[{"data":1,"prerenderedAt":137},["ShallowReactive",2],{"jobs-list-en":3,"thesis-list-en":4},[],[5,30,47,64,81,98,119],{"id":6,"title":7,"author":8,"body":9,"date":20,"description":15,"extension":21,"meta":22,"navigation":23,"path":24,"seo":25,"status":26,"stem":27,"type":28,"__hash__":29},"thesis\u002Fen\u002Fthesis\u002F20251001_schwass.md","Multi-Backbone Embedding-Fusion: An Attention-Based Deep Learning Approach for Fusing Heterogeneous Computer Vision Backbones","Jacob Schwaß",{"type":10,"value":11,"toc":16},"minimark",[12],[13,14,15],"p",{},"This thesis investigates the fusion of latent representations from multiple heterogeneous computer vision backbones as an approach to explicitly structure and stabilize the resulting embedding space. The objective is to analyze the conditions under which learning-based fusion yields more discriminative, semantically consistent, and data-efficient representations. Five pre-trained backbone architectures are employed as pure feature extractors, whose embeddings are projected into a shared latent space using learnable adapters and fused through an attention-based model combining multi-head self-attention and additive attention. The evaluation is conducted on ImageNet100 and on a domain-specific dataset for ragweed detection and comprises classification, structural, robustness, efficiency, and few-shot analyses. Complementary visualization and interpretability studies include UMAP projections, correlation analyses of backbone embeddings, and Gradient × Input investigations. The results demonstrate that learning-based fusion achieves substantial gains particularly when the underlying backbones exhibit moderate or heterogeneous representational quality. Fusion improves cluster coherence and the global structure of the embedding space, whereas classical approaches such as averaging show structural limitations. Simple concatenation attains comparable performance in some cases but at the cost of significantly increased dimensionality and without explicit information integration or compression. Under domain shifts, fusion stabilizes existing feature spaces without fundamentally extending generalization capability. In few-shot settings, fused representations improve data efficiency, although advantages diminish as training data increases. Overall, the thesis provides a nuanced assessment of embedding fusion: it does not constitute a universal replacement for strong individual backbones but represents an effective approach for structurally integrating heterogeneous representations in practical computer vision applications.",{"title":17,"searchDepth":18,"depth":18,"links":19},"",2,[],"2025-10-01","md",{},true,"\u002Fen\u002Fthesis\u002F20251001_schwass",{"title":7,"description":15},"closed","en\u002Fthesis\u002F20251001_schwass","master","vXGppMOtJK8T75jJnhTVLcbfds9oMm22P7qrcfMWV3Y",{"id":31,"title":32,"author":33,"body":34,"date":41,"description":38,"extension":21,"meta":42,"navigation":23,"path":43,"seo":44,"status":26,"stem":45,"type":28,"__hash__":46},"thesis\u002Fen\u002Fthesis\u002F20250117_lehn.md","Developing a Domain-Specific Foundation Model with Adapters and Self-Supervised Learning for Efficient Plant Detection in Grasslands","Mirko Lehn",{"type":10,"value":35,"toc":39},[36],[13,37,38],{},"The rise of foundation models has enabled generalizable solutions across diverse computer vision tasks. However, their performance often declines in specialized domains where data is limited, annotations are sparse, and subtle visual cues dominate. This thesis targets such a scenario: detecting individual plants in grassland fields where low inter-class variance, weak contrast, and small objectsizes pose persistent challenges. Instead of discarding prior work like DETReg, which was criticized for its lack of classification and reliance on generic object priors, this thesis revisits and extends it. Building on the Deformable Detection Transformer (Deformable DETR), the proposed method introduces an unsupervised, domainaware pretraining pipeline that directly addresses DETReg’s limitations. It uses custom pseudo boxes tailored to the target domain and gradually incorporates teacher model predictions via an Exponential Moving Average (EMA)-based training loop. Semantic learning is introduced early through a binary classification signal, and regularization mechanisms are applied to stabilize training and avoid collapse. While results show only marginal gains when abundant labeled data is available, the benefits become pronounced under few-shot conditions. The proposed pretraining strategy consistently outperforms models initialized