TCI-Former: Thermal Conduction-Inspired Transformer for Infrared Small Target Detection

TL;DR

ISTD faces challenges from extremely small targets in low-contrast infrared imagery. The paper introduces TCI-Former, a thermal conduction-inspired transformer that derives the Pixel Movement Differential Equation (PMDE) from thermodynamics and implements two modules, TCIA and TCBM, to capture target body features and refine boundaries. Experimental results on NUAA-SIRST and IRSTD-1k show state-of-the-art performance in IoU, $nIoU$, and object-level metrics with efficient model complexity, validating the effectiveness of the thermodynamics-inspired approach. This framework provides a physically motivated perspective on ISTD and demonstrates how heat-diffusion analogies can guide precise, boundary-aware small-target segmentation.

Abstract

Infrared small target detection (ISTD) is critical to national security and has been extensively applied in military areas. ISTD aims to segment small target pixels from background. Most ISTD networks focus on designing feature extraction blocks or feature fusion modules, but rarely describe the ISTD process from the feature map evolution perspective. In the ISTD process, the network attention gradually shifts towards target areas. We abstract this process as the directional movement of feature map pixels to target areas through convolution, pooling and interactions with surrounding pixels, which can be analogous to the movement of thermal particles constrained by surrounding variables and particles. In light of this analogy, we propose Thermal Conduction-Inspired Transformer (TCI-Former) based on the theoretical principles of thermal conduction. According to thermal conduction differential equation in heat dynamics, we derive the pixel movement differential equation (PMDE) in the image domain and further develop two modules: Thermal Conduction-Inspired Attention (TCIA) and Thermal Conduction Boundary Module (TCBM). TCIA incorporates finite difference method with PMDE to reach a numerical approximation so that target body features can be extracted. To further remove errors in boundary areas, TCBM is designed and supervised by boundary masks to refine target body features with fine boundary details. Experiments on IRSTD-1k and NUAA-SIRST demonstrate the superiority of our method.

Figures (5)

• Figure 1: Conversion process between the image field and the thermal field.The feature map evolution process in the image field can be analogous to the thermal conduction process in the thermal field. The upper part depicts the image field and presents the from-coarse-to-fine feature map evolution in the ISTD process. The upper right corner shows the change of pixel value in a 2-D image micro-element. The lower part shows the thermal conduction process of a 3-D micro-element in the thermal field, where thermal energy is conducted spontaneously from high-temperature areas to low-temperature areas.
• Figure 2: Overall architecture of our TCI-Former with an encoder-decoder structure. The encoder is composed of several TCIT blocks. Each TCIT block contains two key components: Thermal Conduction-Inspired Attention (TCIA) and Thermal Conduction Boundary Module (TCBM), which are both devised based on our derived pixel movement differential equation (PMDE). PMDE is inspired by the thermal conduction differential equation (TCDE) in heat dynamics.
• Figure 3: Overall architecture of our proposed Thermal Conduction-Inspired Attention (TCIA), which is devised based on finite difference method and PMDE derived from the TCDE in heat dynamics.
• Figure 4: Result visualization of different ISTD methods.
• Figure 5: Visualization of (a) TCIA and TCBM branch outputs and (b) intermediate stage feature map evolution.