Fruit Disease Detection using Color, Texture

Analysis and ANN

Ashwini Awate, Damini Deshmankar, Prof. Samadhan Sonavane
Gayatri Amrutkar, Utkarsha Bagul Dept. Of Computer Engineering
Dept. of Computer Engineering Sandip Institute of Technology and Research Center
Sandip Institute of Technology and Research Center Sandip Foundation, Mahiravani, Nasik, Maharashtra
Sandip Foundation, Mahiravani, Nasik, Maharashtra 422213.
422213.

Abstract – Now-a-days as there is prohibitive demand for Three fruits namely grapes, apple and pomegranate have
agricultural industry, effective growth and improved yield of fruit is been used for research in this paper. Types of fruits and their
necessary and important. For this purpose farmers need manual respective diseases are as follows: [1]
monitoring of fruits from harvest till its progress period. But 1. Grapes:
manual monitoring will not give satisfactory result all the times and
a) Black Rot: For grapes, black rot is most widely
they always need satisfactory advice from expert. So it requires
proposing an efficient smart farming technique which will help for occurring and severe disease. In this disease, fungus attacks
better yield and growth with less human efforts. We introduce a canes, tendrils, leaves and fruit and it is most devastating in
technique which will diagnose and classify external disease within hot and moist areas. Now-a-days using combination of sound
fruits. Traditional system uses thousands of words which lead to cultural practices, fungicides and resistant varieties it is
boundary of language. Whereas system that we have come up with, possible to control black rot. First sign of black rot is visible
uses image processing techniques for implementation as image is on leaves as black border forms around the edge and small
easy way for conveying. In the proposed work, OpenCV library is yellowish spot is formed, after that spots enlarges. Also center
applied for implementation. K-means clustering method is applied of lesions become reddish-brown. Spore furnish structure of
for image segmentation, the images are catalogue and mapped to
the fungus appear as a minute black dots and inside margin of
their respective disease categories on basis of four feature vectors
color, morphology, texture and structure of hole on the fruit. The lesion they are organize in a ring pattern. Outline of lesion is
system uses two image databases, one for implementation of query oval having purple to black color. Mostly symptoms appear
images and the other for training of already stored disease images. after half growth of grapes. [1]
Artificial Neural Network (ANN) concept is used for pattern
matching and classification of diseases.

Keywords:-OpenCV, K-means clustering, SURF, Artificial

Neural Network.

I. INTRODUCTION
Fig. 1 Black rot
The studies of fruit or plant can be determined by b) Powdery Mildew: It is originated by fungus
observable patterns of specific plant and it is critical to Uncinulanecator; it is also called as Oidium. Only grapes and
monitor health and detect disease within a plant. Through other few related species are affected by this fungus. It is most
proper management strategies such as pesticides, fungicides frequently occurring disease on grapes. Primary symptom of
and chemical applications one can facilitates control of powdery mildew is whitish or greenish powdery patches
diseases which interns improve quality. There are various appearing on the underneath of basal leaves. It also causes leaf
techniques available such as spectroscopic and imaging curling, withering along with blotched or deformation of badly
technology, applied to achieve superior plant disease control infected leaves. Old infections appear on dormant canes as
and management. [1] With smart farming today’s farmer can reddish brown. In case of premature powdery mildew
use decision tools and automation techniques which inflammation can cause less sugar content and small berry
seamlessly integrate product, knowledge and services for size. Cracking and scarring of berries can make fruit
better productivity, grading and surplus yield. The purpose of unsuitable for any purpose. One should be aware that many
this paper is to monitor diseases on fruits and suggest better wine makers have low tolerance for powdery mildew on the
solution for healthy yield and productivity with the help of grapes. [1][14]
Artificial Neural Network concept. System uses two image
databases, one for training of already stored infected area
image and other for execution of query images. [1]

