Pest incidence forecasting based on Internet of Things and Long Short-Term Memory Network

The infestation of litchi stink bugs (Tessaratoma papillosa) has always had a significant impact on the yield of longan plantations. Pest control is critical for farmers to detect and timely suppress the occurrence of pests while effectively reducing damages. Environmental factors, climate change in particular, have contributed to the growing population of pests whereas weather can vary in different terrains, locations, and time. Due to the geographical and topographical conditions of Taiwan, this study focuses on investigating fruit plantations on sloping land in subtropics with distinct seasonal changes. The article aims at forecasting meteorological data based on Long short-term memory network (LSTM) and identifying the correlation between pest infestation and environmental factors through Machine Learning (ML). In this section, the structure and experimental process of the research will be outlined. At the first stage, meteorological information of the experimented site is obtained through the self-designed IoT (Internet of Things) system and wireless long-distance transmission technology. Since meteorological information forecasted is displayed in time series, multi-layer LSTM and bidirectional LSTM are used to solve the problem. Finally, environmental data and field surveys conducted for pest surveillance will be employed to forecast the severity of pest infestation through KNN, SVM, and random forest models. The result of the experiment shows that LSTM performs well in weather forecasting with 96% R-Squared values whereas the accuracy rate of pest prediction conducted by Machine Learning (ML) is 85%. The study verifies that meteorological factors do affect pest incidence. For example, the population of litchi stink bugs increase easily under suitable temperature, humidity, and sunlight. LSTM is superior in providing solutions for long-range dependence in statistics. This article shall present regions with shifting weather patterns, meteorological conditions and time length forecasted corresponding to the oceanic climate, as well as the correlation between pest population and environmental factors.