ISSN (online) 2583-455X

BOHR International Journal of Computer Science

2022, Vol. 2, No. 1, pp. 36–39
https://doi.org/10.54646/bijcs.008
www.bohrpub.com

Weed Detection Using Convolutional Neural Network

M. S. Hema∗ , V. Abhilash, V. Tharun and D. Madukar Reddy

Department of Information Technology, Anurag University Hyderabad, Hyderabad, India

∗ Corresponding author: [email protected]

Abstract. Precision agriculture relies heavily on information technology, which also aids agronomists in their work.
Weeds usually grow alongside crops, reducing the production of that crop. They are controlled by herbicides. The
pesticide may harm the crop as well if the type of weed is not identified. To control weeds on farms, it is required
to identify and classify them. A convolutional network or CNN, a deep learning-based computer vision technology,
is used to evaluate images. A methodology is proposed to detect weeds using convolutional neural networks. There
were two primary phases in this proposed methodology. The first phase is image collection and labeling, in which
the features for images to be labeled for the base images are extracted. In the second phase, the convolutional
neural network model is constructed by 20 layers to detect the weed. CNN architecture has three layers, namely,
the convolutional layer, the pooling layer, and the dense layer. The input image is given to a convolutional layer
to extract the features from the image. The features are given to the pooling layer to compress the image to reduce
the computational complexity. The dense layer is used for final classification. The performance of the proposed
methodology is assessed using agricultural dataset images taken from the Kaggle database.
Keywords: CNN, weed, precision agriculture, segmentation.

INTRODUCTION come in handy in finding the correct classification and

projection.
Farmers’ main issue is detecting weeds in the crop during
the irrigation process. Manual work for detecting the crop
and weeds takes a long time, and more human effort is LITERATURE SURVEY
required to complete this procedure. Weed identification
in plants has become more challenging in recent years. So An object-oriented algorithm is proposed to detect weeds
far, there have not been many efforts put into identifying in the agricultural field [1]. A comprehensive and critical
weeds while growing crops. Traditional methods for iden- survey on image-based plant segmentation techniques is
tifying agricultural weeds focused on directly identifying presented. In this context, “segmentation” refers to the pro-
the weed; nevertheless, there are substantial differences cess of classifying an image into the plant and non-plant
in weed species. In contrast to this method, this study pixels [2]. Non-chemical weed control is important both
proposes a revolutionary technique that combines deep for the organic production of vegetables and for achiev-
learning with imaging technology. The dataset was first ing ecologically sustainable weed management. Estimates
trained using the CNN model. We can categorize and have shown that the yield of vegetables may be decreased
predict the provided input image as a weed or crop once by 45%–95% in the case of weed–vegetable competition.
the training is completed. The major goal of this project Non-chemical weed control in vegetables is desired for
is to use the CNN deep learning algorithm to detect and several reasons [3]. The mixer is used to spray pesticides to
classify weeds and crops. alleviate weeds from crops [4]. Robot technology is used to
To solve this challenge, we are developing our project find weeds, and pesticides are used to alleviate weeds [5].
to detect weeds utilizing the CNN method. Our findings A comprehensive review of research dedicated to applica-
will aid in crop and weed classification, saving humans tions of machine learning in agricultural production sys-
time and effort. Manual labor may overlook proper tems is presented. The analyzed works were categorized
classification on occasion, which is where our project will as follows: (a) crop management, including applications

