Abstract
Deep learning, robotics, AI, and automation have lots of applications that are beneficial to society at large. In fact, nearly every sector, such as transportation, industries, manufacturing, healthcare, education, retail, and home automation, are adopting AI, machine learning, IoT, and robotics to their advantage. Of course, agriculture is no exception. The chapter starts with an introduction to the applications of deep learning in agriculture. Next, a comprehensive survey of the research work done in recent years is provided. It is followed by the description of various techniques of deep learning (DL). The next section briefly describes the traditional ways of weed detection and removal. Next, the architecture of deep learning for weed detection and removal is presented along with the associated code. Further, the chapter goes on to discuss the pros and cons of this approach. Finally, the chapter concludes by citing the important points discussed in this study.
TopLiterature Survey
This section presents a survey of state-of-the-art papers on weed detection. The survey comprises of papers on Image Processing, UAV, and Deep Learning.
In image processing, the line segmentation is most basic technique for distinguishing two or more types of objects in an image. Mao et al., (2007) describes a method for weed detection that is based on the optimized segmentation line of weed and crop. In this study, the spectral information in the visible light captured by the CCD camera is used. Basically, the use of both G-R and H-S Optimized Segmentation line helps in detecting the flixweed (tansy mustard which is a broadleaf weed and can be highly toxic to cattle) and separates it from the wheat crop. Cheng et al., (2015) presented a method that automatically discriminates rice and weed using a feature-based Machine Learning (ML) agent. Basically, the intelligent agent used in this study makes use of sensors in order to automate a process that separates rice plants from weeds. The algorithm used for this purpose is Harris Corner Detection algorithm that determines the points of interest. Further, the proposed approach extracts multiple features surrounding each point. Then a Machine Learning algorithm is used for training. Finally, a clustering algorithm removes the noise.