Characterising the Behavioural Differences between Insecticide Susceptible and Insecticide Resistant Mosquito Species - 1=Sensors and instrumentation

与蚊子有关的疾病每年造成大约100万人死亡，因此减少疾病传播的研究至关重要。蚊子是影响人类和动物的一些最致命的病毒和寄生虫感染的病媒，如疟疾、登革热、基孔肯雅热、西尼罗河热和许多其他疾病。长效杀虫蚊帐（LLINs）和室内滞留喷洒（IRS）是目前使用的主要干预方法，迄今为止效果良好。然而，人们越来越关注蚊子种群中杀虫剂耐药性的增加。美国疾病控制与预防中心的一份报告发现，蚊子的抗药性增加了1000多倍，这使得目前的一些杀虫剂变得无效。人们已经对蚊子的生理抗性进行了一些研究，但对行为抗性的了解要少得多。分类和了解杀虫剂敏感和抗药性蚊子之间的行为差异将有助于了解蚊子的行为，并进一步有助于开发新的有效干预治疗方法。华威大学和利物浦热带医学学院此前开发了识别和跟踪蚊子飞行的系统。拟议的研究将通过利用这些实验数据来确定杀虫剂易感蚊子（IS）和杀虫剂抗性蚊子（IR）的关键行为飞行特征以及它们之间的差异，从而进一步推进该项目。为了做到这一点，机器学习模型，如人工神经网络、随机森林决策树和k近邻将被开发、训练和评估。一个扩展目标可以使用这些分类器来开发一个实时分析蚊子物种的系统，以进一步协助研究。博士项目的主要目标是：-收集或获得合适的蚊子飞行训练数据集（2D或3D轨迹）。—生成和选择各种特征进行分析。识别合适的机器学习分类器，并构建、测试和评估它们。-确定易受杀虫剂影响和抗药性蚊子的行为特征差异最大。-扩展目标将是开发一个使用低成本相机系统的实时分类系统，或探索（与LSTM）如何利用已识别的行为特征来提高干预方法的有效性。该项目将从量化蚊子飞行的各种特征和相关统计数据开始。然后使用筛选方法、包装方法和嵌入方法等方法执行特征选择。这些算法的组合将被使用，因为每一个都有不同的优点和缺点。这样做的结果将是一组独特的特征，并包含相关信息，使该模型能够对蚊子进行分类。项目的最后阶段将包括分类，这需要构建、测试和评估机器学习分类器。分类器可能包括随机森林、支持向量机和naïve贝叶斯，它们在鸟类分类研究中表现良好。然而，需要对所使用的机器学习模型进行进一步研究。[1] Worldmosquitoprogram.org。2021. 蚊子传播的疾病|世界蚊子计划。[在线]可在：https://www.worldmosquitoprogram.org/en/learn/mosquito-borne-diseases[2] J. Parker， N. et al Scientific Reports, vol. 5, no. 5。13392年,2015年。[3]to<s:1>, k.h.等人，《新发传染病》，第20卷，第2期。2014。[4]张志强，张志强，等。模式识别技术的研究进展。中文信息学报，2004,(1):53-62。

Mosquito-related diseases are responsible for approximately 1 million deaths every year and so research into reducing disease transmission is vital. Mosquitoes are a disease-vector for some of the deadliest viral and parasitic infections affecting both humans and animals, such as malaria, dengue, Chikungunya, West Nile Fever and many others [1]. Long-lasting insecticidal bed nets (LLINs) and indoor residual spraying (IRS) are the principle intervention methods currently used and have worked effectively to date [2]. However, there is growing concern of increasing insecticide resistance within mosquito populations. A report by the Center for Disease Control and Prevention found that resistance in mosquitoes has increased by over 1000-fold, which renders some current insecticides ineffective [3]. There has been some research into the physiological resistance of mosquitoes, but behavioural resistance is much less understood. Classifying and understanding the differences in behaviour between insecticide susceptible and insecticide resistant mosquito species will assist with understanding mosquito behaviour and further aid the development of novel effective intervention treatments.The University of Warwick and the Liverpool of Tropical Medicine have previously developed systems to identify and track the flight of mosquitoes [2]. The proposed study would take this project further by utilising this experimental data to identify key behavioural flight features of insecticide susceptible (IS) and insecticide resistant (IR) mosquitoes and the differences between them. To perform this, machine learning models such as artificial neural networks, random forest decision trees and k-nearest neighbours would be developed, trained, and evaluated. A stretch target could use these classifiers to develop a system for real-time analysis of mosquito species to further assist with research.The principal objectives of the PhD project are: - To collect or obtain suitable training datasets of mosquito flight (2D or 3D tracks). - To generate and select various features for analysis. - To identify suitable machine learning classifiers, and to build, test and evaluate them. - To determine the behavioural features that differ the most between the insecticide susceptible and insecticide resistant mosquito species. - Stretch targets would be to develop a system for real-time classification using a low-cost camera system or explore (with LSTM) how the identified behavioural features could be exploited to improve the efficacy of intervention methods.The project will begin by quantifying various features and associated statistics from measurements of mosquito flight. Features selection is then performed, using methods such as: filter methods, wrapper methods and embedded methods. A combination of these algorithms will be used, as each has various advantages and disadvantages. The result of this will be a set of features that are unique and hold relevant information that enable the model to classify the mosquitoes. The final stage of the project would include the classification, which entails building, testing, and evaluating the machine learning classifiers. The classifiers may include random forests, support vector machines and naïve bayes - which worked well in a study on the classification of birds [4]. However, further study into the set of machine learning models used would need to take place. [1] Worldmosquitoprogram.org. 2021. Mosquito-borne diseases | World Mosquito Program. [online] Available at: https://www.worldmosquitoprogram.org/en/learn/mosquito-borne-diseases[2] J. Parker, N. et al Scientific Reports, vol. 5, no. 13392, 2015.[3] Toé, K. H. et al., Emerging Infectious Diseases, vol. 20, no. 10, 2014.[4] Atanbori, J et al Pattern Recognition Letters, 81, pp.53-62.