Sampling method for fire point data based on multi-source remote sensing data fusion

Utilizing deep learning for fire point detection is currently one of the most popular approaches in satellite remote sensing. However, even with appropriate models, the variable quality of fire point training data can significantly impact detection accuracy. Regarding this issue, the study introduces a sampling method for fire point data that combines VIIRS fire point product J1 with high spatial resolution Sentinel-2 data. It enhances the OTSU algorithm using Sentinel-2 data features to identify burned areas and performs spatio-temporal matching with fire point product J1, classifying and sampling through post-validation methods. The sampling method was validated and evaluated using fire point verification data obtained through on-site manual verification and visual interpretation. The results demonstrate that the proposed method can more accurately and efficiently sample true and false fire point data, generate a large number of high-quality samples, and provide data support for deep learning fire point detection models.