36
Weed Detection Using Convolutional Neural Network 37

on yield prediction, disease detection, weed detection crop the growth of crops. By combining image processing and
quality, and species recognition; (b) livestock management, CNN in drones, they get different accuracies depending
including applications on animal welfare and livestock on the processing, which is from 98.8% with CNN to 85%
production; (c) water management; and (d) soil manage- using Histograms of Oriented Gradients (HOG) [21]. The
ment [6]. ML and DL techniques have been used for weed authors combined Hough transform with simple linear
detection and recognition and thus for weed management. iterative clustering (SLIC). This method focuses on the
In 2018, Kamilaris and Prenafeta-Bold’u (2018) published detection of crop lines [22]. A threshold based on the
a survey of 40 research papers that applied DL techniques classification values of the area for a crop or a weed is
to address various agricultural problems, including weed proposed [23].
detection. The study reported that DL techniques out-
performed more than traditional image processing meth- PROPOSED METHODOLOGY
ods [7]. It discussed ten components that are essential and
possible obstructions to developing a fully autonomous The main objective of the proposed methodology is to
mechanical weed management system [8]. The authors detect weeds. The convolutional neural network is pro-
focused on different machine vision and image process- posed for weed detection. The architecture of the proposed
ing techniques used for ground-based weed detection [9]. methodology is shown in Figure 1.
Fernández-Quintanilla et al. (2018) reviewed technologies The convolution layer is used to extract the features
that can be used to monitor weeds in crops. They explored from the image. The rectified linear unit (ReLU) activation
different remotely sensed and ground-based weed mon- function is used in the convolutional layer. The ReLU
itoring systems in agricultural fields. They reported that helps to break up the linearity even further, compensat-
weed monitoring is essential for weed management. They ing for any linearity that may be imposed on an image
foresaw that the data collected using different sensors during the convolution process. As a result, ReLU aids
could be stored in cloud systems for timely use in relevant in avoiding the exponential growth of the computation
contexts [10]. They used the VGG-16 model for classifying required to run the neural network. As the size of the
crop plants and weeds. They also trained the model with CNN rises, the computational cost of adding more ReLUs
one dataset containing sunflower crops and evaluated it grows linearly. Another nonlinear activation function that
with two different datasets with carrot and sugar beet has gained prominence in the deep learning sector is the
crops [11]. The author describes in “Weed detection using ReLU function. The key benefit of employing the ReLU
image processing” how they can detect and separate weed- function over other activation functions is that it does not
affected areas from the crop plants using image process- simultaneously stimulate all of the neurons.
ing [12]. A methodology is proposed to detect weeds using The feature maps’ dimensions are reduced by using
image processing techniques. The properties are extracted pooling layers. As a result, the number of parameters to
from the image and weed is detected from the extracted
features [13]. Machine vision uses unique image processing
techniques. Weeds in agricultural fields have been detected
by their properties such as size, shape, spectral reflectance,
and texture features [14]. Two methods are proposed for
weed detection: crop row detection in images from agri-
culture fields with high weed difficulty and to further
differentiate between weed and crop [15]. The author
proposed in “Crop and weed detection based on texture
and size features and automatic spraying of herbicides”
how they developed the image processing algorithm for
yield finding and management of weeds [16]. A computer
vision application to detect unwanted weeds in early-
stage crops is proposed [17]. A novel approach for weed
classification using curvelet transform and Tamura texture
feature (CTTTF) with RVM classification methodology is
proposed [18]. Robots are working collaboratively with
humans and learning from them how to realize basic agri-
culture tasks such as weed detection, watering, or seeding
(Marinoudi, et al., 2019) [19]. Weed detection using deep
learning has been proposed [20]. It is proposed that the
implementation of image processing in drones instead of
robots so that they not only detect weeds but also monitor Figure 1. Weed detection architecture.
38 M. S. Hema et al.