978-1-4673-7910-6/15/$31.00 2015
c IEEE 970
 c) Apple Blotch: It is most common “summer disease
” of apples in the northwest. Apple blotch is caused due to two
different organisms. Economical loss and commercial quality
damage are some of the harsh outcomes of this disease. Signs
of disease appear as dark greenish-blue spots on the surface of
Fig. 2 Powdery mildew contaminated fruit. One to many nearly circular colonies are
c) Downy Mildew: It is deeply devastating infection develop individually. These symptoms occur 3-4 weeks later
on grapevine. During bloom, summer, rainfall and if the once leaflet falls. Large and unshaped colonies widens over
temperature goes to 10˚C (50˚F) in grape-growing areas across the fruit. [3]
the world downy mildew occurs. First symptom of downy
mildew can be seen on leaves after 6 to 7 days once the plant
is infected. It is caused by Plasmoparaviticola. Almost all
green portion of the grape are susplectible. Oil spots (yellow
circular spots) along with an oily appearance occur on the
leaves. Greater number of oil spots may be found under
favourable weather condition. Young oil spots on young foliar Fig. 6 Apple Blotch
are surrounded by brownish-yellow halo. Because of its 3. Pomegranate:
downy growth it is named as Downy mildew. [14] a) Bacterial Blight: This disease was first recorded in
Delhi(India) in the year 1952. Until 1998 Bacterial Blight was
considered a lower economic threat. However now-a-days this
disease occurs widely and has been recorded in all [Link]
disease occurs in all pomegranate-growing states like
Maharashtra, Karnataka, and Andhra Pradesh. Sepals, twigs
Fig. 3 Downy Mildew and pomegranate are affected by Bacterial blight. Preliminary
2. Apple: symptoms for the disease can be black colored spots
a) Apple Scab: It causes most devastating apple surrounded by bacterial slime. 90% yield of pomegranate
infection. It occurs throughout in the apple-growing areas. depletes due to bacterial blight. Fruits crack due to this
During the bloom in cool and wet weather apple scab is more disease. [4] [13]
severe but it is not reasonably significant in dry or warm
climates. Signs of apple scab are visible on leaves, petals,
flowers, husk, fruit, young shoots and bud scales of apple tree.
Mostly infection on the fruit and leaves are common and
obvious. [1]

Fig. 7 Bacterial Blight

b) Aspergillus Fruit Rot: Alias of alternaria fruit rot.
It appear when flower begins to open after the rainfall and it
infect the internal portion of pomegranate. Tiny off-color in
Fig. 4 Apple Scab the skin and less weight due to internal decay are some of the
b) Apple Rot: It is a fungus caused due to exterior signs of disease. But this problem usually is not
Botryosphaeriaobtusa. It attacks on leaves, bark and fruits of apparent until harvesting or during fruit sorting. Without any
apple tree. The first symptoms of apple rot appears on outer external symptoms fungus may grow within the fruit. Mostly,
surfaces of leaves 1 to 3 weeks after petiole fall as small, infected fruit show some yellowish to brownish-red
purple blotch after which centre turn brown tan and yellowish discoloration and are slightly off-color such as a pale red. [13]
brown. Second stage of apple rot occur after few weeks. In
this stage secondary enlargement of leaf spots occurs. Leaf
that are highly infected drop from the tree. [9]
Apple rot disease occur in three forms: 1) leaf blotch
on apple trees 2) fruit rot on apple trees 3) limb canker on
apple tree. [1]
Fig. 8 Aspergillus Fruit Rot
c) Gray Mold: Gray mold is also called as Botrytis
cinetea. This disease is more active and commonly occured
during post-harvest wash and spread when kept at room
temperature. Gray Mold damage flower part of pomegranate
and affects the fruit until its ripening. Once the fruit is washed
Fig. 5 Apple Rot or stored at high humidity, condensation or water on the
blossom tissues activates the fungal mycelium to begin

