MediLeaf: A Herbal Plant Classification System

Kim Brian Mutia Raymond Balcita Ghinecel Carza Sarmiento

Cavite State University Cavite State University Cavite State University
CCAT Campus CCAT Campus CCAT Campus
[email protected] [email protected] [email protected]

ABSTRACT Statement of the Problem

In the Philippines, plants play significant roles in the lives of Nowadays many people did not know how to identify or
early Filipinos since some plants were already using as medicine classify herbal plants and aware of using it. It`s difficult for
as far back as the beginning of humankind. However, this them to identify leaf what kind of plant is they need to use to
generation gradually loses the knowledge of traditional uses and their illness. Some people going to the library to research
importance of the medicinal plant. In addition, some people were about the plant.
not that knowledgeable to recognize a plant and their uses. Thus, Objective of the Study
this study focuses on creating a system that will able to classify a
herbal plant according to its uses and property. This paper The general objective of this study is to design and create a
presents an effective way to develop PC based software that gives system for herbal plant classification using computer vision.
user the ability to identify plant species based on photographs of It specifically aims to:
the plant's leaf taken with mobile phone and any type of camera. 1. Classify herbal plants with the use of Artificial
The algorithm acquires a color image of the leaf using a camera Neural Network Algorithm and Multilayer
and performs the classification by comparing the acquired image Processing;
with the stored database image. The method proposed in the paper 2. Identify the best learning algorithm to be used in
performs preprocessing to obtain the binary contour of the leaf, classification herbal plants; and
then morphological features of the leaf are obtain to generate 3. Test the accuracy of computer vision system in
feature vector of the leaf, then classifies the species based on the classifying herbal plants.
combination of featured vector and gives it uses and property.
Scope and Limitation
INTRODUCTION The scope of the system is that it can run to the different type
Nowadays, some people are aware of using herbal plants and of operating systems like Windows and Linux. The system can
some are not. They have difficulties to identify what kind of use 24/7 and easy to use. The system can help people to classify
herbal plant they need to their illness and it is hard for them to and identify what are they need to their illness and it can give
classify what is the species of that herbal plant. Sometimes people accurate information. This system cannot run to the mac operating
were relying with what their relatives or elders say about using the system. It can use only when you have a photograph of a leaf. The
herbal plant without knowing the property and the proper use of system could not run using any mobile devices. It can only
the given plant. This leads to wrong medication and may continue process 1 piece of leaf on the image and those that are taken from
to serious complications. afar will not be accept by the system. The system will recognized
photographs with close proximity or High definition on
In this generation, in order to identify the species and uses of resolution of the image. The system can only accept images or
herbal plant people will asked with the elder one but the accuracy photograph as input.
of the answer is not that reliable these days because they have no
scientific basis that the herbal plant being asked is suitable with
the illness given. Other current solution is that people will look on Time and Place of the Study
the internet and search for the uses of that given herbal plant but The researchers will conducted at Navarro, General Trias,
not all the information given on the internet is consistent and can Cavite in February 14, 2018
be trusted.
Definition of Terms
With this, the study aimed to create a PC based system that Acquisition: An asset or object bought or obtained, typically by
will give the property and uses of a given herbal plant through a library or museum.
photograph. This is to give reliable source and accurate Herbal Plant: A plants that is use to relieve or cure some
information about it. This will also help the user to identify the illness.
plant without pressure at all. This could serve also as a guide and Recognition: The action or process of recognizing or being
learning tool to people who are quite not aware of herbal plants. recognized, in particular.
Species: A group of animals or plants that are similar and can
produce young animals or plants.
Review of Related Literature
Use of artificial neural networks in colour measurements
The most popular class of multilayer feed-forward ANNs is the
multilayer perceptron or multilayer processing(MLP).

A feed-forward neural network is biologically inspired

classification algorithm. It consists of a number of simple neuron-
like processing units, organized in layers and every unit in a layer
is connected with all the units in the previous layer. These
connections are not all equal, as each connection may have
different strength. The weights on these connections encode the
knowledge of a network. Often the units in a neural network are
called nodes.