from unrelated domains such as COCO. Moreover, it is designed to operate with minimal supervision and for reducing not only labeled data requirements but also the need for large volumes of unlabeled data by leveraging curriculum schedules and early semantic cues. Lightweight adapter modules are optionally integrated to stabilize the training in the early phase and to test additional possibilities for data reduction, but are not decisive for the final performance. This work presents a scalable and interpretable approach to domain-specific pre-training, aimed at data efficiency and practical applicability in real world, lowresource scenarios. Applications include ecological monitoring, automated weed control, and other tasks requiring finegrained recognition in complex natural environments.",{"title":17,"searchDepth":18,"depth":18,"links":40},[],"2025-01-17",{},"\u002Fen\u002Fthesis\u002F20250117_lehn",{"title":32,"description":38},"en\u002Fthesis\u002F20250117_lehn","xMWB2alLbwUnH_spkc6W_o3dee5FkUidAYRPlTH2AaA",{"id":48,"title":49,"author":50,"body":51,"date":58,"description":55,"extension":21,"meta":59,"navigation":23,"path":60,"seo":61,"status":26,"stem":62,"type":28,"__hash__":63},"thesis\u002Fen\u002Fthesis\u002F20241224_reuter.md","Semantic Interoperability in the Web of Things: Design and Validation of a Reference Data Model for Monitoring Applications in the Grasslands Agricultural Domain","Sebastian Reuter",{"type":10,"value":52,"toc":56},[53],[13,54,55],{},"There currently exists no ontology for the pest management subdomain of precision agriculture. This prevents the development of interoperable Internet of Things (IoT) solutions, and cements the existing heterogeneous landscape of vertical solutions. In the context of developing a weed detection device for KI-gestützte hochautomatisierte Unkrautbekämpfung im Grünland (KIhUG), a research project at the Technische Hochschule Mittelhessen (THM) a partial ontology has been designed with the goal of making the device highly interoperable with third-party weed management systems. The ontology and the accompanying set of semantic data models proved to be a valid starting point for the design of a full-scale pest management ontology.",{"title":17,"searchDepth":18,"depth":18,"links":57},[],"2024-12-24",{},"\u002Fen\u002Fthesis\u002F20241224_reuter",{"title":49,"description":55},"en\u002Fthesis\u002F20241224_reuter","V4H0n_VFs5oMqVK919LzB2EiN0DD7PqlLeSbvr78GKc",{"id":65,"title":66,"author":67,"body":68,"date":75,"description":72,"extension":21,"meta":76,"navigation":23,"path":77,"seo":78,"status":26,"stem":79,"type":28,"__hash__":80},"thesis\u002Fen\u002Fthesis\u002F20240409_vaupel.md","Generation of Synthetic Training Data for Neural Network based Weed Detection with Stable Diffusion","Dennis Vaupel",{"type":10,"value":69,"toc":73},[70],[13,71,72],{},"Artificial intelligence is becoming a key technology in many industries, including agriculture. Precision farming heavily leverages modern computer vision technology to equip farm machines as well as autonomous robotic platforms with perceptive capabilities. For years, neural networks have been the first choice when it comes to machine vision. Although robust and high-performing, neural networks come with a caveat: Their training requires a great amount of annotated training data. This presents an obstacle to machine learning engineers who need a certain dataset size to reach their required performance metrics. Collecting data themselves is often not economically feasible for agriculture companies. The annotation process is also labor-intensive and takes valuable time. Synthetic training data provide a remedy to this concern. This work develops a framework to generate synthetic image data along with pixel-wise annotations. It presents a multi-stage ensemble reference architecture and demonstrates its potential with an example of weed in grassland. The work concludes with a critical discussion and possible real-world applications.",{"title":17,"searchDepth":18,"depth":18,"links":74},[],"2024-04-09",{},"\u002Fen\u002Fthesis\u002F20240409_vaupel",{"title":66,"description":72},"en\u002Fthesis\u002F20240409_vaupel","0D5M8CeXAdBh6j1-HS0CcdtPJnYxswzKJ4JYoQl-wIg",{"id":82,"title":83,"author":84,"body":85,"date":92,"description":89,"extension":21,"meta":93,"navigation":23,"path":94,"seo":95,"status":26,"stem":96,"type":28,"__hash__":97},"thesis\u002Fen\u002Fthesis\u002F20240301_hohl.md","Deep Learning based Growth Modeling of Plant Phenotypes","Renke Hohl",{"type":10,"value":86,"toc":90},[87],[13,88,89],{},"This master’s thesis investigates the largely