Code for Infrared Small Target Detection with Multi-Branch Perception and Cross-Layer Semantic Fusion

This file contains the code for the Multi-Branch Perception and Cross-Layer Semantic Fusion Network (MPCF-Net) for infrared small target detection, designed to support the development and evaluation of infrared small target detection models. The network employs multi-branch perception and cross-layer semantic fusion strategies to effectively extract edge features and address the semantic disparity issues in cross-layer feature fusion present in existing U-Net-based detection networks. The datasets used in the experiments are sourced from three publicly available infrared small target detection datasets: SIRST, IRSTD, and NUDT-SIRST, which include infrared images with complex backgrounds and annotations for small targets. The file includes the complete MPCF-Net implementation code, which comprises training scripts (train.py), testing scripts (test.py), and network structure definitions (seg_model.py), among others. With this code, researchers can effectively implement and evaluate infrared small target detection models, validate the effectiveness of multi-branch perception and cross-layer semantic fusion methods, and further advance the technology in this field.