Page 78 - IJEEE-2023-Vol19-ISSUE-1
P. 78

74 | Abdulla & Marhoon

          Fig. 8: Curve for training accuracy and loss for (a) proposed method, (b) MobileNet_v2.

          TABLE III

TESTING METRICS FOR PROPOSED METHOD AND MOBILENET_V2.

Models    Testing   Precision  recall   F1-
          accuracy                     score

MobileNet_v2 96.6% 96% 96% 96%

Proposed  97.62%    97%        98% 97%
 method

  • Build Classified Application Part                                                  ab

   At this stage, it is building a classification section for      Fig. 9: Test mobile app to classification google image for
tomato plant diseases in the mobile application using                   tomato diseases with actual classification results.
Android Studio. The proposed model converted to TFLM
provides many advantages, including small size and fast           B. IoT System Phase
conclusion, enabling it to work on a mobile device with                At this stage, we use the IoT system to monitor the
limited memory and computing. Android Studio builds the
application's front end through the activity_main.xml file        environmental conditions of the plant and control irrigation
and the backend through the MainActivity.java file.               remotely. A set of sensors is used to measure environmental
Therefore, built the application's front end to contain two       conditions and programmed using a microcontroller to send
buttons (Gallery) for fetching photos from the gallery to         data via WiFi to the cloud. Then the data is called to the
classify them and (Take Picture) for taking photos using the      mobile application for processing and decision-making.
mobile camera. And the third button (Disease Cause)               Sensors monitor the environmental conditions surrounding
displays the causes of the disease. After completing the step     the tomato plant, which affect it, including temperature and
of building the classification interface, programmed the          humidity, soil moisture, water quality, and carbon dioxide
application's backend to classify tomato leaf diseases using      concentration ( Co!). And use the soil moisture sensor for
TFLM with print classification confidence. Mobile                 remote plant irrigation. Fig.10 shows the flowchart for
application was run on the Android operating system and has       monitoring and irrigation system remotely.
been tested on 114 images for tomato diseases collected from
the internet due to the difficulty of obtaining actual test
samples. As the classification of this image is shown in Fig.
9, where (a) represents the classification image for late blight
disease, and (b) for yellow leaf curl virus. Therefore, when
the farmer discovers that the plant has a disease, he can know
the type of disease through this phase of the application.
   73   74   75   76   77   78   79   80   81   82   83