unexplored domain of image-based plant growth modeling using deep learning, aiming to generate synthetic training data for object detection and segmentation across different plant growth stages. Demonstrating the ability of deep learning models to simulate growth of plants this study introduces a novel class of neural network architectures called Latent Growth Models (LGMs), which attempt to model the temporal aspects of plant growth exclusively within the efficient latent space of pre-trained autoencoders. Specifically, the research of this thesis includes the development and investigation of three distinct LGM architectures, some of which demonstrate superior performance over existing state-of-the-art methods in terms of image quality, while retaining a high degree of realism in the simulated growth, leveraging the Aberystwyth Leaf Evaluation dataset for empirical validation. Furthermore, the architectures achieve comparably fast training times, enhancing their practical applicability and setting a new benchmark in the field. Despite these advancements, this thesis identifies a significant gap in the current research landscape: the absence of specialized metrics capable of evaluating the realism with which plant growth is simulated in image data. This gap highlights a crucial direction for future research, emphasizing the development of robust evaluation tools to quantitatively assess the authenticity and applicability of synthetic image data in advancing the field of plant phenotyping and beyond. Furthermore, the thesis delineates strategies for refining the proposed methodologies in subsequent studies. These primarly consider architectural details and training strategies of the models, with a particular focus on enhancing image-based modeling of plant growth.",{"title":17,"searchDepth":18,"depth":18,"links":91},[],"2024-03-01",{},"\u002Fen\u002Fthesis\u002F20240301_hohl",{"title":83,"description":89},"en\u002Fthesis\u002F20240301_hohl","BQqVEF7au3cDfZMkdfGXgafC3bvU-e0YAVBMlk18N04",{"id":99,"title":100,"author":101,"body":102,"date":112,"description":106,"extension":21,"meta":113,"navigation":23,"path":114,"seo":115,"status":26,"stem":116,"type":117,"__hash__":118},"thesis\u002Fen\u002Fthesis\u002F20231201_koch.md","Generation of Synthetic Model-Based Image Data for Training Artificial Neural Networks","Till Koch",{"type":10,"value":103,"toc":110},[104,107],[13,105,106],{},"To train artificial intelligence (AI) for object detection, large amounts of annotated data are needed. With increasing awareness of AI, diverse application areas are also increasingly opening up. An exemplary field of application is agriculture, especially in the area of weed removal. However, the creation of specific datasets, for example for weed detection, can be time-consuming and cost-intensive.",[13,108,109],{},"To facilitate this process, synthetic model-based image data of grassland is generated in this bachelor's thesis for training artificial neural networks. Both a pipeline for creating this data is described, as well as its applicability tested on real recordings.",{"title":17,"searchDepth":18,"depth":18,"links":111},[],"2023-12-01",{},"\u002Fen\u002Fthesis\u002F20231201_koch",{"title":100,"description":106},"en\u002Fthesis\u002F20231201_koch","bachelor","FtWLKDv0Idk6VpnxmQrp05pvzlRwMAD0n-vCJPg8Ox4",{"id":120,"title":121,"author":122,"body":123,"date":130,"description":127,"extension":21,"meta":131,"navigation":23,"path":132,"seo":133,"status":26,"stem":134,"type":135,"__hash__":136},"thesis\u002Fen\u002Fthesis\u002F20231128_soethe.md","Evaluation and Selection of an Object Detection Model for Weed Detection in Grassland","Nico Soethe",{"type":10,"value":124,"toc":128},[125],[13,126,127],{},"This bachelor's thesis deals with the object detection of ragwort (JKK) in grassland. The goal of the work is to create a detector that can be used under variable environmental conditions. A reliable detector is essential to support farmers in detecting and potentially automating the removal of this poisonous plant, which poses serious dangers in hay production. After creating the dataset, several neural networks were trained and evaluated based on predefined requirements such as detection accuracy and processing speed. Finally, a model was selected that achieved the best results compared to other models and experimental setups.",{"title":17,"searchDepth":18,"depth":18,"links":129},[],"2023-11-28",{},"\u002Fen\u002Fthesis\u002F20231128_soethe",{"title":121,"description":127},"en\u002Fthesis\u002F20231128_soethe","Bachelor","5yP8l4BZTWJGEtHKz09Q71CwzDuvvidk0Fi0O0QRYSs",1793017186364]