Deep Learning for Environmental Remote Sensing Image Understanding: Analyzing Dust and Anthropogenic Objects Across Varying Scales and Densities

Steady advancements in sensor technologies and platforms have significantly increased the collection of electro-optical data over large areas of the Earth’s surface. To manage this vast amount of data, deep learning methods offer promising solutions for various tasks such as classification, object detection, semantic segmentation, and density estimation. Despite ongoing challenges, this work focuses on analyzing objects and particles at different scales and densities. Analyzing airborne dust is particularly suitable for this purpose. However, developing an efficient and robust algorithm for airborne dust analyzing entails several hurdles. Airborne dust can be opaque or translucent, exhibit considerable variation in density, and possess indistinct boundaries. Moreover, distinguishing dust from other atmospheric phenomena, such as fog or clouds, is particularly challenging. In this work, multiple neural networks are developed for the analyzing of dust. The initial focus is on density estimation, but to improve the approach, other methods and tasks are considered, particularly object detection. Not only are neural network architectures examined, but attention is also given to training and inference optimization. Instead of dust, anthropogenic objects such as ships or airplanes are primarily considered, which can vary greatly in their scale and object density. One of the strengths of this study is the combination of methods from density estimation and object detection. It is demonstrated that the integration of both tasks can lead to improved object detection, better object counting, and ultimately enhance dust density estimation. The proposed methods are evaluated on multiple datasets and compared with state-of-the-art techniques. This work aspires to advance the research of analyzing objects and particles in remote sensing across various object sizes and scales, especially for dust monitoring. Analyzing of airborne dust holds substantial potential benefits for climate protection, environmentally sustainable construction, scientific research, and various other fields.