Adaptive AI-enabled Mobile Intelligence Solutions for Climate-smart Pest Management

在英国，各种各样的害虫影响可耕地作物（例如，小麦、大麦、油菜籽、马铃薯）。虫害会对农作物造成重大损害，平均产量损失15- 20%，在高水平虫害下，损失率可达80%。农药通常用于作物以提供保护并限制产量损失。这些施药往往是在保险的基础上进行的（喷洒是作为应急措施，以减轻潜在的产量损失），而不是在规定的基础上进行的（施药是在正确识别虫害水平超过损害阈值之后进行的）。保险喷雾剂的应用可以增加昆虫种群的进化压力，导致抗性昆虫种群的发展。目前，英国农作物病虫害的防治主要依靠专家的人工检查和向农民或种植者提供适当的病虫害防治措施的建议。然而，作物害虫的个体（种植者主导的）识别是有限的，因为准确的害虫识别需要分类学培训。一个流行的解决方案是使用人工智能（AI）技术来自动识别害虫。然而，由于农作物种类繁多，害虫种类繁多，这些解决方案在实际应用中的准确性和鲁棒性较低。本项目将研究将情境和视觉信息与自适应人工智能技术整合到移动的解决方案中的技术可行性，该解决方案提供：1）快速检测和量化农作物害虫; 2）有效预测可接受的害虫阈值; 3）估算农药在气候智能型虫害控制方面的相应功效。我们将探索优化谢菲尔德大学（UoS）现有的深度学习害虫检测模型PestNet，并采用子结构最优传输机制识别多种农作物害虫;评估自适应连续学习技术，以融合来自现有数据集的图像特征和上下文信息，支持气候变化害虫量化;探讨如何结合已确定的害虫品种的抗药性状况的公认害虫阈值水平，以产生信息丰富的害虫管理产出。该项目将利用现有的资源和技术，包括害虫管理和害虫管理。数据集（3.2K注释的小麦害虫图像）和来自Spark-Soft的移动云平台，以及谢菲尔德大学（UoS）的PestNet模型。Spark-Soft将与UoS合作管理该项目的开发活动。

In the UK a diverse range of pests affect arable crops (e.g., wheat, barley, rapeseed, potatoes). Insect pests can cause significant damage to arable crops, with average yield of losses of 15-20%, with this increasing to 80% under high levels of infestation. Pesticides are often applied to crops to provide protection and to limit yield losses. These applications are often done on an insurance basis (a spray is applied as contingency to mitigate potential yield loss) rather than a prescriptive basis (application follows correct pest identification with levels exceeding damage thresholds). The application of insurance sprays can increase the evolutionary pressure on insect populations, leading to the development of resistant insect populations. There is increasing demand for intelligent systems that can help farmers grow more sustainably through smart pesticide use to improve farm resilience while supporting more sustainable practices.Currently, the prevention and control of UK crops pests heavily reply on experts' manual inspections and recommendations to farmers or growers on appropriate pest control measures. However, individual (grower-led) identification of crop pests is limited as accurate pest identification requires taxonomic training. A popular solution for this is the use of artificial intelligence (AI) techniques for automated, image-based identification of pests. However, these solutions suffer from reduced accuracy and robustness in real-world applications due to multiplicity of crops and variety of pests.This project will investigate the technical feasibility of integrating contextual and visual information with adaptive AI technique into a mobile solution that offers: 1) rapid detection and quantification of arable crops pests; 2) efficient forecasting of accepted pest thresholds; 3) estimation of the corresponding efficacy of a pesticide for climate-smart pest control. We will explore optimising existing deep learning pest detection model PestNet by the University of Sheffield (UoS) with substructural optimal transport mechanisms in recognising multiple arable crops pests; evaluate adaptive continuous learning techniques for fusing image features and contextual information from existing datasets for supporting climate-change pest quantification; explore how this can be combined with accepted pest threshold levels on the pesticide resistance status of the identified pest species to produce an informative pest management output.The project will build on existing resources and technologies including pest data set (3.2K annotated wheat pest images) and mobile-cloud platforms from Spark-Soft, and PestNet model at the University of Sheffield (UoS). Spark-Soft will manage the development activity of this project in collaboration with UoS.