learn and the amount of processing in the network are CONCLUSION

both reduced. The pooling layer summarizes the features
found in a specific region of the feature map created by Using the CNN model of deep learning, a system that can
the convolution layer. The pooling layer is a crucial layer classify weeds and crops was implemented. The features
that performs down-sampling on the feature maps from are extracted from the input images using a convolutional
the previous layer, resulting in new feature maps with layer. A pooling layer is used to downsize the image.
a reduced resolution. This layer significantly decreases Finally, the dense layer is used for classification. Further-
the input’s spatial dimension. The output from the final more, it can be extended in the future to help in detecting
pooling or convolutional layer, which is flattened and then weeds from large crops or plants and can be improved
fed into the fully connected layer, is the input to the to make work with more types of crops and weeds for
fully connected layer. In the neural network models that accurate classification and to reduce human efforts. It will
predict a multinomial probability distribution, the Softmax be easier to find any weed or crop with less human effort.
function is utilized as the activation function in the output
layer. Softmax is used as the activation function for multi-
class classification problems requiring class membership
REFERENCES
on more than two labels. CNNs are trained to recognize
and extract the best features from images that are relevant [1] T. W. Berge, A. H. Aastveit, and H. Fykse, “Evaluation of an
to the problem at hand. Their biggest advantage is this. algorithm for automatic detection of broad-leaved weeds in spring
Because of their effectiveness as a classifier, CNN’s last lay- cereals,” Precis. Agricult, vol. 9, no. 6, pp. 391–405, Dec. 2008.
ers are fully connected. Because CNNs include FC layers, [2] E. Hamuda, M. Glavin, and E. Jones, “A survey of image processing
these two architectures are not competing as much as you techniques for plant extraction and segmentation in the field,” Com-
put. Electron. Agricult. vol. 125, pp. 184–199, Jul. 2016.
might imagine. Finally, we will finish things up and give a
[3] H. Mennan, K. Jabran, B. H. Zandstra, and F. Pala, “Non-chemical
quick overview of the section’s main concepts. weed management in vegetables by using cover crops: A review,”
Agronomy, vol. 10, no. 2, p. 257, Feb. 2020.
[4] X. Dai, Y. Xu, J. Zheng, and H. Song, “Analysis of the variability
of pesticide concentration downstream of inline mixers for direct
nozzle injection systems,” Biosyst. Eng., vol. 180, pp. 59–69, Apr.
RESULTS AND DISCUSSION 2019.
[5] D. C. Slaughter, D. K. Giles, and D. Downey, “Autonomous robotic
The agriculture image is taken from the Kaggle database. It weed control systems: A review,” Comput. Electron. Agricult, vol. 61,
consists of images of 1,000 weeds and crops. Out of which, no. 1, pp. 63–78, Apr. 2008.
80% of the images are used for training and 20% of the [6] K. Liakos, P. Busato, D. Moshou, S. Pearson, and D. Bochtis,
“Machine learning in agriculture: A review,” Sensors, vol. 18, no. 8,
images are used for validation.
p. 2674, Aug. 2018
Figure 2 shows the performance of the proposed [7] Kamilaris, A. and Prenafeta-Boldú, F. X. (2018). Deep learning in
approach. The accuracy of the training set improves when agriculture: A survey. Computers and electronics in agriculture, 147,
the number of epochs increases. When the number of 70–90.
epochs is less, the validation set accuracy is high; but, the [8] Merfield, C. N. (2016). Robotic weeding’s false dawn? Ten require-
accuracy increases when the number of epochs increases. ments for fully autonomous mechanical weed management. Weed
Research, 56(5), 340–344. https://doi.org/10.1111/wre.12217
[9] Wang, A., Zhang, W. and Wei, X. (2019). A review on weed detection
using ground-based machine vision and image processing tech-
niques. Computers and Electronics in Agriculture.
[10] Fawakherji, M., Youssef, A., Bloisi, D., Pretto, A. and Nardi, D.
(2019). Crop and weeds classification for precision agriculture using
context-independent pixel-wise segmentation. 2019 Third IEEE
International Conference on Robotic Computing (IRC), 146–152.
[11] Fernández-Quintanilla, C., Peñna, J., Andújar, D., Dorado, J., Ribeiro,
A. and López-Granados, F. (2018). Is the current state of the art
of weed monitoring suitable for site-specific weed management in
arable crops? Weed research, 58(4), 259–272. d electronics in agricul-
ture, 158, 226–240.
[12] Ajinkya Paikekari, Vrushali Ghule, et al., “Weed detection using
image processing” International Research Journal of Engineering
and Technology (March 2016).
[13] Riya Desai, et al., “Removal of weeds using Image Processing”
International Journal of Advanced Computer Technology (IJACT)
(2016).
[14] Ashintosh K Shinde, et al., “Crop detection by machine vision
for weed management” International Journal of Advances in
Figure 2. Accuracy of training and validation dataset. Engineering & Technology (July, 2014).
Weed Detection Using Convolutional Neural Network 39

[15] A Satish kumar, et al., “Detection of weeds in a crop row using image [21] Daman, M., Aravind, R. and Kariyappa, B., 2015. Design and Devel-
processing” Imperial Journal of Interdisciplinary Research (2016). opment of Automatic Weed Detection and Smart Herbicide Sprayer
[16] Amruta A. Aware, “Crop and weed detection based on texture and Robot. IEEE Recent Advances in Intelligent Computational Systems
size features and automatic spraying of herbicides” International (RAICS), pp. 257–261.
Journal of Advanced Research (2016). [22] Liang, W.-C., Yang, Y.-J. and Chao, C.-M., 2019. Low-Cost Weed
[17] Batriz Nathlia, et al., “A computer vision application to detect Identification System Using Drones. Candarw, Volume 1, pp. 260–
unwanted weed in early stage crops” WSEAS (2016). 263.
[18] P. Prema, “A Novel approach for weed classification using curvelet [23] Bah, M. D., Hafiane, A. and Canals, R., 2017. Weeds detection in
transform and tamura texture feature (CTTTF) with RVM classifica- UAV imagery using SLIC and the hough transform. 2017 Seventh
tion” International Journal of Applied Engineering Research (2016). International Conference on Image Processing Theory, Tools and
[19] Marinoudi, V., Sorensen, C., Pearson, S. and Bochtis, D., 2019. Applications (IPTA), pp. 1–6.
Robotics and labour in agriculture. A context consideration. Biosys-
tems Engineering, pp. 111–121.
[20] Dankhara, F., Patel, K. and Doshi, N., 2019. Analysis of robust weed
detection techniques based on the Internet of Things (IoT). Coimbra,
Portugal, Elsevier B.V, pp. 696–701.