Vespa Mandarinia Recognition and Prediction Strategy Based on Geographic Location and Image Recognition

Abstract

The paper conduct data analysis and information mining from the following four aspects: the correlation between geographic location and the spread of Vespa mandarinia, image recognition with the help of convolutional neural networks, species richness prediction and accurate recognition optimization, Vespa mandarinia reproduction law and reproduction characteristics, and then propose a reliable prediction model of Vespa mandarinia spread range and a model to predict the possibility of misclassification. First, consider whether the spread of pests in a period of time can be predicted. We arrange the 14 Positive IDs in chronological order and use a time series model to predict the spread range, quantify the results as longitude and latitude, and compare the results. The optimal degree is calculated. The results show that this model can accurately predict the staged emergence of Vespa mandarinias. Next, we select 1/6 high-resolution images based on the provided data set files and image files, based on the SNP selection algorithm of the χ2 test, and perform data labeling preprocessing to create a model that can predict the likelihood of a mistaken classification. Further borrow the geographic location of model one and the image recognition of model two, and use both as factors influencing the possibility of whether the newly submitted report is confirmed as a Vespa mandarinia. With the help of the Euclidean distance within the bounds of the rectangular frame, it is compared with the distance between two adjacent points to determine the impact factor 1, which is conducive to predicting the geographical distribution of the Vespa mandarinia through time series analysis. With the help of the image recognition model, the output of the convolutional neural network is used as the second impact factor, which is conducive to high-performance identification of Vespa mandarinias through real pictures. Further analysis shows that the weight ratio of the two impact factors is 3:7. Finally, we analyzed the Vespa mandarinia’s cycle reproduction law, combined with the possible genetic variation and species evolution within the population, and concluded that if there are other new reports, they should be added to the training set to rebuild the model. training. At the same time, the possibility of the elimination of this pest in Washington State is discussed. Through the statistical historical data of the monthly detection report, and further combining with time prediction, until the forecast report stabilizes to 0 within a certain period, or in a certain period If reports are submitted within the cycle, but all non-Vespa mandarinias are judged, it can be deemed that Washington State has eliminated this pest.