. Results are shown in
Figure 1.
Stop - 22. Histogram Analysis :An image histogram is a device that shows
the distribution of intensities image and simply a count of the gray levels (A
single value extracted by operation from different color values (R, G, B) of a
pixel) in the image, The image histograms of three ore categories are shown in
Figure 2, Image histogram analysis of three different kind images show the
difference in occurrences of different gray level intensities and it was shifting
towards left to right for steel gray manganese ores, reddish brown ferruginous
art and white alumina rich low grade ores respectively due to variation in the
increased frequency of high gray level values in the images (Figure 2).
Step 2.3. Image texture analysis: Image texture represents the orientation
of gray values in a image. There are various method to quantify the image
texture and we select the oldest and most reliable method suggested by Haralic
et. at. Four texturai features (Entropy, Energy, homogeneity and contrast) were
calculated from gray level co-occurrence matrix and these play a significant role
in ore type recognition and ore classification (Table 1).

6
A co-occurrence matrix is a square matrix with elements corresponding to the
relative frequency of occurrence of pairs of gray level of pixels separated by a
certain distance in a given direction. Formally, the elements of a GxG gray level
co-occurrence matrix Pd for a displacement vector d = (dx, dy) is defined as;

where I (.r) denote an image of size NxN with G gray values, (r,s), (tr v) NxN,
(t, v)= (r + dx, s + dy) and 1.1 is the cardinallty of a set. Harallick, Sharmugan
and O instein proposed fourteen measures of textural features which are derived
from the co-occurrence matrices, and each represents certain image properties
as coarseness, contrast, and homogenatty and texture complexity. Gray level co-
occurrence matrix was derived from the images and given textural features was
extracted.
(1) Entropy: It gives a measure of complexity of the image. Complex
textiles tend to have higher entropy.

(2) Contrast: It is a measure of the amount of local variations present in an
image. The higher the value of contrast is the sharper the structural
variations in the image.

-7-
(3) Energy: It is a measure of the homogeneity of art image and a suitable
measure for detection of disorders in textures. For homogeneous
textures value of energy burns out to be small compared to non-
homogenous ones,

(4) Homogenety: It is a parameter which represents monotonic of image
texture.

In Eqs. (2) - (5), P (i, j) refers to the normalized entry of the to-occurrence
matrices. That is p (i, j) = Pd (0, j) / R where R is the total number of pixel pairs
(i, j). For a displacement vector d = (dx, dy) and image of size NxM, R is given
by (N - dx) (M-dy). A data base was prepared for various ore categories on the
basis of the results shown in the table 1.
Table 1: Textural Features
Parameter
Measure
Order
Energy
Homoogeneity
EHigh > EFeruginuous >Low
Entropy
Complexity
EnFeruginuous > EnLow EnHigh
Homogeneity
Monotonicity
HHigh > HFeruginuous >ELow
Contrast'
Local variations
CLow > CFeruginuous > CHigh

8
Step 3: Ore Class Reorganisation by Artificial Neural Network: Artificial
Neural Networks are relatively crude electronic models based on the neural
structure of the twain. It learns from examples and used for nonlinear function
approximation, classification and pattern reorganization (figure 3). A data base
prepared containing (average color values (red, green and blue), entropy,
contrast, homogeneity and energy) for three ore categories using various images
of ores with small variations in ore quality. This database was used to train the
artificial neural network.. The trained neural network is able to detect new
images of various categories upto a satisfactory accuracy. Overall system
accuracy was 88.71 % and descriptive results are given in table 2.
Table 2: Neural network classification
Type of Imagesi
Ferruginous
ore
Low grade
High grade
High grade
9
4
122
Ferruginous
120
12
11
Low grade
6
100
1
Misclassified
15
16
12
Correctly classified
120
100
122
Total
135
116
134
Accuracy (%)
88.89
66.21
91.04
Misclasiffication (%)
11.11
13.79
8.95

Step 4: Expert System Development: An expert system was developed by
combining the all three steps explained in previous section. A digital camera is
needed to for image producing and image processing algorithms; to extract the
useful features from the images. The prepared datasets used to train the
artificial neural network. The trained neural network can label the new unknown
image of the feed ore. On the basis of this classification the ore blending can be
controlled to achieve desired grade of ores. Process methodology and expert
system for online implementations is shown in Figure 4 and 5,
It's Utility: The developed method and system will be suitable for developing
cost effective online ore sorting and ore classification system to improve the ore
quality and ore blending operation in ferruginous manganese ores.
Non Obvious: The same process and the system can be used in much other
type of oras and racks for on line quality control, separation and classification.
Estimated market value If patented: High for all the Indian ferromanganese
producers which produce alloy from ferruginous lumpy manganese ores. Many
other mineral industries which need ore sorting operation at various stages of
the process.

-10-
1. A device for on-line quality control, separation and classification of ores,
ore-blends and rocks to improve the quality of ore feed to a plant such as
ferromanganese plant, consisting of digital camera used to produce digital
images of ores, ore blends, and rocks and an artificial neural network
(ANN) which is an interconnected group of artificial neurons that uses a
computational model for information processing based on a connectionist
approach to computation.
2. A device as claimed in claim 1, wherein the digital image is a matrix
representing a two dimensional image as a finite set of digital values,
called pixels and the value of any matrix represents the brightness of that
point.
3. A device as claimed in claim 2, wherein the said representation of image
is processed by various operations to extract useful features from an
image and the image is pre-processed to remove the noise and to
improve the contrast for better visibility of object.

-12-
8. A process for on-line quality control, separation and classification of ores,
ore blends, rocks used in 3 plant such as ferromanganese plant consisting
of acquisition of digital image of ores, ore blends, and rocks selected from
feed material by using digital camera, categorizing the image as of high
grade ore, ferruginous ore and low grade ore, ore processing the image
to remove the noise and to improve the contrast for better visibility of
objects, and getting the result in respect of quality, grade, classification
and pattern re-organization of the ore, ore-blend and rock by using
artificial neural network.
9. The process as claimed in claim 8, wherein the pre-processing to remove
the now and to improve the contrast for better visibility of objects is
done by computer using any software or programming of any high level
computer language (C, C++, Java etc.).
10. The process as claimed in claim 9, wherein the images are processed to
measure the colour quantity which is defined by its wave length and the
amount of light that is brightness.

-13-
11 .The process as claimed in claim 9, wherein the distribution of intensities in
an image is measured by operation from different colour valves in the
image and it is a count of the gray levels in the image and the orientation
of gray valves in an image is also measured.
12.The process as claimed in claim 8, where in Artificial Neural Networks
are trained by using the data base when is prepared containing average
colour valves, entropy, contrast, homogeneity and energy for various ore
categories using various images of ores with small variations in one
quality.
13. The process as claimed in claim 12, wherein the trained neural network is
used to detect new images of various categories of ores upto to
satisfactory accuracy.
14. A device for on-line quality control, separation and classification of ores,
ore blends, and rocks used in a plant such as ferromanganese plant as
described in the Complete Specification and shown in the accompanying
figures.

-14
15. A process for on-line quality control, separation, and classification of ore,
ore blends, and rocks used in a plant such as ferromanganese plant, as
described in the Complete Specification and shown in the accompanying
figures.

The invention relates to Expert system for ore sorting. It is a device or a process
for on-line quality control, separation and classification of ores, ore-blends and
rocks to improve the quality of ore feed to a plant. It consists of digital camera
used to produce digital images of ores etc, and an artificial neural network which
is an inter connected group of artificial neurons that uses a computational model
for information processing based on a connectionist approach to computation.