Abstract

This study proposes a ‘Smart Apple Guardian (SAG)’ architecture for apple diseases, leveraging Internet of Things (IoT) sensors and artificial intelligence (AI) technologies. Specifically, the proposed architecture employs a deep learning model—incorporating CNNs, RNNs, and LSTMs—to target anthracnose and Marssonina blotch and predict disease outbreaks using environmental data (e.g., temperature, humidity, precipitation, and wind speed) collected through weather and soil IoT sensors. The training dataset was constructed based on disease occurrence history data from 2013 to 2024 from the Rural Development Administration (RDA). A key contribution of this study is the proposal of an Optimal Input Data Window for AI that integrates the existing disease-specific incubation periods (11 days for anthracnose and 21 days for Marssonina blotch) into a 15-day model. Experimental results show that the training dataset with the proposed 15-day incubation period showed higher prediction accuracy than the datasets with existing disease-specific incubation periods. Specifically, the RNN model achieved the best performance in Anthracnose prediction, with an accuracy of 0.9727, a specificity of 0.9455, and a sensitivity of 1.0000. In contrast, the LSTM model achieved superior performance in Marssonina blotch prediction, with an accuracy of 0.9200, a specificity of 0.8982, and a sensitivity of 0.9418. In conclusion, this study demonstrates that applying a unified incubation period can improve both the predictive accuracy and data efficiency of AI models, providing a robust framework for precision agriculture.