2015 International Conference on Green Computing and Internet of Things (ICGCIoT) 971
growing. The typical grayish coating of spores and pathogen help of multi-class support vector machine. It achieves
sporulats on the flower parts are developed. Eventually the precision up to 93%.
fungus will outspread inside fruit tissue the crown tissue will
be colonized. Infected fruit stored at high moisture. Fruit Detection using Improved Multiple Features based
Algorithm, Authors (Hetal N. Patel, Dr. R. K. Jain, and Dr. M.
V. Joshi), 2011: [5]
This gives improved solution for locating the fruits
on the plant based on multiple features. Multiple feature
extortion technique can include steps like extraction of color
Fig. 9 Gray Mold and intensity feature, extraction of orientation feature,
extraction of edge feature, extraction of area from feature
II. LITERATURE SURVEY maps. The process is entirely automatic and it can work
Image Processing for Smart Farming: Detection of Disease without user involvement. To improve output it considers
and Fruit Grading, Authors (Monica Jhuria, Ashwani Kumar, numerous features.
Rushikesh Borse), 2013: [1]
As there is a need of high yield in agricultural Tomato quality evaluation with image processing: A review,
industries improved yield of fruit is important, for this there is Authors(Abraham Gastélum-Barrios, Rafael A. Bórquez-
a need of automated technique which will find disease on López, Enrique Rico-García, Manuel Toledano-Ayala and
fruits. For this artificial neural network methodology is Genaro M. Soto-Zarazúa), 2011: [6]
suggested which can be helpful to categories fruit infection. All over the world there is excessive requirement for
K-Means clustering is applied to find diseased area on the fruit tomato. Therefore grade assessment of tomato is prime task
but it has disadvantage of sizable estimation load. It will using image processing it can be acquire. Worldwide study of
encourage agronomist to build better production and make tomato production is done to accomplish the target. It is useful
correct time to time judgment. to obtain tomato quality, good color, pattern, size and
composition. Instead of manual testing we can achieve fast
A Review of Image Processing For Pomegranate Disease and accurate testing in laboratories for tomato grading.
Detection, Authors (Manisha A. Bhange, Prof. H. A.
Hingoliwala), 2015: [4] Fast and Accurate Detection and Classification of Plant
The process suggests a solution for the recognition of Diseases, Authors (H. Al-Hiary, S. Bani-Ahmad, M. Reyalat,
pomegranate fruit disease. In this process, web based M. Braik, and Z. ALRahamneh), 2011: [7]
technique is applied to help non experts in identifying fruit Improved solution for automated diagnosis and
diseases which depends on the picture representing the grading of plant leaves disorder can be diagnose with help of
symptoms of the fruit. Farmers can take image of fruit disease K-Means Clustering procedure. It uses SGDM Matrix for Hue
and upload it on the system. After this farmers would be able Saturation. Also Otsu method is applied for masking pixels
to see if the fruit is affected by bacterial blight or not. based on certain threshold values. It uses color concur
technique for extracting features of leaf but it is unable
A Cost Effective Tomato Maturity Grading System using support huge complicated network structure.
Image Processing for Farmers, Authors (Sudhir Rao
Rupangadi, Ranjani B.S., Prathik Nagaraj, Varsha G Bhat),
2014: [2] III. PROPOSED SYSTEM
This system classifies ripeness of fruit based on its
color or texture. It involves current techniques mainly manual A. Introduction
inspection which leads to errorious classification; it results in The purpose of proposed system is to supervised the
economic losses due to inferior produce in the market chain. diseases on fruit and suggest alternate solution for healthy
The short comings are several methodologies which require yield and good productivity. Labeling of border pixel can be
highly expensive setups and complicated procedures; overall achieved by image segmentation this can be done by K-Means
accuracy is achieved up to 98%. clustering technique. Trained database of infected image has
been generated using Neural Network. Feature vectors such as
Adapted Approach for Fruit Disease Identification using image color, morphology, texture and structure of hole are
Images, Authors (Shiv Ram Dubey, Anand Singh Jalal): [3] applied for extracting features of each image and for diagnosis
This adaptive approach is validated on the basis of of disease morphology gives accurate result. SURF algorithm
experiments. The approach consist of steps and that are stated used as locator and descriptor for extracting the features.
as; first step is k-means clustering technique which is applied Using extracted features Scope of Interest can be calculated
for defect segmentation and second step involves some state of and extraction can be followed as its first step after which
art features that are extracted from segmented image and then refinement and analysis is done.
segmented image are classified into one of classes with the Family of Artificial Neural Network is inspired by
biological Neural Networks which is used to evaluate or

972 2015 International Conference on Green Computing and Internet of Things (ICGCIoT)
appraisal functions that depends upon huge number of inputs Sum of the primitive image inwards a rectangle which can be
and they are generally unknown. They are systems of calculated quickly using the integral image, requiring four
interdependent "neurons" and utilities from inputs for calculations at the junction of the rectangle.
computing and are having a potential of machine learning Algorithm is mainly divided as -
along with pattern recognition in adaptive nature. This is 1. Scope point Detector
convenient technique which reduces human effort and gives 2. Local surrounding descriptor
90% accurate result. 3. Matching.
For starting this process, initially non-uniform weights are
fixed and then training begins. Supervised and unsupervised Blob Analysis - Blob detection methods are intended at
are two methodologies used for training. Supervised training detecting scope of interest surrounded by digital image that
mechanism provides the network with the specific output varies in properties, such as color or brightness comparison
either by manually "grading" the network's performance or by with its vicinity regions. Region of an image in which some
providing the desired outputs accomplished by the inputs properties are approximately constant can be called as Blob.
while Individual training can be achieved by network that The basic scenario of the Blob Analysis solution consists of
takes inputs without external help. Supervised training the successive stages:
approach is used by bulk of networks whereas unsupervised 1. Extraction: It is primary step of image thresholding
training is applied to execute some initial characteristics on technique which inspects a region corresponding to single
inputs. Basically database server is used for comparison of object or objects.
extracted image with trained database which in turns diagnose 2. Refinement: Extracted region contain various kind of loud
and classify disease of fruits. sound due to degraded quality of image. Region
transformation techniques are used in refinement step.
3. Analysis: It is ultimate stage for refined region to evaluate
IV. METHODOLOGY & compute the outcome. If the region shows multiple objects
then divide it into separate blobs for inspection.
Image Acquisition - Image acquisition can be widely
described as the activity of restoring an image from some Pattern Matching - It is the procedure of examining stated
origin, usually a hardware-based source which can be proceed successions of tokens for the existence of the elements of
along with processes that need to appear afterward. Image some pattern. In proposed system ANN that is artificial neural
acquisition is consistently the initial condition for the work network concept is applied for pattern matching which interns
flow series of image processing because as processing is classifies disease.
possible only with the help of an image. The image obtained is Comparison between patterns can be accepted using
entirely natural and is the consequence of any hardware which artificial neural network. ANNs can be evaluated as Artificial
was handled to produce it. intelligence (AI) neural network structure.

