A Method for Predicting Trap Counts of Stored-Grain Insects Based on Machine Learning

The growth of stored-grain insects leads to the loss of quantity and quality of stored grains. Electronic probe traps are the most commonly used devices for real-time monitoring the insect occurrence in grain storage practice. A method for predicting the future trap counts of stored-grain insects captured by electronic probe traps based on machine learning is proposed. The feature vector is constructed by integrating five modalities of features, including the number of trap counts, the grain temperature and the air relative humidity of the trapping site (where the electronic probe trap is located), the semantic information of the trapping site, and the calculated result of the degree-day model associated with the development of insect populations. An Insect Trap Counts Predictor based on Multi-scale Feature Cross (ITCP-MFC) is proposed. The prediction of the total number of trap counts obtained by an electronic probe trap in the next five days is realized. Through K-fold (K=10) cross-validation, the mean of determination coefficient(R
²
) reaches 0.867. The proposed method can provide reliable evidence for estimating the development trend of insects in grain bulks.