Introducing
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
Weed identification Using Deep Learning and image Processing in Vegetables Plantation
36
Weed Identification Using Deep Learning and Image Processing in Vegetable Plantation
students :
Ghada Fathi Aziz
Jomana Ali Sleebek
Alaa Adel Alhaddad
Outlines
outlines
- Abstract
- Introduction
- Procedural steps of the proposed method
- Vegetable detection using Deep Learning
- CenterNet and Predictive ability performance
- Weed Identification Utilizing Image
Processing
-Conclusion
Abstract
- Weed identification in vegetable plantation is more challenging than crop weed identification
due to their random plant spacing
- Traditional methods of crop weed identification used to be mainly focused on identifying weed
directly
- This paper proposes a new method combines deep learning and image processing technology.
Abstract
Introduction
Introduction
- Vegetable is considered one of the most nutrient-dense food
all around the world due to its sufficient vitamins, minerals
and antioxidants
why weed is harmful ?
- the yield of vegetable decreased by 45% - 95% in the case of
weed-vegetable competition
- Excessive use of chemical herbicides results in over-application in areas of low or no weed infestation
- Thus, hand weeding is still the primary option for weed control in vegetable plantation at present
Introduction (cont.)
What is the main objective ?
- firstly identify and segment the vegetable using deep learning
then the remaining green objects in the segmented image were
considered as weeds
- This strategy can largely reduce the complexity of weed
detection, thereby enhancing the weed identification
performance and accuracy.
Introduction (cont.)
Procedural steps
of the proposed method
The approach proposed for the identification of weeds is composed of two stages :
The first stage consists of the state-of-art CenterNet algorithm for detecting bok choy.
Images of bok choy are collected and used as input data for training the neural network. The trained neural network is used for detecting bok choy and drawing bounding boxes around them
Procedural steps of the proposed method
Procedural steps
of the proposed method (cont.)
In the second stage, the remaining green objects falling out of bounding boxes were considered as weeds and to extract them from background a color index-based segmentation is performed using color information
Flow diagram for proposed weed detection methodology
Flow diagram
VEGETABLE DETECTION USING DEEP LEARNING
VEGETABLE DETECTION USING DEEP LEARNING
1) IMAGE AUGMENTATION
The training data set contained 1150 images, these images were then expanded to 11500 images.
The collected images were pre-processed in terms of color, brightness, rotation, and image definition.
2) IMAGE ANNOTATION
Manual annotation was applied by drawing bounding boxes onto the vegetable (CenterNet Algorithm).
CenterNet
CenterNet model is a cutting edge and brand-new
object detector, which is anchor-free and depends on the key points estimation.
It is based on the insight that box predictions can be sorted for relevance based on the location of their centers, rather than their overlap with the object
CenterNet
PERFORMANCE OF THE VEGETABLE DETECTION
For object detection task, the results of testing can be divided into four types, including :
true positive (TP): represents the bounding boxes containing target vegetables that are correctly identified
true negative (TN): indicates target vegetables identified as vegetables but no bounding boxes were drew
false positive (FP):means the bounding boxes without target
vegetables that are incorrectly identified as target
vegetables
false negative (FN):indicates target vegetables not
identified as vegetables and no bounding boxes
were drew
performance indications of predictive ability
The precision, recall and F1 score are used as the performance indices of predictive ability. Precision and recall are defined as follows:
predictive ability performance
Introduction
Once the vegetable was found, the remaining green objects falling out of the bounding boxes were marked as weeds.
To extract weeds from other elements of the scene (i.e. soil, straws, stones, and other residues), color index-based segmentation using a binary-coded genetic algorithms (GAs)
identifying weed in RGB color space for the outdoor field conditions was studied and implemented.
WEED IDENTIFICATION UTILIZING IMAGE
PROCESSING
-Introduction
-The three parameters x, y and z are in the range
(−255, 255), while T is limited to [0, 1024) [26]. This will lead to 510 × 510 × 510 × 1024 possible combinations.
problem!
an efficient searching algorithm is necessary to solve
this problem. GAs behave well in exploiting accumulated information of an initially unknown domain in a highly efficient way.
Therefore, GA was selected to design a search
engine in this work.
1) COLOR INDEX-BASED SEGMENTATION
Introduction (cont.)
Segmentation refers to the process of finding a
plane that intersect the RGB color cube, thereby classifying
image into vegetation and non-vegetation pixels.
The plane is defined by the equation:
xR + yG + zB = T
2) GENETIC ALGORITHMS
Introduction (cont.)
-Introduction
-The three parameters x, y and z are in the range
(−255, 255), while T is limited to [0, 1024]. This will lead to 510 × 510 × 510 × 1024 possible combinations.
problem!
an efficient searching algorithm is necessary to solve
this problem. GAs behave well in exploiting accumulated information of an initially unknown domain in a highly efficient way.
Therefore, GA was selected to design a search
engine in this work.
Procedures of a genetic algorithm is shown in Figure. To start the algorithm, an initial population was generated randomly
3) CHROMOSOME
An 88-bit binary string, or a chromosome, was used to represent parameters in Equation that encoded by permutation method.
4) POPULATION SIZE
The basic element of a GA is called individual, which is characterized by a set of parameters (variables) known as Genes.
In this work,
a population size of 200 was used to generate color index parameters.
5) SELECTION
Roulette wheel selection, which selects
the individual with the highest fitness randomly picked individuals, was chosen and implemented in GA
6) CROSSOVER AND MUTATION
-A crossover probability of 0.8 was selected
-The mutation rate of 0.2 was used in this application.
7) EVALUATION
Bayesian classification error (BCE) was applied for function
evaluation. For each given color, BCE (r, g, b) was defined
as:
CONT.
where p1 (r, g, b) and p2 (r, g, b) are the distribution probabilities of weed and background in the color space, respectively
where Cw is the number of occurrences of color (r, g, b) in
weed pixels. Cb is the number of occurrences of color (r, g,
b) in background pixels. Ct
is the total number of pixels in
the reference images.
Obviously, the theoretical minimum value of BCE (r, g, b)
is:
8) STOPPING CRITERION
The algorithm terminates whenever one of the two conditions was satisfied:
(a) if the iterations number reached 2000.
(b) if the best fitness value which is equal to the predefined threshold (theoretical minimum error) of acceptance
was located.
Conclusion
we proposed an approach to identify weeds in
vegetable plantation using deep learning and image processing.
The algorithm was depicted in two steps. A CenterNet model was trained to detect vegetables. The trained
CenterNet achieved a precision of 95%
Then the remaining green objects in the color image were considered
as weeds. To extract weeds from the background, a color index was determined and evaluated through Genetic algorithms (GAs) .
In this way, the model focuses on identifying only the vegetables and thus avoid handling various weed species.
CONCLUSION
Conclusion (cont.)
Thank You!
END
To Students Have any question?
GOOGLE IT !
Reference
Paper of Weed Identification using Deep Learning and Image Processing in Vegetable Plantation
-College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjing 210037 China
-Corresponding author: Yong Chen
(e-mail: chenyongjsnj@163.com).