Image Segmentation - It is the method for segregation of A. Architecture

digital image into several segments. The primary aim of
segmentation is to clarify and/or convert the rendering of an
image into something that is further relevant and easier for
analyses. Objects and bounding line of images are located by
using image segmentation.
Pixels with similar label portion share distinguishing features
for allocating a label to each pixel in an image. For this we are
using K-Means Clustering methodology.

Feature Extraction - Four feature vectors are considered

namely color, texture, morphology and structure of hole of the
fruits. For describing huge data set sometime enormous
resources are required.
Algorithm used for extracting the features is as follow:
SURF (Speed up Robust Feature) algorithm is applied for
extracting the features. SURF algorithm used as local
descriptor and blob detector.

Fig. 10 Architecture of Proposed System

(1)

2015 International Conference on Green Computing and Internet of Things (ICGCIoT) 973
B. Algorithm of Proposed System b) Morphology: Erosion concept is applied for acquiring
boundaries of all database images.
Input – Images of Various Fruits
Output – Detection of Fruit Disease Erosion= {Z/ (Y) z ‫ك‬X} (6)

Step 1: Accept image using android phone from user: Image Boundary= Original image - Eroded image (7)
(Color, Morphology, Texture, Structure of Hole) Where,
Step 2: Extraction of Feature Vectors X is erosion which indicates database images and Y
E (n) = [C (n) + M (n) + T (n) +H (n)] as input image which is set of each points Z such that Y
Here, converted by Z and contained in X in morphology. Entire
C = Color knowledge about structure is symbolized with the help
M = Morphology discrete cosine conversion considering few co-efficient.
T = Texture
H = Structure of Hole c) Texture: Visual patterns describe texture property, each
E = Extraction of features having similarity. Texture identification is done by modeling
n = No. of images textures as two-dimensional deviation of gray level.
Step 3: Calculating ROI:
Let E (n) be set of Extracted Images and ψab (x) =| a |(-½) ψ (x – b/a) (8)
If<Fruit Detected> Then,
Then
E (n)
Else (9)
Reject Where,
Ψ(x) = mother wavelet
Step 4: Pattern Matching ψab(x) = daughter wavelet
Let T be set Trained Database
If<E (n) = = T>
Then V. CONCLUSION
Classification The innovative outcome suggests that the advanced
Detection approach is a worth, which can distinctly support an accurate
Else diagnosis of fruit diseases in a minor computational effort. It
Go To Step (2) also dedicates future study on automatically estimating the
Step 5: Stop. severity of the disease.
C. Mathematical Model An image processing derived solution is proposed for
Feature Extraction of Image Classification: Four feature detection of grape, apple and pomegranate fruit disease. For
vectors such as color, morphology, texture and structure of Grape -Black Rot, Powdery Mildew, Downy Mildew; For
hole are used as learning database images for extracting the Apple -Apple Scab, Apple Rot, Apple Blotch; For
features. Pomegranate -Bacterial Blight, Aspergillus Fruit Rot, Gray
a) Color: It is the most valuable properties used by human for Mold diseases are detected and classified. Once diseases are
object discrimination. As RGB color space is affected by light detected proper treatments are suggested accordingly.
and angle of image which has been captured so there is need
for conversion into HSI color space. It would also promote Indian Farmers to do smart farming
which helps to take time to time decisions which also save
time and reduce loss of fruit due to diseases. The leading
(2) objective of our paper is to enhance the value of automatic
fruit disease detection.
Here,
References
(3) [1] Monica Jhuria, Ashwini Kumar, Rushikesh Borse
“Image Processing for Smart Farming: Detection of
Disease and Fruit Grading” Proceeding of the 2013
IEEE Second International Conference on Image
Processing.
(4) [2] Sudhir Rao Rupanagudi, Ranjani B.S., Prathik
Nagaraj ,Varsha G. Bhat “A Cost Effective Tomato
(5) Maturity Grading System using Image Processing for

