A Mosquito Distribution Prediction Method Based on High-Resolution Remote Sensing Maps and the Combination of Deep Learning and Machine Learning

Abstract

Objective To establish a mosquito distribution prediction model using high-resolution remote sensing maps and machine learning deep learning to determine the risk distribution of mosquito-borne infectious disease vectors and provide a decision-making basis for precise prevention and control. Method From March to November 2017–2021, a mosquito survey was conducted at 20 mosquito monitoring points in Wuxi City. High-resolution remote sensing image semantic segmentation was used to identify the ground features of a 250 meter radius buffer zone at the mosquito monitoring points. The typical correlation between the number of mosquitoes and ground features was analyzed, and an XGBoost model was established. Results A total of 5 types of mosquitoes were observed through the investigation, namely, Culex pipiens pallens, Culex trituberculatus, Anopheles sinensis, Aedes albopictus, and Aedes disturbance, representing 56.5%, 31.7%, 5.0%, 4.0%, and 1.7% of the observed mosquitos, respectively. The accuracy of the land feature data obtained by machine recognition and manual interpretation reaches 70%. The correlation coefficient between the ground feature data and mosquito distribution is 0.892. The accuracy, recall, precision, and F1-score of the dominant species prediction models are 0.916, 0.875, 0.857, and 0.865, respectively, while those for the common species are 0.758, 0.669, 0.733, and 0.699, respectively. Conclusion The prediction model established by machine learning deep learning can effectively predict the distribution of mosquitoes and provide a decision-making basis for precise prevention and control.