"Aedes Vigilax" Detection from Buzz: Deep One-Class Classification

Poor or excessive nutrient management may result in generation of mosquitos in vineyard which is a potential impact of vineyard on residential area. Some species of mosquitos are real threat for human society. For instance, a linkage was observed between vineyards and West Nile virus which spreads via mosquitos [1]. Thus, continuous effective monitoring system is required to ensure mitigation of mosquito borne diseases originated from orchards and vineyards. Numerous image-based machine learning (ML) approaches has been utilized in mosquito systematics, but considering the small body size, these models often required high resolution images and sophisticated pre-processing algorithm to result in high accuracy. Moreover, those classifiers often do not generalize well across different datasets due to a relatively small number of Aedes samples. In this paper, we adopt a one-class perspective for mosquito detection, where the detection classifier is trained with Aedes vigilax mosquito class samples only, which is a major coastal pest species for NSW and more northern areas, and also for parts of coastal SA. Our model employs a BERT-BiLSTM module for feature extraction and a one-class SVM for classification. A comprehensive evaluation with a benchmarking dataset demonstrates the better performance of our model than existing approaches.

Reference

[1] Crowder, David W.,Dykstra, Elizabeth A., Brauner, Jo Marie, Duffy, Anne, Reed, Caitlin, Martin, Emily, Peterson, Wade, Carrière, Yves, Dutilleul, Pierre and Owen, Jeb P. (2013). West Nile Virus Prevalence across Landscapes Is Mediated by Local Effects of Agriculture on Vector and Host Communities. PLOS ONE, 8:1, 1-8.