BIMFM: A Bidimensional Intrinsic Mode Functions Mixup Strategy for Thermal Imagery Data Augmentation

Infrared imaging has the advantages of good concealment, wide night vision range, strong detection capability, and long operating range, so infrared target detection and recognition has been widely used in smart cars, weather forecasting, security fields in recent years. However, due to the privacy of infrared image datasets and the high cost of the infrared imaging equipment, the data is difficult to obtain. Therefore, the classification model cannot be fully trained, which limits the recognition effect. This paper proposes an infrared image augmentation method based on two-dimensional empirical mode decomposition (BEMD). The method generates new image samples by randomly combining the two-dimensional intrinsic mode functions (BIMFs) of infrared images after two-dimensional mode decomposition. The experimental validations on two machine learning models and one deep learning model show that the data augmentation method can effectively increase data size and improve the image recognition results, thus having great application potential and practical value.