For those who have strayed here and don't know me yet, here are a few words about me: I'm Jakob and I'm currently studying Geoinformatics and Spatial Data Science at the University of Münster in my master's degree. Here I am especially interested in remote sensing, object detection and location privacy. I also work as a student assistant at IVV5 and help to establish and maintain a functioning IT infrastructure for the employees of the university. Privately, I am involved in machine learning, as well as image processing.
- ❗ Opened issue #794 in r-lidar/lidR
In this project, I built a web application with four other fellow students that allows spatial predictions to be made using models generated with machine learning. The service allows the training of models based on own training data as well as the computation of predictions based on models and spatially corresponding automatically provided satellite images. The frontend was developed with Angular in Typescript, the backend with a Node.js express server and R scripts for the predictions.
city-lab is a web application that we developed in a group as part of a study project, the goal was to provide users of urban gardening a tool with which such a garden can be easily managed. In the tool different objects can be created by users, which can be viewed by all users. My main task was the administration of the database structures and the access to them via the javascript framework mongoose.
Machine Learning for insect Monitoring introduces an method employing machine learning to monitor insect populations within complex outdoor settings, underlining the pressing need to address insect decline. The approach encompasses manual annotation, preprocessing of RGB and DVS videos, dataset curation, cross-validation procedures, training detection models using YOLO, and evaluating results based on performance metrics such as [email protected] and [email protected]. Statistical analyses, including paired t-Tests, are employed to assess the efficacy of diverse preprocessing techniques in bolstering detection performance. Through systematic experimentation, the study scrutinizes the impact of various preprocessing methods on insect detection across different input streams. Notably, the integration of temporal filtering and background subtraction on the DVS stream significantly enhances insect detection compared to alternative configurations, underscoring the pivotal role of preprocessing techniques in augmenting detection metrics.
"Playground Hub" is a web application designed to address the social and practical challenges surrounding public playgrounds. With the aim of enhancing the experiences of families and communities, the project focuses on providing a comprehensive platform for discovering, reporting, and managing issues related to playgrounds. For parents and caregivers, the application offers intuitive tools to find suitable playgrounds based on specific criteria and report any concerns they encounter, ensuring the safety and cleanliness of these spaces for children. On the administrative side, city officials benefit from streamlined issue management and insightful analytics, enabling prompt responses to community needs and the efficient maintenance of public spaces. The project emphasizes user-centric design principles, interactive map functionalities, and efficient user interactions to create a seamless experience for both end-users and administrators. Through "Playground Hub," the project seeks to foster community engagement, improve the quality of public playgrounds, and promote the well-being of families and children in urban environments.
This study investigating into the critical role of forests in maintaining global environmental stability, with a specific focus on understanding forest ecosystems and addressing threats to diverse tree species. It proposes the adoption of LiDAR technology to enhance forest monitoring and gain species-specific insights in North Rhine-Westphalia (NRW). The research employs a systematic approach to data acquisition, preprocessing, and analysis, emphasizing the efficient handling of LiDAR data to distinguish tree species in monocultural forests in NRW. The findings highlight the effectiveness of LiDAR data in differentiating tree species, with significant variations observed in derived metrics among species. While recognizing the utility of the Random Forest algorithm, the study acknowledges performance fluctuations based on parameter combinations, stressing the importance of a multifaceted approach for accurate species classification, particularly when considering patch-level information.
You can reach me via a Mail to [email protected]