Computer vision techniques have been applied to solve plant

classification tasks at different levels:
When considering only leaves, methods based on shape, color
and texture have proven effective to discriminate between
different types of leaves. Beghin et al. classify leaves from 10
species based on shape and texture with an average accuracy of
85%. Kumar et al. developed a smartphone application for
classification of leaves from trees in the Northeastern United
States. These approaches share with other work that the input is an
image of a flat leaf captured on a mostly homogeneous
background. In agriculture machine vision can be applied with the
goal of intelligent weed control, but in general the requirement of
single leaf images is not applicable to commercial field situations.
Remote-sensing has been successfully applied to estimate weed Figure 1. Flowchat of MediLeaf
densities and distributions on field level with the goal to regulate
herbicide usage. But for precision agricultural activities that are Image Acquisition
considered here, plant classification must be done on a much finer Input data consists of images captured by smartphone camera.
scale. The camera should be high in resolution to . The camera system is
On ground level, camera based sensing can been applied to positioned approximately 45 cm above the leaf. All images used
identify single plants or groups of a few plants and to classify for training and evaluation of our system were collected.
them. Hemming et al. present a robot and computer vision system
which correctly classifies 51 to 95 % of plants based on color and Image Enhancement
shape features of segmented plants in top down images. They Image should be in a gray scale and gaussian blur for system
conclude that segmentation into individual plants is difficult and can remove the noise in the image and perform the thresholding
needs more research. Astrand & Baerveldt use similar features on using gaussian/otsu and binary inverted to turn the background
segmented plants for classification. They evaluate their system in into white.
greenhouse experiments with large plants (about 5 cm diameter)
but do not present quantitative results on classifi- cation Clear excess noise in the image using Closing is reverse of
performance. Leaf segmentation has also been researched in field Opening, Dilation followed by Erosion. It is useful in closing
settings to enable crop / weed discrimination. Neto et al. present a small holes inside the foreground objects, or small black points on
leaf segmentation technique that performs well on convex leaves the object.
but not for other leaf shapes (like carrots considered here).
After removing the noise in the image and turn in thresholding
Methodology next we find the object on the image using contour in opencv and
The herbal plant classification system is designed as a 6 stage to get the different features of contour like area, perimeter,
process to perform the classifying of herbal plants. In following centroid, bounding box etc. and after that we should crop the
we address and explain each step depth. image.

Feature Extraction
This function could get the important matter including height,
width, and color texture to train that feature with the leaf stored in
the database.

2
Classification C. M. Onyango and J. Marchant. Segmentation of row crop
This is the main function of the system. It includes the plants from weeds using colour and morphology. Computers
Multilayer Perceptron to classify a herbal plant. Multilayer and electronics in agriculture, 39(3):141–155, 2003.
Perceptron is composed of more than one perceptron. They are
composed of an input layer to receive the signal, an output layer J. C. Neto, G. E. Meyer, and D. D. Jones. Individual leaf
that makes a decision or prediction about the input, and in extractions from young canopy images using gustafson–
between those two, an arbitrary number of hidden layers that are kessel clustering and a genetic algorithm. Computers and
the true computational engine of the MLP. MLPs with one hidden Electronics in Agriculture, 51(1):66–85, 2006.
layer are capable of approximating any continuous function.

Displaying Result
We checked the accuracy that it matches the actual image or
not for the database of 300 leaves of different 3 plants species
taken, which are completely different in their shape, colour.

Result and Discussion

We test the accuracy of this software. This software give
accurate result that range 80% to 100% .
We test the accessibility of this software if this easy to use.
We therefore conclude that this software can help many people
to identify herbal plant in a fast way and help botanist of their
research and to the developer who create same of this software.

References
T. Beghin, J. S. Cope, P. Remagnino, and S. Barman. Shape
and texture based plant leaf classification. In Advanced
Concepts for Intelligent Vision Systems, pages 345–353.
Springer, 2010.

N. Kumar, P. N. Belhumeur, A. Biswas, D. W. Jacobs, W. J.

Kress, I. C. Lopez, and J. V. Soares. Leafsnap: A computer
vision system for automatic plant species identification. In
Computer Vision–ECCV 2012, pages 502–516. Springer,
2012.

J.-X. Du, X.-F. Wang, and G.-J. Zhang. Leaf shape based
plant species recognition. Applied mathematics and
computation, 185(2):883–893, 2007.

P. F. Felzenszwalb and D. P. Huttenlocher. Efficient belief

propagation for early vision. International journal of
computer vision, 70(1):41–54, 2006.

K. Thorp and L. Tian. A review on remote sensing of weeds

in agriculture. Precision Agriculture, 5(5):477–508, 2004.
J. Hemming and T. Rath. Computer-vision-based weed
identification under field conditions using controlled
lighting. Journal of Agricultural Engineering Research,
78(3):233– 243, 2001

J. M. Guerrero, G. Pajares, M. Montalvo, J. Romeo, and M.

Guijarro. Support vector machines for crop/weeds
identification in maize fields. Expert Systems with
Applications, 39(12):11149–11155, 2012.