Page 75 - IJEEE-2023-Vol19-ISSUE-1

Abdulla & Marhoon | 71

production. So our objective in this paper was Create a sensors using the web application and printing all sensor
mobile application that classifies tomato diseases based on a values on the mobile application. Kwok & Sun [21] created
deep learning Convolution Neural Network (CNN). And an irrigation system based on plant type recognition from
designing a monitoring system on mobile based on IoT to deep learning, using an algorithm previously trained with
remotely monitor the climatic conditions surrounding the ImageNet. As for building the application that determines the
plant and irrigate plants. Therefore, The paper proposes a type of plant, it was through Docker and Android Studio.
monitoring system via IoT to remotely observe the state of Each plant has a certain percentage of moisture based on
plants regarding the environmental conditions and health which the valve is opened and closed to reach the humidity
status and accordingly the farmer can take action so the limit. Jacob et al.[22] created a system for monitoring and
farmer can discover diseases, reducing the field visit to the classifying plant diseases using sensors and deep learning.
farm because monitoring and watering are done remotely, Where temperature, soil moisture, co2, and smoke sensors
thus improving the quality of the agricultural product. The were used, these values are sent to the application, analyzed,
organization of this paper arranged according to the and alerts are given to the farmer. In detecting diseases, the
following: Inceptionv3 model was used to train the data to discover
I the introduction of this paper. II related works. III the plant diseases, where the model's accuracy in the test was
proposed system for this paper. IV discussion for the result 74.4.
obtained from this work with previous work. V conclusion.
III. THE PROPOSED SYSTEM
II. RELATED WORKS This work established a monitoring system for the tomato
plant agriculture environment in two phases. The first is to
A few researchers have developed applications for remote monitor the plant's health status and detect the types of
plant monitoring and disease detection. Uzhinskiy et al.[15] diseases that affect the tomato plant through deep learning
Using the KNN (K-Nearest Neighbor) classifier to classify models and building an application that classifies tomato leaf
15 types of diseases, they obtained a test accuracy of 86%. diseases. The second is building an IoT system for
To improve the accuracy, they used single-layer perceptron monitoring and control by using the sensors to measure the
estimators with a single input and output layer ending with environmental factors surrounding the plant and irrigation
softmax and Adam (Adaptive Moment Estimation) remotely and using the application to monitor all these
optimizer, obtaining an accuracy of 95.71% for 100 epochs sensors' data. Fig.1 shows the general structure of the
then used Apache Cordova to build Classified mobile proposed system.
application but training of the network was entirely based on
the 935 images collected from the Internet. Valdoria et A. Deep Learning Phase
al.[16] using deep learning and a set of images comprising Deep learning is used to classify tomato plant diseases at
eleven types of plant diseases with a total of 1650 images
spread in the Philippines to build a system for classifying this stage the classification process using deep learning went
plant diseases using Android Studio. The deep learning through several steps, including collecting and pre-
model has been trained on images and published via Docker processing the dataset and training the models, after which
for use in a mobile application by android studio, used only the best model is converted to a Tensorflow Lite Model
few data are available for common plant diseases in the (TFLM). Then build the application that classifies plant
Philippines. Smetanin et al.[17] they built a mobile diseases. The following sections explain the steps in detail.
application to classify plant diseases in two methods. The
first method is inserting the image into the platform and Fig. 1: General structure of the proposed system.
classifying it using deep learning. The second method is by
entering a text with the type of problem and using the BERT
(Bidirectional Encoder Representations from Transformers)
model to classify similar texts. Adedoja et al.[18] the
researchers created a mobile application using React Native
to detect plant diseases by taking a picture of the affected
plant. The training process used the NASNet model transfer
learning method and a data set of 54,309 for 14 crops
containing 32 classes of diseases. Jasim & Al-Tuwaijari,
[19] used a proposed model for CNN to train the dataset on
classifying plant diseases and building a classification
interface in Matlab. In training, the network used a database
from PlantVillage for three varieties (tomato, pepper, and
potato) of plants with a size of 20636 images and obtained
an accuracy of 98%. Muangprathub et al.[20] the
researchers developed an IoT system that works inside the
farm to monitor soil moisture and control water sprinklers
automatically through a mobile application. Analyzing the
data measured by temperature, humidity, and ultrasound