DPAITI 2023 : Data Processing with Artificial Intelligence in Thermal Imagery

Dear Colleagues,

Thermal imaging possesses various advantages over the visible light spectrum, allowing us to not only address challenging lighting conditions (e.g., poor lighting [1]), but also reveal information invisible to the naked eye [2]. For this reason, this imaging domain is continuously gaining more popularity across a broad variety of markets, e.g., in the automotive industry for scene understanding [3] and driver monitoring [4]; in the medical field for evaluation of skin conditions [5] or vital sign extraction [6]; and for smart vision in surveillance [7] and border control [8] applications, just to name a few.

At the same time, it is important to note that thermal imagery has different characteristics than visible light data [9]. First, due to the heat flow in objects, thermal images are more blurred with smooth borders between objects and there is an absence of high-frequency components such as edges and textures [10]; frequently, the lack of color data also makes image processing more challenging [11]. Secondly, ranges of thermal sensors are usually shorter than in the case of standard cameras, allowing them to capture only close-proximity scenes. Finally, the resolution of such data is usually lower due to the higher cost of imaging sensors [12].

Although the research in artificial intelligence is progressing at warp speed, only a few studies have focused on imaging domains other than RGB. Furthermore, models are usually designed with visible light spectrum data in mind, assuming that high-frequency components are present in the input data, which are then directly applied to other datasets. However, this frequently leads to worse accuracy [13,14], as such networks cannot capture specific data characteristics, e.g., more distant relationships between object components in thermal images that require bigger receptive fields [15].

Taking this into account, this Special Issue focuses on increasing the community's awareness of the importance of thermal imagery, its benefits and challenges, as well as the need for careful analysis and design of AI solutions with specific data domains in mind. Proposals addressing various research topics are welcome, including, but not limited to:

Thermal imaging applications in medicine, automotive, aerospace, robotics, and surveillance industries, among others.
AI design for thermal imagery including Neural Architecture Search for domain-specific tasks.
Data translation between imaging domains.
Thermal data generation using AI.

Dr. Alicja Kwasniewska
Dr. M. Hamed Mozaffari
Prof. Dr. Yudong Zhang
Guest Editors