Real-time monitoring of the construction sites is crucial for reliability and safety of the workers and working environment. Despite its importance, low-illumination conditions present a considerable challenge for the effective monitoring in the construction industry. We introduce FFT-Mosaic, an innovative approach which can integrate the generalized two-dimensional Fourier domain processing with mosaic data augmentation. By utilizing a mixed window function and Gaussian masks, it can refine the Fourier transform process, thus significantly improving the quality of the low-illumination images. FFT-Mosaic can enhance the dataset with mosaic data augmentation, which would improve the detection accuracy of small-scale targets. We have conducted a comprehensive analysis of artifact generation in the image reconstruction process and developed the targeted solutions. The experimental results can demonstrate that the FFT-Mosaic method markedly improves the target detection in low-illumination conditions, enriches the training dataset and enhances the model generalization capability.