974 2015 International Conference on Green Computing and Internet of Things (ICGCIoT)
 Farmers” International Conference on Contemporary [12] Savita N. Ghaiwat, Parul Arora “Detection and
Computing and Information ,2014. Classification of Plant Leaf Diseases Using Image
[3] Shiv Ram Dubey, Anand Singh Jalal “Adapted processing Techniques: A Review” International
Approach for Fruit Disease Identification using Journal of Recent Advances in Engineering &
Images”. Technology (IJRAET) ISSN (Online): 2347 - 2812,
[4] Manisha A. Bhange, Prof. H. A. Hingoliwala “A Volume-2, Issue - 3, 2014.
Review of Image Processing for Pomegranate [13] Anand H. Kulkarni, Ashwin Patil R. K. “Applying
Disease Detection” International Journal of Computer image processing technique to detect plant diseases”
Science and Information Technologies, Vol. 6 (1), International Journal of Modern Engineering
2015, 92-94. Research (IJMER) Vol.2, Issue.5, Sep-Oct. 2012 pp-
[5] Hetal N. Patel, Dr. M. V. Joshi “Fruit Detection using 3661-3664 .
Improved Multiple Features based Algorithm” [14] Pradnya Ravindra Narvekar, Mahesh Manik
International Journal of Computer Applications (0975 Kumbhar2, S. N. Patil “Grape Leaf Diseases
– 8887), Volume 13– No.2, January 2011. Detection & Analysis using SGDM Matrix Method”
[6] Abraham Gastélum-Barrios, Rafael A. Borquez- International Journal of Innovative Research in
López, Enrique Rico-García, Manuel Toledano- Computer and Communication Engineering (An ISO
Ayala and Genaro M. Soto-Zarazúa* “Tomato 3237:2007 certified organization ) Vol.2,Issue 3,
Quality Evaluation with Image processing: A review” March 2014.
African Journal of Agricultural Research Vol. 6(14), [15] Tejal Deshpande, Sharmila Sengupta, K. S.
pp. 3333-3339, 18 July, 2011. Raghuvanshi “Grading & Identification of Disease in
[7] H. Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik Pomegranate Leaf and Fruit” International Journal of
and Z. ALRahamneh “Fast and Accurate Detection Computer Science and Information Technologies,
and Classification of Plant Diseases” International Vol. 5 (3), 2014, 4638-4645.
Journal of Computer Applications (0975 – 8887) [16] Vinita Tajane, Prof. N.J. Janwe “Medicinal Plants
Volume 17– No.1, March 2011. Disease Identification Using Canny Edge Detection
[8] P. Vimala Devi and K. Vijayarekha “Machine Vision Algorithm, Histogram Analysis and CBIR”
Application to Locate Fruits, Detect Defects and International Journal of Advanced Research in
Remove Noise: A Review” Vol.7 | No.1 | 104-113| Computer Science and Software Engineering,
January – March | 2014. Volume 4, Issue 6, June 2014.
[9] Shiv Ram Dubey, A. S. Jalal “Detection and [17] Jayamala K. Patil , Raj Kumar “Advances in Image
Classification of Apple Fruit Diseases using Processing for Detection of Plant Disease” Journal of
Complete Local Binary Patterns”. Advanced Bioinformatics Applications and Research
[10] Rashmi Pandey, Sapan Naik, Roma Marfatia “Image ISSN 0976-2604 Vol. 2, Issue 2, June-2011, pp 135-
Processing and Machine Learning for Automated 141.
Fruit Grading System: A Technical Review” [18] Shiv Ram Dubey, Pushkar Dixit, Nishant Singh, Jay
International Journal of Computer Applications (0975 Prakash Gupta “Infected Fruit Part Detection using
– 8887) Volume 81 – No 16, November 2013. K-Means Clustering Segmentation Technique”
[11] Anshuka Srivastava, Swapnil Kumar Sharma International Journal of Artificial Intelligence and
“Development of a Robotic Navigator to Assist the Interactive Multimedia, Vol. 2, 2013.
Farmer in Field” Proceeding of the International
Multi Conference of Engineers and Computer
Scientists 2010 Vol. (2) IMECS 2010 March 17-19,
Hong Kong.

2015 International Conference on Green Computing and Internet of Things (ICGCIoT) 975