Design, construction, and selection of appropriate deep learning techniques for an autonomous lawn mowing robot

In this thesis study, deep learning techniques were investigated for an autonomous lawn mowing robot. A vision-based recognition system, utilizing deep learning, was designed to improve environmental perception and decision-making processes, with a focus on agricultural and landscaping applications. The primary objective of the study is to present an autonomous system that minimizes human intervention, thereby increasing efficiency in terms of labor and time. To ensure that the lawn mowing process is performed safely, accurately, and continuously, grass, soil, and boundary classification was conducted. For this purpose, deep learning models such as VGG16, VGG19, MobileNet, ResNet18, and ResNet50 were trained using transfer learning techniques. Transfer learning enabled the adaptation of pre-trained model weights from large datasets, thereby reducing training time and enhancing model efficiency. The results demonstrated that VGG16 and VGG19 models achieved an accuracy of 1.0. This study highlights the innovative applications of artificial intelligence-based deep learning techniques in autonomous robotic systems, providing efficiency and convenience for modern agriculture and landscaping applications.