The Development of Insect Pest Classification Mobile Application from Images via MobileNet

Insect pests are insect organisms with a vast diversity which has a considerable amount of impact on agriculture. They are known as Plant Destroying Organism because of their nature to hinder plant's growth on a cultivated land. Knowledge regarding the impact and the nature of insect pests can be an important point to be able to prevent them from having a rapid growth. Utilizing scientific knowledge branch regarding the classification of insect pests through imagery could be an idea for an effort towards insect pests’ control. An algorithm system for classifying insect pests through images has been developed in the previous study using MobileNetV2 as the architecture and it has been tested in a controlled environment with the data of 75.000 insect pests’ images and resulted in a data test accuracy of 71,36%. However, the classification algorithm from the previous study has yet to have a media that could provide accessibility for the public to use. The main objective of this study is solely about finding out the performance and reliability of an existing image classification algorithm if it is utilized and embedded inside of an Android application with image classification as its main function. This study develops an Android application as a vessel for the existing algorithm to show whether or not the algorithm can stay accurate and precise in a very different situation with random external factors. Android is chosen to be the operating system for the application in this study because of its popularity in the market. Based on the data provided by Newzoo in 2020, the smartphone ownership of Indonesians reaches 64% and the data provided by GlobalStats statscounter website, in 2021 – 2022 the market share for Android operating system in Indonesia covers 89.42% while IOS only covers about 10.46%. With that in mind, it can be concluded that majority of Indonesian population already have smartphone devices and using Android as the operating system. The application in this study is tested through black box testing for functionality test purpose and user acceptance testing utilizing System Usability Scale to measure end-user's satisfaction about the application. Black Box Testing achieved all success in all existing features. The User Acceptance Testing conducted by 5 participants using their own smartphone devices obtained an effectiveness level of 92.5% and the value obtained through System Usability Scale is 80.5. There is also a testing to determine the accuracy of the classification function and it shows that the embedded artificial intelligence has the accuracy of 49.02% in mobile which is a decrease compared to the test accuracy in a controlled environment.