A study on web-based real-time greenhouse crops pest diagnosis system using artificial intelligence

Abstract

Background/Objectives: An effort to increase agricultural output is a very important factor in reducing the agricultural population. The convergence of ICT technology makes it possible to computerize and mechanize the cultivation of greenhouse crops, but the damage from disease and harmful insects is very serious, resulting in a decrease in production. Therefore, the diagnosis and prevention of pests are necessary, and studies on the pests prediction system using artificial intelligence have been conducted. In this paper, we propose a pests diagnosis system using web - based artificial intelligence. Methods/Statistical analysis: It is a model to diagnose the correct name of pests through the characteristics of pests to minimize the damage caused by pests that may occur during the cultivation of greenhouse crops and to make appropriate initial responses. It is a situation that needs pre-diagnosis and prevention of pests, so we constructed a data set using the data on pests generated during the cultivation of red peppers among the professional data on pests registered in the national crop pest management system and learned the pests data through the TensorFlow.JS library. Findings: Since it is necessary to diagnose and prevent pests in advance, we proposed a web based artificial intelligence system to diagnose pests in real time and made it easy to enter and use data via JavaScript, allowing users to use a web browser instead of a console entry. This makes it possible to utilize the powerful functions provided on the web, use the system in an environment in which an internet connection is possible and handle the data entry and modification easily. We can improve the utilization of diagnostic results by using tools such as tables and graphs through a web browser, accumulate the data and results used in the diagnosis of pests by linking with the web based database and improve the accuracy by re-learning the model. Based on three diseases, the prediction model was tested five times, and the prediction model accurately diagnosed the disease through the input data. It was also found that it is possible to accurately predict the disease through the feature data of the disease even if there are some errors in the entry process. Improvements/Applications: Future studies should continue to be carried out to improve the response speed to analysis request and subdivision work for the feature information of a disease in order to exclude the possibility of errors in the diagnosis of pests when the feature data of the disease is similar.