Design and implementation of AI system for fast and reliable weed control

博士研究课题的研究计划这里提出的研究问题是设计和实现一个用于杂草控制的快速人工智能系统。我们的目标是创建一个满足几个关键要求的可靠系统。-实时功能-快速有效地在不同类型的作物中进行杂草检测-隔离感兴趣的区域以更好地定位位置-导航机器人到受影响的区域并找到可能的最佳方法来根除不需要的生长-在不同的角度和情况下可用在不同的要求上，这些要求不会立即成为问题，但属于额外的任务，这些要求充当了系统可扩展和高效的催化剂。-确定杂草的类型--杂草的形态特征--增强现实(对场景进行3D建模，从而更好地识别和研究)--使用现有的硬件创建除草机制--确定农作物田地的污染程度，以确定优先次序并对其采取行动。在开展手头研究的同时，可以进行一些可选的添加，以进一步提高农业的质量。-使用实时图像分类来检测作物中的视觉不一致，及早检测真菌或细菌感染-识别受影响的有害生物区域，以避免虫害和限制化学物质的使用为什么它很重要？全球的农业产业非常多样化。虽然许多亚洲国家依靠廉价劳动力和劳动力来耕种土地和维持农作物，但西方世界更多地依赖于科学进步和重型机械。这两个部分似乎都在进步，越来越依赖于基因开发和化学品的使用。在这些情况下使用的化学品主要落在杀虫剂和除草剂部分。这需要大量的人力来识别这些化学品并将其应用于受影响的地区。更便宜和更快的选择是使用大量除草剂，这将为污染和土壤质量退化打开大门。这可能会对环境造成长期影响，使土地在不久的将来无法使用。在这种情况下，最好的解决方案是使用人工智能识别受影响的地区，并在那里应用集中数量的除草剂，以尽早根除杂草，从而将对环境的影响降至最低。这就是为什么需要一种准确的方法来检测杂草的位置，并建立一种自动化的实时方法来根除杂草。如果不广泛审查现行标准、简单方法、方法论而不广泛审查现行标准，这个项目的简单方法将是采用任务划分和隔离的方法。这些方法作为单独的独立方法工作，它们独立运行，并依赖于其他方法的输入和输出。在这种情况下，主要的方法是数据集的准备、分割、特征提取和分类方法。项目中使用的数据集将取决于工业合作伙伴(小型机器人公司)的可用性。数据集将是该项目的关键因素，因为作物的角度和类型将指导其他任务的进行。准备工作将遵循图像可能的正常化以及图像与实时数据的比较情况。确保数据尽可能接近实际应用。该数据集还将为分段任务的开始指明方向。分割任务将是一个简单的大小调整任务，其中感兴趣区域将从完整的图像中分离出来。T

Research plan for PhD Research question The research question being asked here to solve is to design and implement a fast AI system for weed control. The goal is to create a reliable system which satisfies a few key requirements. - Real-time functionality - Quick and efficient weed detection among different types of crops - Isolate the region of interest to pinpoint the location better - Navigate the robots to the affected area and find the best possible method to eradicate the unwanted growth - Usable in different angles and circumstances There are additional requirements which does not come into question immediately but falls under additional tasks which act as catalysts for the system to be scalable and efficient. - Identifying the type of weed - Morphological properties of the weed - Augmented reality (3D modelling of the scenario leading to better identification and research) - Creating a mechanism for the weed to be removed using the hardware available - Identifying the level of contamination in the crop fields to determine the priority and acting on it There are a few optional additions that can be made that can further enhance the quality of agriculture while working on the research at hand. - Using the real-time image classification to detect visual inconsistencies in the crops to detect fungal or bacterial infections early - Identifying pests affected areas to avoid infestation and limiting chemical use Why is it important? The agriculture industry is very diverse across the globe. While a lot of the Asian countries rely on cheap labour and manpower to cultivate the land and maintain the crops, the western world relies more on scientific advancement and heavy machinery. Both parts are seemingly progressing and relying more and more on genetic development and usage of chemicals. The chemicals in use in these cases fall predominantly on the pesticide and herbicide portion. This requires tremendous amount of manpower to identify and apply these chemicals to the affected areas. The cheaper and quicker option would be the application of mass herbicide which opens the gates for pollution and degradation of soil quality. This can leave long-term impacts on the environment which can render the land unusable in the near future. The best solution in this case is to identify the affected areas using artificial intelligence and apply concentrated amounts of herbicide there to eradicate the weeds early and often so the impact on the environment in minimal. This is why an accurate method for detecting the location of the weeds and establishing an automated real-time method to eradicate them is necessary. Methodology Without extensively reviewing the current standards, the simple methodology,Methodology Without extensively reviewing the current standards, the simple methodology for this project would be to proceed using a task division and isolation method. The methods works as separate standalone methods which function independently and relies on the inputs and outputs of the other methods. The main methods in this case would be the dataset preparation, segmentation, feature extraction and the classification methods. The datasets used in the project would be subject to the availability from the industrial partner (Small robot company). The datasets would be the key factors of the project as the angles and the types of crop would navigate the way to go forward with the other tasks. The preparation would follow the likely normalisation of the images and how the images would compare to real-time data. Making sure the data is as close to the actual application as possible. The dataset would also guide the way for the segmentation task to commence. The segmentation task would be a simple resizing task where the region of interest would be isolated from the complete image. T